2020-06-15 14:18:57 +00:00
|
|
|
|
/*-*- mode:c;indent-tabs-mode:nil;c-basic-offset:2;tab-width:8;coding:utf-8 -*-│
|
2023-12-08 03:11:56 +00:00
|
|
|
|
│ vi: set et ft=c ts=2 sts=2 sw=2 fenc=utf-8 :vi │
|
2020-06-15 14:18:57 +00:00
|
|
|
|
╞══════════════════════════════════════════════════════════════════════════════╡
|
|
|
|
|
|
│ Copyright 2020 Justine Alexandra Roberts Tunney │
|
|
|
|
|
|
│ │
|
2020-12-28 01:18:44 +00:00
|
|
|
|
│ Permission to use, copy, modify, and/or distribute this software for │
|
|
|
|
|
|
│ any purpose with or without fee is hereby granted, provided that the │
|
|
|
|
|
|
│ above copyright notice and this permission notice appear in all copies. │
|
2020-06-15 14:18:57 +00:00
|
|
|
|
│ │
|
2020-12-28 01:18:44 +00:00
|
|
|
|
│ THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL │
|
|
|
|
|
|
│ WARRANTIES WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED │
|
|
|
|
|
|
│ WARRANTIES OF MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE │
|
|
|
|
|
|
│ AUTHOR BE LIABLE FOR ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL │
|
|
|
|
|
|
│ DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR │
|
|
|
|
|
|
│ PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER │
|
|
|
|
|
|
│ TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR │
|
|
|
|
|
|
│ PERFORMANCE OF THIS SOFTWARE. │
|
2020-06-15 14:18:57 +00:00
|
|
|
|
╚─────────────────────────────────────────────────────────────────────────────*/
|
|
|
|
|
|
#include "dsp/core/core.h"
|
|
|
|
|
|
#include "dsp/core/half.h"
|
|
|
|
|
|
#include "dsp/core/illumination.h"
|
|
|
|
|
|
#include "dsp/mpeg/mpeg.h"
|
|
|
|
|
|
#include "dsp/scale/scale.h"
|
|
|
|
|
|
#include "dsp/tty/quant.h"
|
|
|
|
|
|
#include "dsp/tty/tty.h"
|
|
|
|
|
|
#include "libc/assert.h"
|
|
|
|
|
|
#include "libc/calls/calls.h"
|
|
|
|
|
|
#include "libc/calls/internal.h"
|
|
|
|
|
|
#include "libc/calls/struct/framebufferfixedscreeninfo.h"
|
|
|
|
|
|
#include "libc/calls/struct/framebuffervirtualscreeninfo.h"
|
|
|
|
|
|
#include "libc/calls/struct/iovec.h"
|
|
|
|
|
|
#include "libc/calls/struct/itimerval.h"
|
|
|
|
|
|
#include "libc/calls/struct/sigaction.h"
|
|
|
|
|
|
#include "libc/calls/struct/siginfo.h"
|
|
|
|
|
|
#include "libc/calls/struct/sigset.h"
|
Make improvements
- Every unit test now passes on Apple Silicon. The final piece of this
puzzle was porting our POSIX threads cancelation support, since that
works differently on ARM64 XNU vs. AMD64. Our semaphore support on
Apple Silicon is also superior now compared to AMD64, thanks to the
grand central dispatch library which lets *NSYNC locks go faster.
- The Cosmopolitan runtime is now more stable, particularly on Windows.
To do this, thread local storage is mandatory at all runtime levels,
and the innermost packages of the C library is no longer being built
using ASAN. TLS is being bootstrapped with a 128-byte TIB during the
process startup phase, and then later on the runtime re-allocates it
either statically or dynamically to support code using _Thread_local.
fork() and execve() now do a better job cooperating with threads. We
can now check how much stack memory is left in the process or thread
when functions like kprintf() / execve() etc. call alloca(), so that
ENOMEM can be raised, reduce a buffer size, or just print a warning.
- POSIX signal emulation is now implemented the same way kernels do it
with pthread_kill() and raise(). Any thread can interrupt any other
thread, regardless of what it's doing. If it's blocked on read/write
then the killer thread will cancel its i/o operation so that EINTR can
be returned in the mark thread immediately. If it's doing a tight CPU
bound operation, then that's also interrupted by the signal delivery.
Signal delivery works now by suspending a thread and pushing context
data structures onto its stack, and redirecting its execution to a
trampoline function, which calls SetThreadContext(GetCurrentThread())
when it's done.
- We're now doing a better job managing locks and handles. On NetBSD we
now close semaphore file descriptors in forked children. Semaphores on
Windows can now be canceled immediately, which means mutexes/condition
variables will now go faster. Apple Silicon semaphores can be canceled
too. We're now using Apple's pthread_yield() funciton. Apple _nocancel
syscalls are now used on XNU when appropriate to ensure pthread_cancel
requests aren't lost. The MbedTLS library has been updated to support
POSIX thread cancelations. See tool/build/runitd.c for an example of
how it can be used for production multi-threaded tls servers. Handles
on Windows now leak less often across processes. All i/o operations on
Windows are now overlapped, which means file pointers can no longer be
inherited across dup() and fork() for the time being.
- We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4()
which means, for example, that posix_spawn() now goes 3x faster. POSIX
spawn is also now more correct. Like Musl, it's now able to report the
failure code of execve() via a pipe although our approach favors using
shared memory to do that on systems that have a true vfork() function.
- We now spawn a thread to deliver SIGALRM to threads when setitimer()
is used. This enables the most precise wakeups the OS makes possible.
- The Cosmopolitan runtime now uses less memory. On NetBSD for example,
it turned out the kernel would actually commit the PT_GNU_STACK size
which caused RSS to be 6mb for every process. Now it's down to ~4kb.
On Apple Silicon, we reduce the mandatory upstream thread size to the
smallest possible size to reduce the memory overhead of Cosmo threads.
The examples directory has a program called greenbean which can spawn
a web server on Linux with 10,000 worker threads and have the memory
usage of the process be ~77mb. The 1024 byte overhead of POSIX-style
thread-local storage is now optional; it won't be allocated until the
pthread_setspecific/getspecific functions are called. On Windows, the
threads that get spawned which are internal to the libc implementation
use reserve rather than commit memory, which shaves a few hundred kb.
- sigaltstack() is now supported on Windows, however it's currently not
able to be used to handle stack overflows, since crash signals are
still generated by WIN32. However the crash handler will still switch
to the alt stack, which is helpful in environments with tiny threads.
- Test binaries are now smaller. Many of the mandatory dependencies of
the test runner have been removed. This ensures many programs can do a
better job only linking the the thing they're testing. This caused the
test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb
- long double is no longer used in the implementation details of libc,
except in the APIs that define it. The old code that used long double
for time (instead of struct timespec) has now been thoroughly removed.
- ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing
backtraces itself, it'll just print a command you can run on the shell
using our new `cosmoaddr2line` program to view the backtrace.
- Crash report signal handling now works in a much better way. Instead
of terminating the process, it now relies on SA_RESETHAND so that the
default SIG_IGN behavior can terminate the process if necessary.
- Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
|
|
|
|
#include "libc/calls/struct/timespec.h"
|
2022-09-13 06:10:38 +00:00
|
|
|
|
#include "libc/calls/struct/winsize.h"
|
2020-06-15 14:18:57 +00:00
|
|
|
|
#include "libc/calls/termios.h"
|
|
|
|
|
|
#include "libc/calls/ucontext.h"
|
2023-11-12 06:32:12 +00:00
|
|
|
|
#include "libc/cxxabi.h"
|
2020-06-15 14:18:57 +00:00
|
|
|
|
#include "libc/errno.h"
|
2020-12-09 23:04:54 +00:00
|
|
|
|
#include "libc/fmt/conv.h"
|
|
|
|
|
|
#include "libc/fmt/itoa.h"
|
Make improvements
- Every unit test now passes on Apple Silicon. The final piece of this
puzzle was porting our POSIX threads cancelation support, since that
works differently on ARM64 XNU vs. AMD64. Our semaphore support on
Apple Silicon is also superior now compared to AMD64, thanks to the
grand central dispatch library which lets *NSYNC locks go faster.
- The Cosmopolitan runtime is now more stable, particularly on Windows.
To do this, thread local storage is mandatory at all runtime levels,
and the innermost packages of the C library is no longer being built
using ASAN. TLS is being bootstrapped with a 128-byte TIB during the
process startup phase, and then later on the runtime re-allocates it
either statically or dynamically to support code using _Thread_local.
fork() and execve() now do a better job cooperating with threads. We
can now check how much stack memory is left in the process or thread
when functions like kprintf() / execve() etc. call alloca(), so that
ENOMEM can be raised, reduce a buffer size, or just print a warning.
- POSIX signal emulation is now implemented the same way kernels do it
with pthread_kill() and raise(). Any thread can interrupt any other
thread, regardless of what it's doing. If it's blocked on read/write
then the killer thread will cancel its i/o operation so that EINTR can
be returned in the mark thread immediately. If it's doing a tight CPU
bound operation, then that's also interrupted by the signal delivery.
Signal delivery works now by suspending a thread and pushing context
data structures onto its stack, and redirecting its execution to a
trampoline function, which calls SetThreadContext(GetCurrentThread())
when it's done.
- We're now doing a better job managing locks and handles. On NetBSD we
now close semaphore file descriptors in forked children. Semaphores on
Windows can now be canceled immediately, which means mutexes/condition
variables will now go faster. Apple Silicon semaphores can be canceled
too. We're now using Apple's pthread_yield() funciton. Apple _nocancel
syscalls are now used on XNU when appropriate to ensure pthread_cancel
requests aren't lost. The MbedTLS library has been updated to support
POSIX thread cancelations. See tool/build/runitd.c for an example of
how it can be used for production multi-threaded tls servers. Handles
on Windows now leak less often across processes. All i/o operations on
Windows are now overlapped, which means file pointers can no longer be
inherited across dup() and fork() for the time being.
- We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4()
which means, for example, that posix_spawn() now goes 3x faster. POSIX
spawn is also now more correct. Like Musl, it's now able to report the
failure code of execve() via a pipe although our approach favors using
shared memory to do that on systems that have a true vfork() function.
- We now spawn a thread to deliver SIGALRM to threads when setitimer()
is used. This enables the most precise wakeups the OS makes possible.
- The Cosmopolitan runtime now uses less memory. On NetBSD for example,
it turned out the kernel would actually commit the PT_GNU_STACK size
which caused RSS to be 6mb for every process. Now it's down to ~4kb.
On Apple Silicon, we reduce the mandatory upstream thread size to the
smallest possible size to reduce the memory overhead of Cosmo threads.
The examples directory has a program called greenbean which can spawn
a web server on Linux with 10,000 worker threads and have the memory
usage of the process be ~77mb. The 1024 byte overhead of POSIX-style
thread-local storage is now optional; it won't be allocated until the
pthread_setspecific/getspecific functions are called. On Windows, the
threads that get spawned which are internal to the libc implementation
use reserve rather than commit memory, which shaves a few hundred kb.
- sigaltstack() is now supported on Windows, however it's currently not
able to be used to handle stack overflows, since crash signals are
still generated by WIN32. However the crash handler will still switch
to the alt stack, which is helpful in environments with tiny threads.
- Test binaries are now smaller. Many of the mandatory dependencies of
the test runner have been removed. This ensures many programs can do a
better job only linking the the thing they're testing. This caused the
test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb
- long double is no longer used in the implementation details of libc,
except in the APIs that define it. The old code that used long double
for time (instead of struct timespec) has now been thoroughly removed.
- ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing
backtraces itself, it'll just print a command you can run on the shell
using our new `cosmoaddr2line` program to view the backtrace.
- Crash report signal handling now works in a much better way. Instead
of terminating the process, it now relies on SA_RESETHAND so that the
default SIG_IGN behavior can terminate the process if necessary.
- Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
|
|
|
|
#include "libc/intrin/kprintf.h"
|
2022-09-06 14:04:13 +00:00
|
|
|
|
#include "libc/intrin/safemacros.internal.h"
|
|
|
|
|
|
#include "libc/intrin/xchg.internal.h"
|
2020-06-15 14:18:57 +00:00
|
|
|
|
#include "libc/log/check.h"
|
|
|
|
|
|
#include "libc/log/log.h"
|
2021-03-01 07:42:35 +00:00
|
|
|
|
#include "libc/macros.internal.h"
|
2020-06-15 14:18:57 +00:00
|
|
|
|
#include "libc/math.h"
|
2022-09-06 14:04:13 +00:00
|
|
|
|
#include "libc/mem/alg.h"
|
|
|
|
|
|
#include "libc/mem/arraylist.internal.h"
|
2020-06-15 14:18:57 +00:00
|
|
|
|
#include "libc/mem/mem.h"
|
|
|
|
|
|
#include "libc/nexgen32e/bench.h"
|
|
|
|
|
|
#include "libc/nexgen32e/x86feature.h"
|
|
|
|
|
|
#include "libc/nt/console.h"
|
|
|
|
|
|
#include "libc/nt/runtime.h"
|
|
|
|
|
|
#include "libc/runtime/runtime.h"
|
|
|
|
|
|
#include "libc/sock/sock.h"
|
2022-08-06 10:51:50 +00:00
|
|
|
|
#include "libc/sock/struct/pollfd.h"
|
2020-06-15 14:18:57 +00:00
|
|
|
|
#include "libc/stdio/internal.h"
|
2022-09-06 14:04:13 +00:00
|
|
|
|
#include "libc/stdio/rand.h"
|
2020-06-15 14:18:57 +00:00
|
|
|
|
#include "libc/stdio/stdio.h"
|
|
|
|
|
|
#include "libc/str/str.h"
|
2022-09-06 14:04:13 +00:00
|
|
|
|
#include "libc/str/strwidth.h"
|
|
|
|
|
|
#include "libc/str/unicode.h"
|
2020-06-15 14:18:57 +00:00
|
|
|
|
#include "libc/sysv/consts/af.h"
|
|
|
|
|
|
#include "libc/sysv/consts/auxv.h"
|
|
|
|
|
|
#include "libc/sysv/consts/clock.h"
|
|
|
|
|
|
#include "libc/sysv/consts/ex.h"
|
|
|
|
|
|
#include "libc/sysv/consts/exit.h"
|
|
|
|
|
|
#include "libc/sysv/consts/f.h"
|
|
|
|
|
|
#include "libc/sysv/consts/fd.h"
|
|
|
|
|
|
#include "libc/sysv/consts/fileno.h"
|
|
|
|
|
|
#include "libc/sysv/consts/ipproto.h"
|
|
|
|
|
|
#include "libc/sysv/consts/itimer.h"
|
|
|
|
|
|
#include "libc/sysv/consts/map.h"
|
|
|
|
|
|
#include "libc/sysv/consts/mlock.h"
|
|
|
|
|
|
#include "libc/sysv/consts/o.h"
|
|
|
|
|
|
#include "libc/sysv/consts/ok.h"
|
|
|
|
|
|
#include "libc/sysv/consts/poll.h"
|
|
|
|
|
|
#include "libc/sysv/consts/prio.h"
|
|
|
|
|
|
#include "libc/sysv/consts/prot.h"
|
|
|
|
|
|
#include "libc/sysv/consts/sa.h"
|
|
|
|
|
|
#include "libc/sysv/consts/shut.h"
|
|
|
|
|
|
#include "libc/sysv/consts/sig.h"
|
|
|
|
|
|
#include "libc/sysv/consts/splice.h"
|
|
|
|
|
|
#include "libc/sysv/consts/termios.h"
|
|
|
|
|
|
#include "libc/sysv/consts/w.h"
|
|
|
|
|
|
#include "libc/sysv/errfuns.h"
|
Make improvements
- Every unit test now passes on Apple Silicon. The final piece of this
puzzle was porting our POSIX threads cancelation support, since that
works differently on ARM64 XNU vs. AMD64. Our semaphore support on
Apple Silicon is also superior now compared to AMD64, thanks to the
grand central dispatch library which lets *NSYNC locks go faster.
- The Cosmopolitan runtime is now more stable, particularly on Windows.
To do this, thread local storage is mandatory at all runtime levels,
and the innermost packages of the C library is no longer being built
using ASAN. TLS is being bootstrapped with a 128-byte TIB during the
process startup phase, and then later on the runtime re-allocates it
either statically or dynamically to support code using _Thread_local.
fork() and execve() now do a better job cooperating with threads. We
can now check how much stack memory is left in the process or thread
when functions like kprintf() / execve() etc. call alloca(), so that
ENOMEM can be raised, reduce a buffer size, or just print a warning.
- POSIX signal emulation is now implemented the same way kernels do it
with pthread_kill() and raise(). Any thread can interrupt any other
thread, regardless of what it's doing. If it's blocked on read/write
then the killer thread will cancel its i/o operation so that EINTR can
be returned in the mark thread immediately. If it's doing a tight CPU
bound operation, then that's also interrupted by the signal delivery.
Signal delivery works now by suspending a thread and pushing context
data structures onto its stack, and redirecting its execution to a
trampoline function, which calls SetThreadContext(GetCurrentThread())
when it's done.
- We're now doing a better job managing locks and handles. On NetBSD we
now close semaphore file descriptors in forked children. Semaphores on
Windows can now be canceled immediately, which means mutexes/condition
variables will now go faster. Apple Silicon semaphores can be canceled
too. We're now using Apple's pthread_yield() funciton. Apple _nocancel
syscalls are now used on XNU when appropriate to ensure pthread_cancel
requests aren't lost. The MbedTLS library has been updated to support
POSIX thread cancelations. See tool/build/runitd.c for an example of
how it can be used for production multi-threaded tls servers. Handles
on Windows now leak less often across processes. All i/o operations on
Windows are now overlapped, which means file pointers can no longer be
inherited across dup() and fork() for the time being.
- We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4()
which means, for example, that posix_spawn() now goes 3x faster. POSIX
spawn is also now more correct. Like Musl, it's now able to report the
failure code of execve() via a pipe although our approach favors using
shared memory to do that on systems that have a true vfork() function.
- We now spawn a thread to deliver SIGALRM to threads when setitimer()
is used. This enables the most precise wakeups the OS makes possible.
- The Cosmopolitan runtime now uses less memory. On NetBSD for example,
it turned out the kernel would actually commit the PT_GNU_STACK size
which caused RSS to be 6mb for every process. Now it's down to ~4kb.
On Apple Silicon, we reduce the mandatory upstream thread size to the
smallest possible size to reduce the memory overhead of Cosmo threads.
The examples directory has a program called greenbean which can spawn
a web server on Linux with 10,000 worker threads and have the memory
usage of the process be ~77mb. The 1024 byte overhead of POSIX-style
thread-local storage is now optional; it won't be allocated until the
pthread_setspecific/getspecific functions are called. On Windows, the
threads that get spawned which are internal to the libc implementation
use reserve rather than commit memory, which shaves a few hundred kb.
- sigaltstack() is now supported on Windows, however it's currently not
able to be used to handle stack overflows, since crash signals are
still generated by WIN32. However the crash handler will still switch
to the alt stack, which is helpful in environments with tiny threads.
- Test binaries are now smaller. Many of the mandatory dependencies of
the test runner have been removed. This ensures many programs can do a
better job only linking the the thing they're testing. This caused the
test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb
- long double is no longer used in the implementation details of libc,
except in the APIs that define it. The old code that used long double
for time (instead of struct timespec) has now been thoroughly removed.
- ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing
backtraces itself, it'll just print a command you can run on the shell
using our new `cosmoaddr2line` program to view the backtrace.
- Crash report signal handling now works in a much better way. Instead
of terminating the process, it now relies on SA_RESETHAND so that the
default SIG_IGN behavior can terminate the process if necessary.
- Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
|
|
|
|
#include "libc/thread/thread.h"
|
2024-05-05 06:05:36 +00:00
|
|
|
|
#include "libc/time.h"
|
2022-09-13 06:10:38 +00:00
|
|
|
|
#include "libc/x/xsigaction.h"
|
2023-07-03 02:57:43 +00:00
|
|
|
|
#include "third_party/getopt/getopt.internal.h"
|
2020-06-15 14:18:57 +00:00
|
|
|
|
#include "third_party/stb/stb_image_resize.h"
|
|
|
|
|
|
#include "tool/viz/lib/graphic.h"
|
|
|
|
|
|
#include "tool/viz/lib/knobs.h"
|
|
|
|
|
|
#include "tool/viz/lib/ycbcr.h"
|
|
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
|
* @fileoverview MPEG Video Player for Terminal.
|
|
|
|
|
|
*/
|
|
|
|
|
|
|
|
|
|
|
|
#define GAMMADELTA 0.1
|
|
|
|
|
|
#define NETBUFSIZ (2 * 1024 * 1024)
|
|
|
|
|
|
#define MAX_FRAMERATE (1 / 60.)
|
|
|
|
|
|
|
|
|
|
|
|
#define USAGE \
|
2020-12-25 00:11:38 +00:00
|
|
|
|
" [FLAGS] MPG\n\
|
2020-06-15 14:18:57 +00:00
|
|
|
|
Renders motion picture to teletypewriters.\n\
|
|
|
|
|
|
\n\
|
|
|
|
|
|
Flags & Keyboard Shortcuts:\n\
|
|
|
|
|
|
-s stats\n\
|
|
|
|
|
|
-t true color\n\
|
|
|
|
|
|
-d dithering\n\
|
|
|
|
|
|
-3 ibm cp437 rendering\n\
|
|
|
|
|
|
-4 unicode rendering\n\
|
|
|
|
|
|
-a ansi quantization\n\
|
|
|
|
|
|
-x xterm256 quantization\n\
|
|
|
|
|
|
-A assume ansi ansi palette\n\
|
|
|
|
|
|
-T assume tango ansi palette\n\
|
|
|
|
|
|
-v increases verbosity [flag]\n\
|
|
|
|
|
|
-L PATH redirects stderr to path [flag]\n\
|
|
|
|
|
|
-y yes to interactive prompts [flag]\n\
|
2023-12-23 05:39:27 +00:00
|
|
|
|
-h or -? shows this information [flag]\n\
|
2020-06-15 14:18:57 +00:00
|
|
|
|
UP/DOWN adjust volume [keyboard]\n\
|
|
|
|
|
|
CTRL+L redraw [keyboard]\n\
|
|
|
|
|
|
CTRL+Z suspend [keyboard]\n\
|
|
|
|
|
|
CTRL+C exit [keyboard]\n\
|
2021-02-03 23:53:33 +00:00
|
|
|
|
q quit [keyboard]\n\
|
2020-06-15 14:18:57 +00:00
|
|
|
|
\n\
|
|
|
|
|
|
Effects Shortcuts:\n\
|
|
|
|
|
|
\n\
|
2021-02-03 23:53:33 +00:00
|
|
|
|
S Toggle Swing (TV, PC)\n\
|
|
|
|
|
|
Y Toggle Black/White Mode\n\
|
|
|
|
|
|
p Toggle Primaries (BT.601, BT.709)\n\
|
|
|
|
|
|
g +Gamma G -Gamma\n\
|
|
|
|
|
|
l +Illumination L -Illumination\n\
|
|
|
|
|
|
k +LumaKernel K -LumaKernel\n\
|
|
|
|
|
|
j +ChromaKernel J -ChromaKernel\n\
|
2020-10-11 04:18:53 +00:00
|
|
|
|
CTRL-G {Unsharp,Sharp}\n\
|
2020-06-15 14:18:57 +00:00
|
|
|
|
\n\
|
|
|
|
|
|
Environment Variables:\n\
|
|
|
|
|
|
SOX overrides location of SoX executable\n\
|
|
|
|
|
|
FFPLAY overrides location of FFmpeg ffplay executable\n\
|
|
|
|
|
|
ROWS=𝑦 sets height [inarticulate mode]\n\
|
|
|
|
|
|
COLUMNS=𝑥 sets width [inarticulate mode]\n\
|
|
|
|
|
|
TERM=dumb inarticulate mode\n\
|
|
|
|
|
|
\n\
|
|
|
|
|
|
Notes:\n\
|
|
|
|
|
|
\n\
|
|
|
|
|
|
Your video printer natively supports .mpg files. If your videos are\n\
|
|
|
|
|
|
in a different format, then it's fast and easy to convert them:\n\
|
|
|
|
|
|
\n\
|
|
|
|
|
|
ffmpeg -i movie.mkv movie.mpg\n\
|
|
|
|
|
|
\n\
|
|
|
|
|
|
The terminal fonts we recommend are PragmataPro, Bitstream Vera Sans\n\
|
|
|
|
|
|
Mono (known as DejaVu Sans Mono in the open source community), Menlo,\n\
|
|
|
|
|
|
and Lucida Console.\n\
|
|
|
|
|
|
\n\
|
|
|
|
|
|
On Linux, playing audio requires either `sox` or `ffplay` being on\n\
|
|
|
|
|
|
the $PATH. Kitty is the fastest terminal. Alacritty also has a fast\n\
|
|
|
|
|
|
display. GNOME Terminal and xterm both work well in 256-color or ANSI\n\
|
|
|
|
|
|
mode.\n\
|
|
|
|
|
|
\n"
|
|
|
|
|
|
|
|
|
|
|
|
#define CTRL(C) ((C) ^ 0100)
|
|
|
|
|
|
#define ALT(C) ((033 << 010) | (C))
|
|
|
|
|
|
#define ARGZ(...) ((char *const[]){__VA_ARGS__, NULL})
|
2020-07-01 02:55:47 +00:00
|
|
|
|
#define MOD(X, Y) ((X) - (ABS(Y)) * ((X) / ABS(Y)))
|
2020-06-15 14:18:57 +00:00
|
|
|
|
|
2022-03-18 11:20:10 +00:00
|
|
|
|
#define BALLOC(B, A, N, NAME) \
|
|
|
|
|
|
({ \
|
2021-09-04 04:14:26 +00:00
|
|
|
|
INFOF("balloc/%s %,zu bytes", NAME, N); \
|
2023-07-06 13:57:28 +00:00
|
|
|
|
*(B) = pvalloc(N); \
|
2020-06-15 14:18:57 +00:00
|
|
|
|
})
|
|
|
|
|
|
|
Make improvements
- Every unit test now passes on Apple Silicon. The final piece of this
puzzle was porting our POSIX threads cancelation support, since that
works differently on ARM64 XNU vs. AMD64. Our semaphore support on
Apple Silicon is also superior now compared to AMD64, thanks to the
grand central dispatch library which lets *NSYNC locks go faster.
- The Cosmopolitan runtime is now more stable, particularly on Windows.
To do this, thread local storage is mandatory at all runtime levels,
and the innermost packages of the C library is no longer being built
using ASAN. TLS is being bootstrapped with a 128-byte TIB during the
process startup phase, and then later on the runtime re-allocates it
either statically or dynamically to support code using _Thread_local.
fork() and execve() now do a better job cooperating with threads. We
can now check how much stack memory is left in the process or thread
when functions like kprintf() / execve() etc. call alloca(), so that
ENOMEM can be raised, reduce a buffer size, or just print a warning.
- POSIX signal emulation is now implemented the same way kernels do it
with pthread_kill() and raise(). Any thread can interrupt any other
thread, regardless of what it's doing. If it's blocked on read/write
then the killer thread will cancel its i/o operation so that EINTR can
be returned in the mark thread immediately. If it's doing a tight CPU
bound operation, then that's also interrupted by the signal delivery.
Signal delivery works now by suspending a thread and pushing context
data structures onto its stack, and redirecting its execution to a
trampoline function, which calls SetThreadContext(GetCurrentThread())
when it's done.
- We're now doing a better job managing locks and handles. On NetBSD we
now close semaphore file descriptors in forked children. Semaphores on
Windows can now be canceled immediately, which means mutexes/condition
variables will now go faster. Apple Silicon semaphores can be canceled
too. We're now using Apple's pthread_yield() funciton. Apple _nocancel
syscalls are now used on XNU when appropriate to ensure pthread_cancel
requests aren't lost. The MbedTLS library has been updated to support
POSIX thread cancelations. See tool/build/runitd.c for an example of
how it can be used for production multi-threaded tls servers. Handles
on Windows now leak less often across processes. All i/o operations on
Windows are now overlapped, which means file pointers can no longer be
inherited across dup() and fork() for the time being.
- We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4()
which means, for example, that posix_spawn() now goes 3x faster. POSIX
spawn is also now more correct. Like Musl, it's now able to report the
failure code of execve() via a pipe although our approach favors using
shared memory to do that on systems that have a true vfork() function.
- We now spawn a thread to deliver SIGALRM to threads when setitimer()
is used. This enables the most precise wakeups the OS makes possible.
- The Cosmopolitan runtime now uses less memory. On NetBSD for example,
it turned out the kernel would actually commit the PT_GNU_STACK size
which caused RSS to be 6mb for every process. Now it's down to ~4kb.
On Apple Silicon, we reduce the mandatory upstream thread size to the
smallest possible size to reduce the memory overhead of Cosmo threads.
The examples directory has a program called greenbean which can spawn
a web server on Linux with 10,000 worker threads and have the memory
usage of the process be ~77mb. The 1024 byte overhead of POSIX-style
thread-local storage is now optional; it won't be allocated until the
pthread_setspecific/getspecific functions are called. On Windows, the
threads that get spawned which are internal to the libc implementation
use reserve rather than commit memory, which shaves a few hundred kb.
- sigaltstack() is now supported on Windows, however it's currently not
able to be used to handle stack overflows, since crash signals are
still generated by WIN32. However the crash handler will still switch
to the alt stack, which is helpful in environments with tiny threads.
- Test binaries are now smaller. Many of the mandatory dependencies of
the test runner have been removed. This ensures many programs can do a
better job only linking the the thing they're testing. This caused the
test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb
- long double is no longer used in the implementation details of libc,
except in the APIs that define it. The old code that used long double
for time (instead of struct timespec) has now been thoroughly removed.
- ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing
backtraces itself, it'll just print a command you can run on the shell
using our new `cosmoaddr2line` program to view the backtrace.
- Crash report signal handling now works in a much better way. Instead
of terminating the process, it now relies on SA_RESETHAND so that the
default SIG_IGN behavior can terminate the process if necessary.
- Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
|
|
|
|
#define TIMEIT(OUT_NANOS, FORM) \
|
|
|
|
|
|
do { \
|
|
|
|
|
|
struct timespec Start = timespec_real(); \
|
|
|
|
|
|
FORM; \
|
|
|
|
|
|
(OUT_NANOS) = timespec_tonanos(timespec_sub(timespec_real(), Start)); \
|
2020-06-15 14:18:57 +00:00
|
|
|
|
} while (0)
|
|
|
|
|
|
|
|
|
|
|
|
typedef bool (*openspeaker_f)(void);
|
|
|
|
|
|
|
|
|
|
|
|
enum Sharp {
|
|
|
|
|
|
kSharpNone,
|
|
|
|
|
|
kSharpUnsharp,
|
|
|
|
|
|
kSharpSharp,
|
|
|
|
|
|
kSharpMAX,
|
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
|
|
enum Blur {
|
|
|
|
|
|
kBlurNone,
|
|
|
|
|
|
kBlurBox,
|
|
|
|
|
|
kBlurGaussian,
|
|
|
|
|
|
kBlurMAX,
|
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
|
|
struct NamedVector {
|
|
|
|
|
|
char name[8];
|
|
|
|
|
|
const double (*vector)[3];
|
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
|
|
struct VtFrame {
|
|
|
|
|
|
size_t i, n;
|
|
|
|
|
|
union {
|
2023-07-06 13:57:28 +00:00
|
|
|
|
void *b;
|
2020-06-15 14:18:57 +00:00
|
|
|
|
char *bytes;
|
|
|
|
|
|
};
|
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
|
|
struct FrameCountRing {
|
|
|
|
|
|
size_t i, n;
|
|
|
|
|
|
float p[64]; /* seconds relative to starttime_ */
|
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
|
|
struct FrameBuffer {
|
|
|
|
|
|
void *map;
|
|
|
|
|
|
size_t size;
|
|
|
|
|
|
char *path;
|
|
|
|
|
|
int fd;
|
|
|
|
|
|
struct FrameBufferFixedScreenInfo fscreen;
|
|
|
|
|
|
struct FrameBufferVirtualScreenInfo vscreen;
|
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
|
|
static const struct itimerval kTimerDisarm = {
|
|
|
|
|
|
{0, 0},
|
|
|
|
|
|
{0, 0},
|
|
|
|
|
|
};
|
|
|
|
|
|
|
2022-07-20 21:01:15 +00:00
|
|
|
|
static const struct itimerval kTimerHalfSecondSingleShot = {
|
2020-06-15 14:18:57 +00:00
|
|
|
|
{0, 0},
|
|
|
|
|
|
{0, 500000},
|
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
|
|
static const struct NamedVector kPrimaries[] = {
|
|
|
|
|
|
{"BT.601", &kBt601Primaries},
|
|
|
|
|
|
{"BT.709", &kBt709Primaries},
|
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
|
|
static const struct NamedVector kLightings[] = {
|
|
|
|
|
|
{"A", &kIlluminantA}, {"C", &kIlluminantC},
|
|
|
|
|
|
{"D50", &kIlluminantD50}, {"D55", &kIlluminantD55},
|
|
|
|
|
|
{"D65", &kIlluminantD65}, {"D75", &kIlluminantD75},
|
|
|
|
|
|
{"F2", &kIlluminantF2}, {"F7", &kIlluminantF7},
|
|
|
|
|
|
{"F11", &kIlluminantF11}, {"A-10", &kIlluminantAD10},
|
|
|
|
|
|
{"C-10", &kIlluminantCD10}, {"D50-10", &kIlluminantD50D10},
|
|
|
|
|
|
{"D55-10", &kIlluminantD55D10}, {"D65-10", &kIlluminantD65D10},
|
|
|
|
|
|
{"D75-10", &kIlluminantD75D10}, {"F2-10", &kIlluminantF2D10},
|
|
|
|
|
|
{"F7-10", &kIlluminantF7D10}, {"F11-10", &kIlluminantF11D10},
|
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
|
|
static plm_t *plm_;
|
|
|
|
|
|
static float gamma_;
|
|
|
|
|
|
static int volscale_;
|
|
|
|
|
|
static enum Blur blur_;
|
|
|
|
|
|
static enum Sharp sharp_;
|
|
|
|
|
|
static jmp_buf jb_, jbi_;
|
|
|
|
|
|
static double pary_, parx_;
|
|
|
|
|
|
static struct TtyIdent ti_;
|
|
|
|
|
|
static struct YCbCr *ycbcr_;
|
|
|
|
|
|
static bool emboss_, sobel_;
|
|
|
|
|
|
static volatile int playpid_;
|
|
|
|
|
|
static struct winsize wsize_;
|
|
|
|
|
|
static float hue_, sat_, lit_;
|
2023-07-06 13:57:28 +00:00
|
|
|
|
static void *xtcodes_, *audio_;
|
2020-06-15 14:18:57 +00:00
|
|
|
|
static struct FrameBuffer fb0_;
|
|
|
|
|
|
static unsigned chans_, srate_;
|
|
|
|
|
|
static volatile bool ignoresigs_;
|
|
|
|
|
|
static size_t dh_, dw_, framecount_;
|
|
|
|
|
|
static struct FrameCountRing fcring_;
|
|
|
|
|
|
static volatile bool resized_, piped_;
|
|
|
|
|
|
static int lumakernel_, chromakernel_;
|
|
|
|
|
|
static openspeaker_f tryspeakerfns_[4];
|
|
|
|
|
|
static int primaries_, lighting_, swing_;
|
|
|
|
|
|
static uint64_t t1, t2, t3, t4, t5, t6, t8;
|
2020-12-25 00:11:38 +00:00
|
|
|
|
static const char *sox_, *ffplay_, *patharg_;
|
2020-06-15 14:18:57 +00:00
|
|
|
|
static struct VtFrame vtframe_[2], *f1_, *f2_;
|
|
|
|
|
|
static struct Graphic graphic_[2], *g1_, *g2_;
|
Make improvements
- Every unit test now passes on Apple Silicon. The final piece of this
puzzle was porting our POSIX threads cancelation support, since that
works differently on ARM64 XNU vs. AMD64. Our semaphore support on
Apple Silicon is also superior now compared to AMD64, thanks to the
grand central dispatch library which lets *NSYNC locks go faster.
- The Cosmopolitan runtime is now more stable, particularly on Windows.
To do this, thread local storage is mandatory at all runtime levels,
and the innermost packages of the C library is no longer being built
using ASAN. TLS is being bootstrapped with a 128-byte TIB during the
process startup phase, and then later on the runtime re-allocates it
either statically or dynamically to support code using _Thread_local.
fork() and execve() now do a better job cooperating with threads. We
can now check how much stack memory is left in the process or thread
when functions like kprintf() / execve() etc. call alloca(), so that
ENOMEM can be raised, reduce a buffer size, or just print a warning.
- POSIX signal emulation is now implemented the same way kernels do it
with pthread_kill() and raise(). Any thread can interrupt any other
thread, regardless of what it's doing. If it's blocked on read/write
then the killer thread will cancel its i/o operation so that EINTR can
be returned in the mark thread immediately. If it's doing a tight CPU
bound operation, then that's also interrupted by the signal delivery.
Signal delivery works now by suspending a thread and pushing context
data structures onto its stack, and redirecting its execution to a
trampoline function, which calls SetThreadContext(GetCurrentThread())
when it's done.
- We're now doing a better job managing locks and handles. On NetBSD we
now close semaphore file descriptors in forked children. Semaphores on
Windows can now be canceled immediately, which means mutexes/condition
variables will now go faster. Apple Silicon semaphores can be canceled
too. We're now using Apple's pthread_yield() funciton. Apple _nocancel
syscalls are now used on XNU when appropriate to ensure pthread_cancel
requests aren't lost. The MbedTLS library has been updated to support
POSIX thread cancelations. See tool/build/runitd.c for an example of
how it can be used for production multi-threaded tls servers. Handles
on Windows now leak less often across processes. All i/o operations on
Windows are now overlapped, which means file pointers can no longer be
inherited across dup() and fork() for the time being.
- We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4()
which means, for example, that posix_spawn() now goes 3x faster. POSIX
spawn is also now more correct. Like Musl, it's now able to report the
failure code of execve() via a pipe although our approach favors using
shared memory to do that on systems that have a true vfork() function.
- We now spawn a thread to deliver SIGALRM to threads when setitimer()
is used. This enables the most precise wakeups the OS makes possible.
- The Cosmopolitan runtime now uses less memory. On NetBSD for example,
it turned out the kernel would actually commit the PT_GNU_STACK size
which caused RSS to be 6mb for every process. Now it's down to ~4kb.
On Apple Silicon, we reduce the mandatory upstream thread size to the
smallest possible size to reduce the memory overhead of Cosmo threads.
The examples directory has a program called greenbean which can spawn
a web server on Linux with 10,000 worker threads and have the memory
usage of the process be ~77mb. The 1024 byte overhead of POSIX-style
thread-local storage is now optional; it won't be allocated until the
pthread_setspecific/getspecific functions are called. On Windows, the
threads that get spawned which are internal to the libc implementation
use reserve rather than commit memory, which shaves a few hundred kb.
- sigaltstack() is now supported on Windows, however it's currently not
able to be used to handle stack overflows, since crash signals are
still generated by WIN32. However the crash handler will still switch
to the alt stack, which is helpful in environments with tiny threads.
- Test binaries are now smaller. Many of the mandatory dependencies of
the test runner have been removed. This ensures many programs can do a
better job only linking the the thing they're testing. This caused the
test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb
- long double is no longer used in the implementation details of libc,
except in the APIs that define it. The old code that used long double
for time (instead of struct timespec) has now been thoroughly removed.
- ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing
backtraces itself, it'll just print a command you can run on the shell
using our new `cosmoaddr2line` program to view the backtrace.
- Crash report signal handling now works in a much better way. Instead
of terminating the process, it now relies on SA_RESETHAND so that the
default SIG_IGN behavior can terminate the process if necessary.
- Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
|
|
|
|
static struct timespec deadline_, dura_, starttime_;
|
2020-06-15 14:18:57 +00:00
|
|
|
|
static bool yes_, stats_, dither_, ttymode_, istango_;
|
Make improvements
- Every unit test now passes on Apple Silicon. The final piece of this
puzzle was porting our POSIX threads cancelation support, since that
works differently on ARM64 XNU vs. AMD64. Our semaphore support on
Apple Silicon is also superior now compared to AMD64, thanks to the
grand central dispatch library which lets *NSYNC locks go faster.
- The Cosmopolitan runtime is now more stable, particularly on Windows.
To do this, thread local storage is mandatory at all runtime levels,
and the innermost packages of the C library is no longer being built
using ASAN. TLS is being bootstrapped with a 128-byte TIB during the
process startup phase, and then later on the runtime re-allocates it
either statically or dynamically to support code using _Thread_local.
fork() and execve() now do a better job cooperating with threads. We
can now check how much stack memory is left in the process or thread
when functions like kprintf() / execve() etc. call alloca(), so that
ENOMEM can be raised, reduce a buffer size, or just print a warning.
- POSIX signal emulation is now implemented the same way kernels do it
with pthread_kill() and raise(). Any thread can interrupt any other
thread, regardless of what it's doing. If it's blocked on read/write
then the killer thread will cancel its i/o operation so that EINTR can
be returned in the mark thread immediately. If it's doing a tight CPU
bound operation, then that's also interrupted by the signal delivery.
Signal delivery works now by suspending a thread and pushing context
data structures onto its stack, and redirecting its execution to a
trampoline function, which calls SetThreadContext(GetCurrentThread())
when it's done.
- We're now doing a better job managing locks and handles. On NetBSD we
now close semaphore file descriptors in forked children. Semaphores on
Windows can now be canceled immediately, which means mutexes/condition
variables will now go faster. Apple Silicon semaphores can be canceled
too. We're now using Apple's pthread_yield() funciton. Apple _nocancel
syscalls are now used on XNU when appropriate to ensure pthread_cancel
requests aren't lost. The MbedTLS library has been updated to support
POSIX thread cancelations. See tool/build/runitd.c for an example of
how it can be used for production multi-threaded tls servers. Handles
on Windows now leak less often across processes. All i/o operations on
Windows are now overlapped, which means file pointers can no longer be
inherited across dup() and fork() for the time being.
- We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4()
which means, for example, that posix_spawn() now goes 3x faster. POSIX
spawn is also now more correct. Like Musl, it's now able to report the
failure code of execve() via a pipe although our approach favors using
shared memory to do that on systems that have a true vfork() function.
- We now spawn a thread to deliver SIGALRM to threads when setitimer()
is used. This enables the most precise wakeups the OS makes possible.
- The Cosmopolitan runtime now uses less memory. On NetBSD for example,
it turned out the kernel would actually commit the PT_GNU_STACK size
which caused RSS to be 6mb for every process. Now it's down to ~4kb.
On Apple Silicon, we reduce the mandatory upstream thread size to the
smallest possible size to reduce the memory overhead of Cosmo threads.
The examples directory has a program called greenbean which can spawn
a web server on Linux with 10,000 worker threads and have the memory
usage of the process be ~77mb. The 1024 byte overhead of POSIX-style
thread-local storage is now optional; it won't be allocated until the
pthread_setspecific/getspecific functions are called. On Windows, the
threads that get spawned which are internal to the libc implementation
use reserve rather than commit memory, which shaves a few hundred kb.
- sigaltstack() is now supported on Windows, however it's currently not
able to be used to handle stack overflows, since crash signals are
still generated by WIN32. However the crash handler will still switch
to the alt stack, which is helpful in environments with tiny threads.
- Test binaries are now smaller. Many of the mandatory dependencies of
the test runner have been removed. This ensures many programs can do a
better job only linking the the thing they're testing. This caused the
test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb
- long double is no longer used in the implementation details of libc,
except in the APIs that define it. The old code that used long double
for time (instead of struct timespec) has now been thoroughly removed.
- ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing
backtraces itself, it'll just print a command you can run on the shell
using our new `cosmoaddr2line` program to view the backtrace.
- Crash report signal handling now works in a much better way. Instead
of terminating the process, it now relies on SA_RESETHAND so that the
default SIG_IGN behavior can terminate the process if necessary.
- Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
|
|
|
|
static struct timespec decode_start_, f1_start_, f2_start_;
|
2020-12-05 20:20:41 +00:00
|
|
|
|
static int16_t pcm_[PLM_AUDIO_SAMPLES_PER_FRAME * 2 / 8][8];
|
2020-06-15 14:18:57 +00:00
|
|
|
|
static int16_t pcmscale_[PLM_AUDIO_SAMPLES_PER_FRAME * 2 / 8][8];
|
2020-12-05 20:20:41 +00:00
|
|
|
|
static bool fullclear_, historyclear_, tuned_, yonly_, gotvideo_;
|
2023-09-02 03:49:13 +00:00
|
|
|
|
static int homerow_, lastrow_, playfd_, infd_, outfd_, speakerfails_;
|
|
|
|
|
|
static char status_[7][200], logpath_[PATH_MAX], fifopath_[PATH_MAX],
|
|
|
|
|
|
chansstr_[32], sratestr_[32];
|
2020-06-15 14:18:57 +00:00
|
|
|
|
|
|
|
|
|
|
static void OnCtrlC(void) {
|
|
|
|
|
|
longjmp(jb_, 1);
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static void OnResize(void) {
|
|
|
|
|
|
resized_ = true;
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static void OnSigPipe(void) {
|
|
|
|
|
|
piped_ = true;
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static void OnSigChld(void) {
|
|
|
|
|
|
playpid_ = 0, piped_ = true;
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static void StrikeDownCrapware(int sig) {
|
|
|
|
|
|
kill(playpid_, SIGKILL);
|
|
|
|
|
|
}
|
|
|
|
|
|
|
Make improvements
- Every unit test now passes on Apple Silicon. The final piece of this
puzzle was porting our POSIX threads cancelation support, since that
works differently on ARM64 XNU vs. AMD64. Our semaphore support on
Apple Silicon is also superior now compared to AMD64, thanks to the
grand central dispatch library which lets *NSYNC locks go faster.
- The Cosmopolitan runtime is now more stable, particularly on Windows.
To do this, thread local storage is mandatory at all runtime levels,
and the innermost packages of the C library is no longer being built
using ASAN. TLS is being bootstrapped with a 128-byte TIB during the
process startup phase, and then later on the runtime re-allocates it
either statically or dynamically to support code using _Thread_local.
fork() and execve() now do a better job cooperating with threads. We
can now check how much stack memory is left in the process or thread
when functions like kprintf() / execve() etc. call alloca(), so that
ENOMEM can be raised, reduce a buffer size, or just print a warning.
- POSIX signal emulation is now implemented the same way kernels do it
with pthread_kill() and raise(). Any thread can interrupt any other
thread, regardless of what it's doing. If it's blocked on read/write
then the killer thread will cancel its i/o operation so that EINTR can
be returned in the mark thread immediately. If it's doing a tight CPU
bound operation, then that's also interrupted by the signal delivery.
Signal delivery works now by suspending a thread and pushing context
data structures onto its stack, and redirecting its execution to a
trampoline function, which calls SetThreadContext(GetCurrentThread())
when it's done.
- We're now doing a better job managing locks and handles. On NetBSD we
now close semaphore file descriptors in forked children. Semaphores on
Windows can now be canceled immediately, which means mutexes/condition
variables will now go faster. Apple Silicon semaphores can be canceled
too. We're now using Apple's pthread_yield() funciton. Apple _nocancel
syscalls are now used on XNU when appropriate to ensure pthread_cancel
requests aren't lost. The MbedTLS library has been updated to support
POSIX thread cancelations. See tool/build/runitd.c for an example of
how it can be used for production multi-threaded tls servers. Handles
on Windows now leak less often across processes. All i/o operations on
Windows are now overlapped, which means file pointers can no longer be
inherited across dup() and fork() for the time being.
- We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4()
which means, for example, that posix_spawn() now goes 3x faster. POSIX
spawn is also now more correct. Like Musl, it's now able to report the
failure code of execve() via a pipe although our approach favors using
shared memory to do that on systems that have a true vfork() function.
- We now spawn a thread to deliver SIGALRM to threads when setitimer()
is used. This enables the most precise wakeups the OS makes possible.
- The Cosmopolitan runtime now uses less memory. On NetBSD for example,
it turned out the kernel would actually commit the PT_GNU_STACK size
which caused RSS to be 6mb for every process. Now it's down to ~4kb.
On Apple Silicon, we reduce the mandatory upstream thread size to the
smallest possible size to reduce the memory overhead of Cosmo threads.
The examples directory has a program called greenbean which can spawn
a web server on Linux with 10,000 worker threads and have the memory
usage of the process be ~77mb. The 1024 byte overhead of POSIX-style
thread-local storage is now optional; it won't be allocated until the
pthread_setspecific/getspecific functions are called. On Windows, the
threads that get spawned which are internal to the libc implementation
use reserve rather than commit memory, which shaves a few hundred kb.
- sigaltstack() is now supported on Windows, however it's currently not
able to be used to handle stack overflows, since crash signals are
still generated by WIN32. However the crash handler will still switch
to the alt stack, which is helpful in environments with tiny threads.
- Test binaries are now smaller. Many of the mandatory dependencies of
the test runner have been removed. This ensures many programs can do a
better job only linking the the thing they're testing. This caused the
test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb
- long double is no longer used in the implementation details of libc,
except in the APIs that define it. The old code that used long double
for time (instead of struct timespec) has now been thoroughly removed.
- ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing
backtraces itself, it'll just print a command you can run on the shell
using our new `cosmoaddr2line` program to view the backtrace.
- Crash report signal handling now works in a much better way. Instead
of terminating the process, it now relies on SA_RESETHAND so that the
default SIG_IGN behavior can terminate the process if necessary.
- Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
|
|
|
|
static struct timespec GetGraceTime(void) {
|
|
|
|
|
|
return timespec_sub(deadline_, timespec_real());
|
2020-06-15 14:18:57 +00:00
|
|
|
|
}
|
|
|
|
|
|
|
2023-10-14 10:06:11 +00:00
|
|
|
|
static char *strntoupper(char *s, size_t n) {
|
|
|
|
|
|
size_t i;
|
|
|
|
|
|
for (i = 0; s[i] && i < n; ++i) {
|
|
|
|
|
|
if ('a' <= s[i] && s[i] <= 'z') {
|
|
|
|
|
|
s[i] -= 'a' - 'A';
|
|
|
|
|
|
}
|
|
|
|
|
|
}
|
|
|
|
|
|
return s;
|
|
|
|
|
|
}
|
|
|
|
|
|
|
2020-06-15 14:18:57 +00:00
|
|
|
|
static int GetNamedVector(const struct NamedVector *choices, size_t n,
|
|
|
|
|
|
const char *s) {
|
|
|
|
|
|
int i;
|
|
|
|
|
|
char name[sizeof(choices->name)];
|
Make improvements
- Every unit test now passes on Apple Silicon. The final piece of this
puzzle was porting our POSIX threads cancelation support, since that
works differently on ARM64 XNU vs. AMD64. Our semaphore support on
Apple Silicon is also superior now compared to AMD64, thanks to the
grand central dispatch library which lets *NSYNC locks go faster.
- The Cosmopolitan runtime is now more stable, particularly on Windows.
To do this, thread local storage is mandatory at all runtime levels,
and the innermost packages of the C library is no longer being built
using ASAN. TLS is being bootstrapped with a 128-byte TIB during the
process startup phase, and then later on the runtime re-allocates it
either statically or dynamically to support code using _Thread_local.
fork() and execve() now do a better job cooperating with threads. We
can now check how much stack memory is left in the process or thread
when functions like kprintf() / execve() etc. call alloca(), so that
ENOMEM can be raised, reduce a buffer size, or just print a warning.
- POSIX signal emulation is now implemented the same way kernels do it
with pthread_kill() and raise(). Any thread can interrupt any other
thread, regardless of what it's doing. If it's blocked on read/write
then the killer thread will cancel its i/o operation so that EINTR can
be returned in the mark thread immediately. If it's doing a tight CPU
bound operation, then that's also interrupted by the signal delivery.
Signal delivery works now by suspending a thread and pushing context
data structures onto its stack, and redirecting its execution to a
trampoline function, which calls SetThreadContext(GetCurrentThread())
when it's done.
- We're now doing a better job managing locks and handles. On NetBSD we
now close semaphore file descriptors in forked children. Semaphores on
Windows can now be canceled immediately, which means mutexes/condition
variables will now go faster. Apple Silicon semaphores can be canceled
too. We're now using Apple's pthread_yield() funciton. Apple _nocancel
syscalls are now used on XNU when appropriate to ensure pthread_cancel
requests aren't lost. The MbedTLS library has been updated to support
POSIX thread cancelations. See tool/build/runitd.c for an example of
how it can be used for production multi-threaded tls servers. Handles
on Windows now leak less often across processes. All i/o operations on
Windows are now overlapped, which means file pointers can no longer be
inherited across dup() and fork() for the time being.
- We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4()
which means, for example, that posix_spawn() now goes 3x faster. POSIX
spawn is also now more correct. Like Musl, it's now able to report the
failure code of execve() via a pipe although our approach favors using
shared memory to do that on systems that have a true vfork() function.
- We now spawn a thread to deliver SIGALRM to threads when setitimer()
is used. This enables the most precise wakeups the OS makes possible.
- The Cosmopolitan runtime now uses less memory. On NetBSD for example,
it turned out the kernel would actually commit the PT_GNU_STACK size
which caused RSS to be 6mb for every process. Now it's down to ~4kb.
On Apple Silicon, we reduce the mandatory upstream thread size to the
smallest possible size to reduce the memory overhead of Cosmo threads.
The examples directory has a program called greenbean which can spawn
a web server on Linux with 10,000 worker threads and have the memory
usage of the process be ~77mb. The 1024 byte overhead of POSIX-style
thread-local storage is now optional; it won't be allocated until the
pthread_setspecific/getspecific functions are called. On Windows, the
threads that get spawned which are internal to the libc implementation
use reserve rather than commit memory, which shaves a few hundred kb.
- sigaltstack() is now supported on Windows, however it's currently not
able to be used to handle stack overflows, since crash signals are
still generated by WIN32. However the crash handler will still switch
to the alt stack, which is helpful in environments with tiny threads.
- Test binaries are now smaller. Many of the mandatory dependencies of
the test runner have been removed. This ensures many programs can do a
better job only linking the the thing they're testing. This caused the
test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb
- long double is no longer used in the implementation details of libc,
except in the APIs that define it. The old code that used long double
for time (instead of struct timespec) has now been thoroughly removed.
- ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing
backtraces itself, it'll just print a command you can run on the shell
using our new `cosmoaddr2line` program to view the backtrace.
- Crash report signal handling now works in a much better way. Instead
of terminating the process, it now relies on SA_RESETHAND so that the
default SIG_IGN behavior can terminate the process if necessary.
- Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
|
|
|
|
#pragma GCC push_options
|
|
|
|
|
|
#pragma GCC diagnostic ignored "-Wstringop-truncation"
|
|
|
|
|
|
strncpy(name, s, sizeof(name));
|
|
|
|
|
|
#pragma GCC pop_options
|
2020-06-15 14:18:57 +00:00
|
|
|
|
strntoupper(name, sizeof(name));
|
|
|
|
|
|
for (i = 0; i < n; ++i) {
|
|
|
|
|
|
if (memcmp(choices[i].name, name, sizeof(name)) == 0) {
|
|
|
|
|
|
return i;
|
|
|
|
|
|
}
|
|
|
|
|
|
}
|
|
|
|
|
|
return -1;
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static int GetPrimaries(const char *s) {
|
|
|
|
|
|
return GetNamedVector(kPrimaries, ARRAYLEN(kPrimaries), s);
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static int GetLighting(const char *s) {
|
|
|
|
|
|
return GetNamedVector(kLightings, ARRAYLEN(kLightings), s);
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static bool CloseSpeaker(void) {
|
|
|
|
|
|
int rc, wstatus;
|
|
|
|
|
|
rc = 0;
|
Make improvements
- Every unit test now passes on Apple Silicon. The final piece of this
puzzle was porting our POSIX threads cancelation support, since that
works differently on ARM64 XNU vs. AMD64. Our semaphore support on
Apple Silicon is also superior now compared to AMD64, thanks to the
grand central dispatch library which lets *NSYNC locks go faster.
- The Cosmopolitan runtime is now more stable, particularly on Windows.
To do this, thread local storage is mandatory at all runtime levels,
and the innermost packages of the C library is no longer being built
using ASAN. TLS is being bootstrapped with a 128-byte TIB during the
process startup phase, and then later on the runtime re-allocates it
either statically or dynamically to support code using _Thread_local.
fork() and execve() now do a better job cooperating with threads. We
can now check how much stack memory is left in the process or thread
when functions like kprintf() / execve() etc. call alloca(), so that
ENOMEM can be raised, reduce a buffer size, or just print a warning.
- POSIX signal emulation is now implemented the same way kernels do it
with pthread_kill() and raise(). Any thread can interrupt any other
thread, regardless of what it's doing. If it's blocked on read/write
then the killer thread will cancel its i/o operation so that EINTR can
be returned in the mark thread immediately. If it's doing a tight CPU
bound operation, then that's also interrupted by the signal delivery.
Signal delivery works now by suspending a thread and pushing context
data structures onto its stack, and redirecting its execution to a
trampoline function, which calls SetThreadContext(GetCurrentThread())
when it's done.
- We're now doing a better job managing locks and handles. On NetBSD we
now close semaphore file descriptors in forked children. Semaphores on
Windows can now be canceled immediately, which means mutexes/condition
variables will now go faster. Apple Silicon semaphores can be canceled
too. We're now using Apple's pthread_yield() funciton. Apple _nocancel
syscalls are now used on XNU when appropriate to ensure pthread_cancel
requests aren't lost. The MbedTLS library has been updated to support
POSIX thread cancelations. See tool/build/runitd.c for an example of
how it can be used for production multi-threaded tls servers. Handles
on Windows now leak less often across processes. All i/o operations on
Windows are now overlapped, which means file pointers can no longer be
inherited across dup() and fork() for the time being.
- We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4()
which means, for example, that posix_spawn() now goes 3x faster. POSIX
spawn is also now more correct. Like Musl, it's now able to report the
failure code of execve() via a pipe although our approach favors using
shared memory to do that on systems that have a true vfork() function.
- We now spawn a thread to deliver SIGALRM to threads when setitimer()
is used. This enables the most precise wakeups the OS makes possible.
- The Cosmopolitan runtime now uses less memory. On NetBSD for example,
it turned out the kernel would actually commit the PT_GNU_STACK size
which caused RSS to be 6mb for every process. Now it's down to ~4kb.
On Apple Silicon, we reduce the mandatory upstream thread size to the
smallest possible size to reduce the memory overhead of Cosmo threads.
The examples directory has a program called greenbean which can spawn
a web server on Linux with 10,000 worker threads and have the memory
usage of the process be ~77mb. The 1024 byte overhead of POSIX-style
thread-local storage is now optional; it won't be allocated until the
pthread_setspecific/getspecific functions are called. On Windows, the
threads that get spawned which are internal to the libc implementation
use reserve rather than commit memory, which shaves a few hundred kb.
- sigaltstack() is now supported on Windows, however it's currently not
able to be used to handle stack overflows, since crash signals are
still generated by WIN32. However the crash handler will still switch
to the alt stack, which is helpful in environments with tiny threads.
- Test binaries are now smaller. Many of the mandatory dependencies of
the test runner have been removed. This ensures many programs can do a
better job only linking the the thing they're testing. This caused the
test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb
- long double is no longer used in the implementation details of libc,
except in the APIs that define it. The old code that used long double
for time (instead of struct timespec) has now been thoroughly removed.
- ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing
backtraces itself, it'll just print a command you can run on the shell
using our new `cosmoaddr2line` program to view the backtrace.
- Crash report signal handling now works in a much better way. Instead
of terminating the process, it now relies on SA_RESETHAND so that the
default SIG_IGN behavior can terminate the process if necessary.
- Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
|
|
|
|
pthread_yield();
|
2020-06-15 14:18:57 +00:00
|
|
|
|
if (playfd_) {
|
|
|
|
|
|
rc |= close(playfd_);
|
|
|
|
|
|
playfd_ = -1;
|
|
|
|
|
|
}
|
|
|
|
|
|
if (playpid_) {
|
|
|
|
|
|
kill(playpid_, SIGTERM);
|
|
|
|
|
|
xsigaction(SIGALRM, StrikeDownCrapware, SA_RESETHAND, 0, 0);
|
2022-07-20 21:01:15 +00:00
|
|
|
|
setitimer(ITIMER_REAL, &kTimerHalfSecondSingleShot, NULL);
|
2020-06-15 14:18:57 +00:00
|
|
|
|
while (playpid_) {
|
|
|
|
|
|
if (waitpid(playpid_, &wstatus, 0) != -1) {
|
|
|
|
|
|
rc |= WEXITSTATUS(wstatus);
|
|
|
|
|
|
} else if (errno == EINTR) {
|
|
|
|
|
|
continue;
|
|
|
|
|
|
} else {
|
|
|
|
|
|
rc = -1;
|
|
|
|
|
|
}
|
|
|
|
|
|
break;
|
|
|
|
|
|
}
|
|
|
|
|
|
playpid_ = 0;
|
|
|
|
|
|
setitimer(ITIMER_REAL, &kTimerDisarm, NULL);
|
|
|
|
|
|
}
|
|
|
|
|
|
return !!rc;
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static void ResizeVtFrame(struct VtFrame *f, size_t yn, size_t xn) {
|
2023-07-30 01:44:15 +00:00
|
|
|
|
BALLOC(&f->b, 4096, 64 + yn * (xn * 32 + 8), __FUNCTION__);
|
2020-06-15 14:18:57 +00:00
|
|
|
|
f->i = f->n = 0;
|
|
|
|
|
|
}
|
|
|
|
|
|
|
Make improvements
- Every unit test now passes on Apple Silicon. The final piece of this
puzzle was porting our POSIX threads cancelation support, since that
works differently on ARM64 XNU vs. AMD64. Our semaphore support on
Apple Silicon is also superior now compared to AMD64, thanks to the
grand central dispatch library which lets *NSYNC locks go faster.
- The Cosmopolitan runtime is now more stable, particularly on Windows.
To do this, thread local storage is mandatory at all runtime levels,
and the innermost packages of the C library is no longer being built
using ASAN. TLS is being bootstrapped with a 128-byte TIB during the
process startup phase, and then later on the runtime re-allocates it
either statically or dynamically to support code using _Thread_local.
fork() and execve() now do a better job cooperating with threads. We
can now check how much stack memory is left in the process or thread
when functions like kprintf() / execve() etc. call alloca(), so that
ENOMEM can be raised, reduce a buffer size, or just print a warning.
- POSIX signal emulation is now implemented the same way kernels do it
with pthread_kill() and raise(). Any thread can interrupt any other
thread, regardless of what it's doing. If it's blocked on read/write
then the killer thread will cancel its i/o operation so that EINTR can
be returned in the mark thread immediately. If it's doing a tight CPU
bound operation, then that's also interrupted by the signal delivery.
Signal delivery works now by suspending a thread and pushing context
data structures onto its stack, and redirecting its execution to a
trampoline function, which calls SetThreadContext(GetCurrentThread())
when it's done.
- We're now doing a better job managing locks and handles. On NetBSD we
now close semaphore file descriptors in forked children. Semaphores on
Windows can now be canceled immediately, which means mutexes/condition
variables will now go faster. Apple Silicon semaphores can be canceled
too. We're now using Apple's pthread_yield() funciton. Apple _nocancel
syscalls are now used on XNU when appropriate to ensure pthread_cancel
requests aren't lost. The MbedTLS library has been updated to support
POSIX thread cancelations. See tool/build/runitd.c for an example of
how it can be used for production multi-threaded tls servers. Handles
on Windows now leak less often across processes. All i/o operations on
Windows are now overlapped, which means file pointers can no longer be
inherited across dup() and fork() for the time being.
- We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4()
which means, for example, that posix_spawn() now goes 3x faster. POSIX
spawn is also now more correct. Like Musl, it's now able to report the
failure code of execve() via a pipe although our approach favors using
shared memory to do that on systems that have a true vfork() function.
- We now spawn a thread to deliver SIGALRM to threads when setitimer()
is used. This enables the most precise wakeups the OS makes possible.
- The Cosmopolitan runtime now uses less memory. On NetBSD for example,
it turned out the kernel would actually commit the PT_GNU_STACK size
which caused RSS to be 6mb for every process. Now it's down to ~4kb.
On Apple Silicon, we reduce the mandatory upstream thread size to the
smallest possible size to reduce the memory overhead of Cosmo threads.
The examples directory has a program called greenbean which can spawn
a web server on Linux with 10,000 worker threads and have the memory
usage of the process be ~77mb. The 1024 byte overhead of POSIX-style
thread-local storage is now optional; it won't be allocated until the
pthread_setspecific/getspecific functions are called. On Windows, the
threads that get spawned which are internal to the libc implementation
use reserve rather than commit memory, which shaves a few hundred kb.
- sigaltstack() is now supported on Windows, however it's currently not
able to be used to handle stack overflows, since crash signals are
still generated by WIN32. However the crash handler will still switch
to the alt stack, which is helpful in environments with tiny threads.
- Test binaries are now smaller. Many of the mandatory dependencies of
the test runner have been removed. This ensures many programs can do a
better job only linking the the thing they're testing. This caused the
test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb
- long double is no longer used in the implementation details of libc,
except in the APIs that define it. The old code that used long double
for time (instead of struct timespec) has now been thoroughly removed.
- ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing
backtraces itself, it'll just print a command you can run on the shell
using our new `cosmoaddr2line` program to view the backtrace.
- Crash report signal handling now works in a much better way. Instead
of terminating the process, it now relies on SA_RESETHAND so that the
default SIG_IGN behavior can terminate the process if necessary.
- Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
|
|
|
|
static float timespec_tofloat(struct timespec ts) {
|
|
|
|
|
|
return ts.tv_sec + ts.tv_nsec * 1e-9;
|
|
|
|
|
|
}
|
|
|
|
|
|
|
2020-06-15 14:18:57 +00:00
|
|
|
|
static void RecordFactThatFrameWasFullyRendered(void) {
|
Make improvements
- Every unit test now passes on Apple Silicon. The final piece of this
puzzle was porting our POSIX threads cancelation support, since that
works differently on ARM64 XNU vs. AMD64. Our semaphore support on
Apple Silicon is also superior now compared to AMD64, thanks to the
grand central dispatch library which lets *NSYNC locks go faster.
- The Cosmopolitan runtime is now more stable, particularly on Windows.
To do this, thread local storage is mandatory at all runtime levels,
and the innermost packages of the C library is no longer being built
using ASAN. TLS is being bootstrapped with a 128-byte TIB during the
process startup phase, and then later on the runtime re-allocates it
either statically or dynamically to support code using _Thread_local.
fork() and execve() now do a better job cooperating with threads. We
can now check how much stack memory is left in the process or thread
when functions like kprintf() / execve() etc. call alloca(), so that
ENOMEM can be raised, reduce a buffer size, or just print a warning.
- POSIX signal emulation is now implemented the same way kernels do it
with pthread_kill() and raise(). Any thread can interrupt any other
thread, regardless of what it's doing. If it's blocked on read/write
then the killer thread will cancel its i/o operation so that EINTR can
be returned in the mark thread immediately. If it's doing a tight CPU
bound operation, then that's also interrupted by the signal delivery.
Signal delivery works now by suspending a thread and pushing context
data structures onto its stack, and redirecting its execution to a
trampoline function, which calls SetThreadContext(GetCurrentThread())
when it's done.
- We're now doing a better job managing locks and handles. On NetBSD we
now close semaphore file descriptors in forked children. Semaphores on
Windows can now be canceled immediately, which means mutexes/condition
variables will now go faster. Apple Silicon semaphores can be canceled
too. We're now using Apple's pthread_yield() funciton. Apple _nocancel
syscalls are now used on XNU when appropriate to ensure pthread_cancel
requests aren't lost. The MbedTLS library has been updated to support
POSIX thread cancelations. See tool/build/runitd.c for an example of
how it can be used for production multi-threaded tls servers. Handles
on Windows now leak less often across processes. All i/o operations on
Windows are now overlapped, which means file pointers can no longer be
inherited across dup() and fork() for the time being.
- We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4()
which means, for example, that posix_spawn() now goes 3x faster. POSIX
spawn is also now more correct. Like Musl, it's now able to report the
failure code of execve() via a pipe although our approach favors using
shared memory to do that on systems that have a true vfork() function.
- We now spawn a thread to deliver SIGALRM to threads when setitimer()
is used. This enables the most precise wakeups the OS makes possible.
- The Cosmopolitan runtime now uses less memory. On NetBSD for example,
it turned out the kernel would actually commit the PT_GNU_STACK size
which caused RSS to be 6mb for every process. Now it's down to ~4kb.
On Apple Silicon, we reduce the mandatory upstream thread size to the
smallest possible size to reduce the memory overhead of Cosmo threads.
The examples directory has a program called greenbean which can spawn
a web server on Linux with 10,000 worker threads and have the memory
usage of the process be ~77mb. The 1024 byte overhead of POSIX-style
thread-local storage is now optional; it won't be allocated until the
pthread_setspecific/getspecific functions are called. On Windows, the
threads that get spawned which are internal to the libc implementation
use reserve rather than commit memory, which shaves a few hundred kb.
- sigaltstack() is now supported on Windows, however it's currently not
able to be used to handle stack overflows, since crash signals are
still generated by WIN32. However the crash handler will still switch
to the alt stack, which is helpful in environments with tiny threads.
- Test binaries are now smaller. Many of the mandatory dependencies of
the test runner have been removed. This ensures many programs can do a
better job only linking the the thing they're testing. This caused the
test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb
- long double is no longer used in the implementation details of libc,
except in the APIs that define it. The old code that used long double
for time (instead of struct timespec) has now been thoroughly removed.
- ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing
backtraces itself, it'll just print a command you can run on the shell
using our new `cosmoaddr2line` program to view the backtrace.
- Crash report signal handling now works in a much better way. Instead
of terminating the process, it now relies on SA_RESETHAND so that the
default SIG_IGN behavior can terminate the process if necessary.
- Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
|
|
|
|
fcring_.p[fcring_.i] =
|
|
|
|
|
|
timespec_tofloat(timespec_sub(timespec_real(), starttime_));
|
2020-06-15 14:18:57 +00:00
|
|
|
|
fcring_.n += 1;
|
|
|
|
|
|
fcring_.i += 1;
|
|
|
|
|
|
fcring_.i &= ARRAYLEN(fcring_.p) - 1;
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static double MeasureFrameRate(void) {
|
|
|
|
|
|
int i, j, n, m;
|
|
|
|
|
|
if (fcring_.n) {
|
|
|
|
|
|
m = ARRAYLEN(fcring_.p);
|
|
|
|
|
|
n = MIN(fcring_.n, m);
|
|
|
|
|
|
i = (fcring_.i - 1) & (m - 1);
|
|
|
|
|
|
j = (fcring_.i - n) & (m - 1);
|
|
|
|
|
|
return n / (fcring_.p[i] - fcring_.p[j]);
|
|
|
|
|
|
} else {
|
|
|
|
|
|
return 0;
|
|
|
|
|
|
}
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static bool ShouldUseFrameBuffer(void) {
|
|
|
|
|
|
return fb0_.fd != -1;
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static bool IsHighDefinition(long yn, long xn) {
|
|
|
|
|
|
return yn * xn >= 1280 * 720;
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static void ComputeColoringSolution(void) {
|
|
|
|
|
|
YCbCrInit(&ycbcr_, yonly_, swing_, gamma_, *kPrimaries[primaries_].vector,
|
|
|
|
|
|
*kLightings[lighting_].vector);
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static void DimensionDisplay(void) {
|
|
|
|
|
|
size_t yn, xn;
|
|
|
|
|
|
double ratio, height, width;
|
|
|
|
|
|
do {
|
|
|
|
|
|
resized_ = false;
|
|
|
|
|
|
if (ShouldUseFrameBuffer()) {
|
|
|
|
|
|
pary_ = 1;
|
|
|
|
|
|
parx_ = 1;
|
|
|
|
|
|
dh_ = fb0_.vscreen.yres;
|
|
|
|
|
|
dw_ = fb0_.vscreen.xres;
|
|
|
|
|
|
wsize_.ws_row = fb0_.vscreen.yres_virtual;
|
|
|
|
|
|
wsize_.ws_col = fb0_.vscreen.xres_virtual;
|
|
|
|
|
|
} else {
|
|
|
|
|
|
pary_ = 1;
|
|
|
|
|
|
parx_ = 1;
|
|
|
|
|
|
wsize_.ws_row = 25;
|
|
|
|
|
|
wsize_.ws_col = 80;
|
|
|
|
|
|
wsize_ = (struct winsize){.ws_row = 40, .ws_col = 80};
|
2023-06-15 00:02:57 +00:00
|
|
|
|
if (tcgetwinsize(outfd_, &wsize_) == -1) {
|
|
|
|
|
|
tcgetwinsize(0, &wsize_);
|
|
|
|
|
|
}
|
2020-06-15 14:18:57 +00:00
|
|
|
|
dh_ = wsize_.ws_row * 2;
|
|
|
|
|
|
dw_ = wsize_.ws_col * 2;
|
|
|
|
|
|
}
|
|
|
|
|
|
ratio = g1_->xn;
|
|
|
|
|
|
ratio /= g1_->yn;
|
|
|
|
|
|
height = dh_;
|
|
|
|
|
|
width = dw_;
|
|
|
|
|
|
height = MIN(height, height * ratio);
|
|
|
|
|
|
width = MIN(width, width * ratio);
|
|
|
|
|
|
yn = height;
|
|
|
|
|
|
xn = width;
|
|
|
|
|
|
yn = ROUNDDOWN(yn, 2);
|
|
|
|
|
|
xn = ROUNDDOWN(xn, 2);
|
|
|
|
|
|
g2_ = resizegraphic(&graphic_[1], yn, xn);
|
2021-09-04 04:14:26 +00:00
|
|
|
|
INFOF("%s 𝑑(%hu×%hu)×(%d,%d): 𝑔₁(%zu×%zu,r=%f) → 𝑔₂(%zu×%zu)",
|
2022-03-18 11:20:10 +00:00
|
|
|
|
"DimensionDisplay", wsize_.ws_row, wsize_.ws_col, g1_->yn, g1_->xn,
|
|
|
|
|
|
ratio, yn, xn);
|
2020-06-15 14:18:57 +00:00
|
|
|
|
BALLOC(&xtcodes_, 64, ((g2_->yn) * g2_->xn + 8) * sizeof(struct TtyRgb),
|
|
|
|
|
|
"xtcodes_");
|
|
|
|
|
|
ResizeVtFrame(&vtframe_[0], (g2_->yn), g2_->xn);
|
|
|
|
|
|
ResizeVtFrame(&vtframe_[1], (g2_->yn), g2_->xn);
|
|
|
|
|
|
f1_ = &vtframe_[0];
|
|
|
|
|
|
f2_ = &vtframe_[1];
|
|
|
|
|
|
if (ttymode_) {
|
|
|
|
|
|
homerow_ = MIN(wsize_.ws_row - HALF(g2_->yn),
|
|
|
|
|
|
HALF(wsize_.ws_row - HALF(g2_->yn)));
|
|
|
|
|
|
}
|
|
|
|
|
|
lastrow_ = homerow_ + HALF(g2_->yn);
|
|
|
|
|
|
ComputeColoringSolution();
|
|
|
|
|
|
} while (resized_);
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static int WriteAudio(int fd, const void *data, size_t size, int deadlinems) {
|
|
|
|
|
|
ssize_t rc;
|
|
|
|
|
|
const char *p;
|
|
|
|
|
|
size_t wrote, n;
|
|
|
|
|
|
p = data;
|
|
|
|
|
|
n = size;
|
|
|
|
|
|
do {
|
|
|
|
|
|
TryAgain:
|
|
|
|
|
|
if ((rc = write(fd, p, n)) != -1) {
|
|
|
|
|
|
wrote = rc;
|
|
|
|
|
|
p += wrote;
|
|
|
|
|
|
n -= wrote;
|
|
|
|
|
|
} else if (errno == EINTR) {
|
|
|
|
|
|
goto TryAgain;
|
|
|
|
|
|
} else if (errno == EAGAIN) {
|
|
|
|
|
|
if (poll((struct pollfd[]){{fd, POLLOUT}}, 1, deadlinems) == 0) {
|
|
|
|
|
|
return etimedout();
|
|
|
|
|
|
}
|
|
|
|
|
|
} else {
|
|
|
|
|
|
return -1;
|
|
|
|
|
|
}
|
|
|
|
|
|
} while (n);
|
|
|
|
|
|
return 0;
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static bool TrySpeaker(const char *prog, char *const *args) {
|
2021-01-25 21:08:05 +00:00
|
|
|
|
int pipefds[2];
|
|
|
|
|
|
CHECK_NE(-1, pipe2(pipefds, O_CLOEXEC));
|
|
|
|
|
|
if (!(playpid_ = fork())) {
|
|
|
|
|
|
dup2(pipefds[0], 0);
|
2021-03-01 07:42:35 +00:00
|
|
|
|
dup2(fileno(__log_file), 1);
|
|
|
|
|
|
dup2(fileno(__log_file), 2);
|
|
|
|
|
|
close(fileno(__log_file));
|
2021-01-25 21:08:05 +00:00
|
|
|
|
execv(prog, args);
|
|
|
|
|
|
abort();
|
2020-06-15 14:18:57 +00:00
|
|
|
|
}
|
2021-01-25 21:08:05 +00:00
|
|
|
|
playfd_ = pipefds[1];
|
|
|
|
|
|
return true;
|
2020-06-15 14:18:57 +00:00
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static bool TrySox(void) {
|
|
|
|
|
|
return TrySpeaker(sox_, ARGZ("play", "-q", "-c", chansstr_, "-traw",
|
|
|
|
|
|
"-esigned", "-b16", "-r", sratestr_, "-"));
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static bool TryFfplay(void) {
|
|
|
|
|
|
return TrySpeaker(ffplay_, ARGZ("ffplay", "-nodisp", "-loglevel", "quiet",
|
|
|
|
|
|
"-fflags", "nobuffer", "-ac", chansstr_,
|
|
|
|
|
|
"-ar", sratestr_, "-f", "s16le", "pipe:"));
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static bool OpenSpeaker(void) {
|
|
|
|
|
|
size_t i;
|
|
|
|
|
|
static bool once, count;
|
|
|
|
|
|
if (!once) {
|
|
|
|
|
|
once = true;
|
|
|
|
|
|
i = 0;
|
Apply clang-format update to repo (#1154)
Commit bc6c183 introduced a bunch of discrepancies between what files
look like in the repo and what clang-format says they should look like.
However, there were already a few discrepancies prior to that. Most of
these discrepancies seemed to be unintentional, but a few of them were
load-bearing (e.g., a #include that violated header ordering needing
something to have been #defined by a 'later' #include.)
I opted to take what I hope is a relatively smooth-brained approach: I
reverted the .clang-format change, ran clang-format on the whole repo,
reapplied the .clang-format change, reran clang-format again, and then
reverted the commit that contained the first run. Thus the full effect
of this PR should only be to apply the changed formatting rules to the
repo, and from skimming the results, this seems to be the case.
My work can be checked by applying the short, manual commits, and then
rerunning the command listed in the autogenerated commits (those whose
messages I have prefixed auto:) and seeing if your results agree.
It might be that the other diffs should be fixed at some point but I'm
leaving that aside for now.
fd '\.c(c|pp)?$' --print0| xargs -0 clang-format -i
2024-04-25 17:38:00 +00:00
|
|
|
|
if (ffplay_)
|
|
|
|
|
|
tryspeakerfns_[i++] = TryFfplay;
|
|
|
|
|
|
if (sox_)
|
|
|
|
|
|
tryspeakerfns_[i++] = TrySox;
|
2020-06-15 14:18:57 +00:00
|
|
|
|
}
|
Make improvements
- Every unit test now passes on Apple Silicon. The final piece of this
puzzle was porting our POSIX threads cancelation support, since that
works differently on ARM64 XNU vs. AMD64. Our semaphore support on
Apple Silicon is also superior now compared to AMD64, thanks to the
grand central dispatch library which lets *NSYNC locks go faster.
- The Cosmopolitan runtime is now more stable, particularly on Windows.
To do this, thread local storage is mandatory at all runtime levels,
and the innermost packages of the C library is no longer being built
using ASAN. TLS is being bootstrapped with a 128-byte TIB during the
process startup phase, and then later on the runtime re-allocates it
either statically or dynamically to support code using _Thread_local.
fork() and execve() now do a better job cooperating with threads. We
can now check how much stack memory is left in the process or thread
when functions like kprintf() / execve() etc. call alloca(), so that
ENOMEM can be raised, reduce a buffer size, or just print a warning.
- POSIX signal emulation is now implemented the same way kernels do it
with pthread_kill() and raise(). Any thread can interrupt any other
thread, regardless of what it's doing. If it's blocked on read/write
then the killer thread will cancel its i/o operation so that EINTR can
be returned in the mark thread immediately. If it's doing a tight CPU
bound operation, then that's also interrupted by the signal delivery.
Signal delivery works now by suspending a thread and pushing context
data structures onto its stack, and redirecting its execution to a
trampoline function, which calls SetThreadContext(GetCurrentThread())
when it's done.
- We're now doing a better job managing locks and handles. On NetBSD we
now close semaphore file descriptors in forked children. Semaphores on
Windows can now be canceled immediately, which means mutexes/condition
variables will now go faster. Apple Silicon semaphores can be canceled
too. We're now using Apple's pthread_yield() funciton. Apple _nocancel
syscalls are now used on XNU when appropriate to ensure pthread_cancel
requests aren't lost. The MbedTLS library has been updated to support
POSIX thread cancelations. See tool/build/runitd.c for an example of
how it can be used for production multi-threaded tls servers. Handles
on Windows now leak less often across processes. All i/o operations on
Windows are now overlapped, which means file pointers can no longer be
inherited across dup() and fork() for the time being.
- We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4()
which means, for example, that posix_spawn() now goes 3x faster. POSIX
spawn is also now more correct. Like Musl, it's now able to report the
failure code of execve() via a pipe although our approach favors using
shared memory to do that on systems that have a true vfork() function.
- We now spawn a thread to deliver SIGALRM to threads when setitimer()
is used. This enables the most precise wakeups the OS makes possible.
- The Cosmopolitan runtime now uses less memory. On NetBSD for example,
it turned out the kernel would actually commit the PT_GNU_STACK size
which caused RSS to be 6mb for every process. Now it's down to ~4kb.
On Apple Silicon, we reduce the mandatory upstream thread size to the
smallest possible size to reduce the memory overhead of Cosmo threads.
The examples directory has a program called greenbean which can spawn
a web server on Linux with 10,000 worker threads and have the memory
usage of the process be ~77mb. The 1024 byte overhead of POSIX-style
thread-local storage is now optional; it won't be allocated until the
pthread_setspecific/getspecific functions are called. On Windows, the
threads that get spawned which are internal to the libc implementation
use reserve rather than commit memory, which shaves a few hundred kb.
- sigaltstack() is now supported on Windows, however it's currently not
able to be used to handle stack overflows, since crash signals are
still generated by WIN32. However the crash handler will still switch
to the alt stack, which is helpful in environments with tiny threads.
- Test binaries are now smaller. Many of the mandatory dependencies of
the test runner have been removed. This ensures many programs can do a
better job only linking the the thing they're testing. This caused the
test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb
- long double is no longer used in the implementation details of libc,
except in the APIs that define it. The old code that used long double
for time (instead of struct timespec) has now been thoroughly removed.
- ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing
backtraces itself, it'll just print a command you can run on the shell
using our new `cosmoaddr2line` program to view the backtrace.
- Crash report signal handling now works in a much better way. Instead
of terminating the process, it now relies on SA_RESETHAND so that the
default SIG_IGN behavior can terminate the process if necessary.
- Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
|
|
|
|
snprintf(fifopath_, sizeof(fifopath_), "%s%s.%d.%d.wav", __get_tmpdir(),
|
2022-05-13 20:31:21 +00:00
|
|
|
|
firstnonnull(program_invocation_short_name, "unknown"), getpid(),
|
|
|
|
|
|
count);
|
2020-06-15 14:18:57 +00:00
|
|
|
|
for (i = 0; i < ARRAYLEN(tryspeakerfns_); ++i) {
|
|
|
|
|
|
if (tryspeakerfns_[i]) {
|
|
|
|
|
|
if (++speakerfails_ <= 2 && tryspeakerfns_[i]()) {
|
|
|
|
|
|
return true;
|
|
|
|
|
|
} else {
|
|
|
|
|
|
speakerfails_ = 0;
|
|
|
|
|
|
tryspeakerfns_[i] = NULL;
|
|
|
|
|
|
}
|
|
|
|
|
|
}
|
|
|
|
|
|
}
|
|
|
|
|
|
return false;
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static void OnAudio(plm_t *mpeg, plm_samples_t *samples, void *user) {
|
|
|
|
|
|
if (playfd_ != -1) {
|
Make improvements
- Every unit test now passes on Apple Silicon. The final piece of this
puzzle was porting our POSIX threads cancelation support, since that
works differently on ARM64 XNU vs. AMD64. Our semaphore support on
Apple Silicon is also superior now compared to AMD64, thanks to the
grand central dispatch library which lets *NSYNC locks go faster.
- The Cosmopolitan runtime is now more stable, particularly on Windows.
To do this, thread local storage is mandatory at all runtime levels,
and the innermost packages of the C library is no longer being built
using ASAN. TLS is being bootstrapped with a 128-byte TIB during the
process startup phase, and then later on the runtime re-allocates it
either statically or dynamically to support code using _Thread_local.
fork() and execve() now do a better job cooperating with threads. We
can now check how much stack memory is left in the process or thread
when functions like kprintf() / execve() etc. call alloca(), so that
ENOMEM can be raised, reduce a buffer size, or just print a warning.
- POSIX signal emulation is now implemented the same way kernels do it
with pthread_kill() and raise(). Any thread can interrupt any other
thread, regardless of what it's doing. If it's blocked on read/write
then the killer thread will cancel its i/o operation so that EINTR can
be returned in the mark thread immediately. If it's doing a tight CPU
bound operation, then that's also interrupted by the signal delivery.
Signal delivery works now by suspending a thread and pushing context
data structures onto its stack, and redirecting its execution to a
trampoline function, which calls SetThreadContext(GetCurrentThread())
when it's done.
- We're now doing a better job managing locks and handles. On NetBSD we
now close semaphore file descriptors in forked children. Semaphores on
Windows can now be canceled immediately, which means mutexes/condition
variables will now go faster. Apple Silicon semaphores can be canceled
too. We're now using Apple's pthread_yield() funciton. Apple _nocancel
syscalls are now used on XNU when appropriate to ensure pthread_cancel
requests aren't lost. The MbedTLS library has been updated to support
POSIX thread cancelations. See tool/build/runitd.c for an example of
how it can be used for production multi-threaded tls servers. Handles
on Windows now leak less often across processes. All i/o operations on
Windows are now overlapped, which means file pointers can no longer be
inherited across dup() and fork() for the time being.
- We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4()
which means, for example, that posix_spawn() now goes 3x faster. POSIX
spawn is also now more correct. Like Musl, it's now able to report the
failure code of execve() via a pipe although our approach favors using
shared memory to do that on systems that have a true vfork() function.
- We now spawn a thread to deliver SIGALRM to threads when setitimer()
is used. This enables the most precise wakeups the OS makes possible.
- The Cosmopolitan runtime now uses less memory. On NetBSD for example,
it turned out the kernel would actually commit the PT_GNU_STACK size
which caused RSS to be 6mb for every process. Now it's down to ~4kb.
On Apple Silicon, we reduce the mandatory upstream thread size to the
smallest possible size to reduce the memory overhead of Cosmo threads.
The examples directory has a program called greenbean which can spawn
a web server on Linux with 10,000 worker threads and have the memory
usage of the process be ~77mb. The 1024 byte overhead of POSIX-style
thread-local storage is now optional; it won't be allocated until the
pthread_setspecific/getspecific functions are called. On Windows, the
threads that get spawned which are internal to the libc implementation
use reserve rather than commit memory, which shaves a few hundred kb.
- sigaltstack() is now supported on Windows, however it's currently not
able to be used to handle stack overflows, since crash signals are
still generated by WIN32. However the crash handler will still switch
to the alt stack, which is helpful in environments with tiny threads.
- Test binaries are now smaller. Many of the mandatory dependencies of
the test runner have been removed. This ensures many programs can do a
better job only linking the the thing they're testing. This caused the
test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb
- long double is no longer used in the implementation details of libc,
except in the APIs that define it. The old code that used long double
for time (instead of struct timespec) has now been thoroughly removed.
- ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing
backtraces itself, it'll just print a command you can run on the shell
using our new `cosmoaddr2line` program to view the backtrace.
- Crash report signal handling now works in a much better way. Instead
of terminating the process, it now relies on SA_RESETHAND so that the
default SIG_IGN behavior can terminate the process if necessary.
- Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
|
|
|
|
DEBUGF("OnAudio() [grace=%,ldns]", timespec_tonanos(GetGraceTime()));
|
2020-06-15 14:18:57 +00:00
|
|
|
|
CHECK_EQ(2, chans_);
|
|
|
|
|
|
CHECK_EQ(ARRAYLEN(pcm_) * 8, samples->count * chans_);
|
|
|
|
|
|
float2short(ARRAYLEN(pcm_), pcm_, (void *)samples->interleaved);
|
|
|
|
|
|
scalevolume(ARRAYLEN(pcm_), pcm_, volscale_);
|
|
|
|
|
|
sad16x8n(ARRAYLEN(pcm_), pcm_, pcmscale_);
|
|
|
|
|
|
DEBUGF("transcoded audio");
|
|
|
|
|
|
TryAgain:
|
|
|
|
|
|
if (WriteAudio(playfd_, pcm_, sizeof(pcm_), 1000) != -1) {
|
|
|
|
|
|
DEBUGF("WriteAudio(%d, %zu) ok [grace=%,ldns]", playfd_,
|
Make improvements
- Every unit test now passes on Apple Silicon. The final piece of this
puzzle was porting our POSIX threads cancelation support, since that
works differently on ARM64 XNU vs. AMD64. Our semaphore support on
Apple Silicon is also superior now compared to AMD64, thanks to the
grand central dispatch library which lets *NSYNC locks go faster.
- The Cosmopolitan runtime is now more stable, particularly on Windows.
To do this, thread local storage is mandatory at all runtime levels,
and the innermost packages of the C library is no longer being built
using ASAN. TLS is being bootstrapped with a 128-byte TIB during the
process startup phase, and then later on the runtime re-allocates it
either statically or dynamically to support code using _Thread_local.
fork() and execve() now do a better job cooperating with threads. We
can now check how much stack memory is left in the process or thread
when functions like kprintf() / execve() etc. call alloca(), so that
ENOMEM can be raised, reduce a buffer size, or just print a warning.
- POSIX signal emulation is now implemented the same way kernels do it
with pthread_kill() and raise(). Any thread can interrupt any other
thread, regardless of what it's doing. If it's blocked on read/write
then the killer thread will cancel its i/o operation so that EINTR can
be returned in the mark thread immediately. If it's doing a tight CPU
bound operation, then that's also interrupted by the signal delivery.
Signal delivery works now by suspending a thread and pushing context
data structures onto its stack, and redirecting its execution to a
trampoline function, which calls SetThreadContext(GetCurrentThread())
when it's done.
- We're now doing a better job managing locks and handles. On NetBSD we
now close semaphore file descriptors in forked children. Semaphores on
Windows can now be canceled immediately, which means mutexes/condition
variables will now go faster. Apple Silicon semaphores can be canceled
too. We're now using Apple's pthread_yield() funciton. Apple _nocancel
syscalls are now used on XNU when appropriate to ensure pthread_cancel
requests aren't lost. The MbedTLS library has been updated to support
POSIX thread cancelations. See tool/build/runitd.c for an example of
how it can be used for production multi-threaded tls servers. Handles
on Windows now leak less often across processes. All i/o operations on
Windows are now overlapped, which means file pointers can no longer be
inherited across dup() and fork() for the time being.
- We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4()
which means, for example, that posix_spawn() now goes 3x faster. POSIX
spawn is also now more correct. Like Musl, it's now able to report the
failure code of execve() via a pipe although our approach favors using
shared memory to do that on systems that have a true vfork() function.
- We now spawn a thread to deliver SIGALRM to threads when setitimer()
is used. This enables the most precise wakeups the OS makes possible.
- The Cosmopolitan runtime now uses less memory. On NetBSD for example,
it turned out the kernel would actually commit the PT_GNU_STACK size
which caused RSS to be 6mb for every process. Now it's down to ~4kb.
On Apple Silicon, we reduce the mandatory upstream thread size to the
smallest possible size to reduce the memory overhead of Cosmo threads.
The examples directory has a program called greenbean which can spawn
a web server on Linux with 10,000 worker threads and have the memory
usage of the process be ~77mb. The 1024 byte overhead of POSIX-style
thread-local storage is now optional; it won't be allocated until the
pthread_setspecific/getspecific functions are called. On Windows, the
threads that get spawned which are internal to the libc implementation
use reserve rather than commit memory, which shaves a few hundred kb.
- sigaltstack() is now supported on Windows, however it's currently not
able to be used to handle stack overflows, since crash signals are
still generated by WIN32. However the crash handler will still switch
to the alt stack, which is helpful in environments with tiny threads.
- Test binaries are now smaller. Many of the mandatory dependencies of
the test runner have been removed. This ensures many programs can do a
better job only linking the the thing they're testing. This caused the
test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb
- long double is no longer used in the implementation details of libc,
except in the APIs that define it. The old code that used long double
for time (instead of struct timespec) has now been thoroughly removed.
- ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing
backtraces itself, it'll just print a command you can run on the shell
using our new `cosmoaddr2line` program to view the backtrace.
- Crash report signal handling now works in a much better way. Instead
of terminating the process, it now relies on SA_RESETHAND so that the
default SIG_IGN behavior can terminate the process if necessary.
- Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
|
|
|
|
samples->count * 2, timespec_tonanos(GetGraceTime()));
|
2020-06-15 14:18:57 +00:00
|
|
|
|
} else {
|
|
|
|
|
|
WARNF("WriteAudio(%d, %zu) failed: %s", playfd_, samples->count * 2,
|
|
|
|
|
|
strerror(errno));
|
|
|
|
|
|
CloseSpeaker();
|
|
|
|
|
|
if (OpenSpeaker()) {
|
|
|
|
|
|
goto TryAgain;
|
|
|
|
|
|
}
|
|
|
|
|
|
}
|
|
|
|
|
|
}
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static void DescribeAlgorithms(char *p) {
|
|
|
|
|
|
if (dither_ && TTYQUANT()->alg != kTtyQuantTrue) {
|
|
|
|
|
|
p = stpcpy(p, " ℍithered");
|
|
|
|
|
|
}
|
|
|
|
|
|
if (yonly_) {
|
|
|
|
|
|
p = stpcpy(p, " grayscaled");
|
|
|
|
|
|
}
|
|
|
|
|
|
p += sprintf(p, " magikarp:%d:%d", lumakernel_, chromakernel_);
|
|
|
|
|
|
switch (TTYQUANT()->alg) {
|
|
|
|
|
|
case kTtyQuantTrue:
|
|
|
|
|
|
p = stpcpy(p, " true-color");
|
|
|
|
|
|
break;
|
|
|
|
|
|
case kTtyQuantXterm256:
|
|
|
|
|
|
p = stpcpy(p, " xterm256");
|
|
|
|
|
|
break;
|
|
|
|
|
|
case kTtyQuantAnsi:
|
|
|
|
|
|
p = stpcpy(p, " aixterm ansi");
|
Apply clang-format update to repo (#1154)
Commit bc6c183 introduced a bunch of discrepancies between what files
look like in the repo and what clang-format says they should look like.
However, there were already a few discrepancies prior to that. Most of
these discrepancies seemed to be unintentional, but a few of them were
load-bearing (e.g., a #include that violated header ordering needing
something to have been #defined by a 'later' #include.)
I opted to take what I hope is a relatively smooth-brained approach: I
reverted the .clang-format change, ran clang-format on the whole repo,
reapplied the .clang-format change, reran clang-format again, and then
reverted the commit that contained the first run. Thus the full effect
of this PR should only be to apply the changed formatting rules to the
repo, and from skimming the results, this seems to be the case.
My work can be checked by applying the short, manual commits, and then
rerunning the command listed in the autogenerated commits (those whose
messages I have prefixed auto:) and seeing if your results agree.
It might be that the other diffs should be fixed at some point but I'm
leaving that aside for now.
fd '\.c(c|pp)?$' --print0| xargs -0 clang-format -i
2024-04-25 17:38:00 +00:00
|
|
|
|
if (istango_)
|
|
|
|
|
|
p = stpcpy(p, " tango");
|
2020-06-15 14:18:57 +00:00
|
|
|
|
break;
|
|
|
|
|
|
default:
|
|
|
|
|
|
break;
|
|
|
|
|
|
}
|
|
|
|
|
|
switch (TTYQUANT()->blocks) {
|
|
|
|
|
|
case kTtyBlocksCp437:
|
|
|
|
|
|
p = stpcpy(p, " ibm cp437");
|
|
|
|
|
|
break;
|
|
|
|
|
|
case kTtyBlocksUnicode:
|
|
|
|
|
|
p = stpcpy(p, " unicode");
|
|
|
|
|
|
break;
|
|
|
|
|
|
default:
|
|
|
|
|
|
break;
|
|
|
|
|
|
}
|
|
|
|
|
|
*p++ = ' ';
|
|
|
|
|
|
*p = '\0';
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static char *StartRender(char *vt) {
|
Apply clang-format update to repo (#1154)
Commit bc6c183 introduced a bunch of discrepancies between what files
look like in the repo and what clang-format says they should look like.
However, there were already a few discrepancies prior to that. Most of
these discrepancies seemed to be unintentional, but a few of them were
load-bearing (e.g., a #include that violated header ordering needing
something to have been #defined by a 'later' #include.)
I opted to take what I hope is a relatively smooth-brained approach: I
reverted the .clang-format change, ran clang-format on the whole repo,
reapplied the .clang-format change, reran clang-format again, and then
reverted the commit that contained the first run. Thus the full effect
of this PR should only be to apply the changed formatting rules to the
repo, and from skimming the results, this seems to be the case.
My work can be checked by applying the short, manual commits, and then
rerunning the command listed in the autogenerated commits (those whose
messages I have prefixed auto:) and seeing if your results agree.
It might be that the other diffs should be fixed at some point but I'm
leaving that aside for now.
fd '\.c(c|pp)?$' --print0| xargs -0 clang-format -i
2024-04-25 17:38:00 +00:00
|
|
|
|
if (!ttymode_)
|
|
|
|
|
|
vt += sprintf(vt, "\r\n\r\n");
|
2020-06-15 14:18:57 +00:00
|
|
|
|
if (fullclear_) {
|
|
|
|
|
|
vt += sprintf(vt, "\e[0m\e[H\e[J");
|
|
|
|
|
|
fullclear_ = false;
|
|
|
|
|
|
} else if (historyclear_) {
|
|
|
|
|
|
vt += sprintf(vt, "\e[0m\e[H\e[J\e[3J");
|
|
|
|
|
|
historyclear_ = false;
|
|
|
|
|
|
}
|
|
|
|
|
|
vt += sprintf(vt, "\e[%hhuH", homerow_ + 1);
|
|
|
|
|
|
return vt;
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static void EndRender(char *vt) {
|
|
|
|
|
|
vt += sprintf(vt, "\e[0m");
|
2023-07-06 13:57:28 +00:00
|
|
|
|
f2_->n = (intptr_t)vt - (intptr_t)f2_->b;
|
2020-06-15 14:18:57 +00:00
|
|
|
|
f2_->i = 0;
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static bool IsNonZeroFloat(float f) {
|
|
|
|
|
|
return fabsf(f) > 0.001f;
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static bool HasAdjustments(void) {
|
|
|
|
|
|
return (IsNonZeroFloat(hue_) || IsNonZeroFloat(sat_) ||
|
|
|
|
|
|
IsNonZeroFloat(lit_)) ||
|
|
|
|
|
|
(emboss_ || sharp_ || blur_ || sobel_ || pf1_ || pf2_ || pf3_ ||
|
|
|
|
|
|
pf4_ || pf5_ || pf6_ || pf7_ || pf8_ || pf9_ || pf10_ || pf11_ ||
|
|
|
|
|
|
pf12_);
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static char *DescribeAdjustments(char *p) {
|
Apply clang-format update to repo (#1154)
Commit bc6c183 introduced a bunch of discrepancies between what files
look like in the repo and what clang-format says they should look like.
However, there were already a few discrepancies prior to that. Most of
these discrepancies seemed to be unintentional, but a few of them were
load-bearing (e.g., a #include that violated header ordering needing
something to have been #defined by a 'later' #include.)
I opted to take what I hope is a relatively smooth-brained approach: I
reverted the .clang-format change, ran clang-format on the whole repo,
reapplied the .clang-format change, reran clang-format again, and then
reverted the commit that contained the first run. Thus the full effect
of this PR should only be to apply the changed formatting rules to the
repo, and from skimming the results, this seems to be the case.
My work can be checked by applying the short, manual commits, and then
rerunning the command listed in the autogenerated commits (those whose
messages I have prefixed auto:) and seeing if your results agree.
It might be that the other diffs should be fixed at some point but I'm
leaving that aside for now.
fd '\.c(c|pp)?$' --print0| xargs -0 clang-format -i
2024-04-25 17:38:00 +00:00
|
|
|
|
if (emboss_)
|
|
|
|
|
|
p = stpcpy(p, " emboss");
|
|
|
|
|
|
if (sobel_)
|
|
|
|
|
|
p = stpcpy(p, " sobel");
|
2020-06-15 14:18:57 +00:00
|
|
|
|
switch (sharp_) {
|
|
|
|
|
|
case kSharpSharp:
|
|
|
|
|
|
p = stpcpy(p, " sharp");
|
|
|
|
|
|
break;
|
|
|
|
|
|
case kSharpUnsharp:
|
|
|
|
|
|
p = stpcpy(p, " unsharp");
|
|
|
|
|
|
break;
|
|
|
|
|
|
default:
|
|
|
|
|
|
break;
|
|
|
|
|
|
}
|
|
|
|
|
|
switch (blur_) {
|
|
|
|
|
|
case kBlurBox:
|
|
|
|
|
|
p = stpcpy(p, " boxblur");
|
|
|
|
|
|
break;
|
|
|
|
|
|
case kBlurGaussian:
|
|
|
|
|
|
p = stpcpy(p, " gaussian");
|
|
|
|
|
|
break;
|
|
|
|
|
|
default:
|
|
|
|
|
|
break;
|
|
|
|
|
|
}
|
Apply clang-format update to repo (#1154)
Commit bc6c183 introduced a bunch of discrepancies between what files
look like in the repo and what clang-format says they should look like.
However, there were already a few discrepancies prior to that. Most of
these discrepancies seemed to be unintentional, but a few of them were
load-bearing (e.g., a #include that violated header ordering needing
something to have been #defined by a 'later' #include.)
I opted to take what I hope is a relatively smooth-brained approach: I
reverted the .clang-format change, ran clang-format on the whole repo,
reapplied the .clang-format change, reran clang-format again, and then
reverted the commit that contained the first run. Thus the full effect
of this PR should only be to apply the changed formatting rules to the
repo, and from skimming the results, this seems to be the case.
My work can be checked by applying the short, manual commits, and then
rerunning the command listed in the autogenerated commits (those whose
messages I have prefixed auto:) and seeing if your results agree.
It might be that the other diffs should be fixed at some point but I'm
leaving that aside for now.
fd '\.c(c|pp)?$' --print0| xargs -0 clang-format -i
2024-04-25 17:38:00 +00:00
|
|
|
|
if (IsNonZeroFloat(hue_))
|
|
|
|
|
|
p += sprintf(p, " hue%+.2f", hue_);
|
|
|
|
|
|
if (IsNonZeroFloat(sat_))
|
|
|
|
|
|
p += sprintf(p, " sat%+.2f", sat_);
|
|
|
|
|
|
if (IsNonZeroFloat(lit_))
|
|
|
|
|
|
p += sprintf(p, " lit%+.2f", lit_);
|
|
|
|
|
|
if (pf1_)
|
|
|
|
|
|
p = stpcpy(p, " PF1");
|
|
|
|
|
|
if (pf2_)
|
|
|
|
|
|
p = stpcpy(p, " PF2");
|
|
|
|
|
|
if (pf3_)
|
|
|
|
|
|
p = stpcpy(p, " PF3");
|
|
|
|
|
|
if (pf4_)
|
|
|
|
|
|
p = stpcpy(p, " PF4");
|
|
|
|
|
|
if (pf5_)
|
|
|
|
|
|
p = stpcpy(p, " PF5");
|
|
|
|
|
|
if (pf6_)
|
|
|
|
|
|
p = stpcpy(p, " PF6");
|
|
|
|
|
|
if (pf7_)
|
|
|
|
|
|
p = stpcpy(p, " PF7");
|
|
|
|
|
|
if (pf8_)
|
|
|
|
|
|
p = stpcpy(p, " PF8");
|
|
|
|
|
|
if (pf9_)
|
|
|
|
|
|
p = stpcpy(p, " PF9");
|
|
|
|
|
|
if (pf10_)
|
|
|
|
|
|
p = stpcpy(p, " PF10");
|
|
|
|
|
|
if (pf11_)
|
|
|
|
|
|
p = stpcpy(p, " PF11");
|
|
|
|
|
|
if (pf12_)
|
|
|
|
|
|
p = stpcpy(p, " PF12");
|
2020-06-15 14:18:57 +00:00
|
|
|
|
*p++ = ' ';
|
|
|
|
|
|
*p++ = '\0';
|
|
|
|
|
|
return p;
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static const char *DescribeSwing(int swing) {
|
|
|
|
|
|
switch (swing) {
|
|
|
|
|
|
case 219:
|
|
|
|
|
|
return "TV";
|
|
|
|
|
|
case 255:
|
|
|
|
|
|
return "PC";
|
|
|
|
|
|
default:
|
|
|
|
|
|
return "??";
|
|
|
|
|
|
}
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static void RenderIt(void) {
|
|
|
|
|
|
long bpf;
|
|
|
|
|
|
double bpc;
|
|
|
|
|
|
char *vt, *p;
|
|
|
|
|
|
unsigned yn, xn;
|
|
|
|
|
|
struct TtyRgb bg, fg;
|
|
|
|
|
|
yn = g2_->yn;
|
|
|
|
|
|
xn = g2_->xn;
|
2023-07-06 13:57:28 +00:00
|
|
|
|
vt = f2_->b;
|
2020-06-15 14:18:57 +00:00
|
|
|
|
p = StartRender(vt);
|
|
|
|
|
|
if (TTYQUANT()->alg == kTtyQuantTrue) {
|
|
|
|
|
|
bg = (struct TtyRgb){0, 0, 0, 0};
|
|
|
|
|
|
fg = (struct TtyRgb){0xee, 0xff, 0xff, 0};
|
|
|
|
|
|
p = stpcpy(p, "\e[48;2;0;0;0;38;2;255;255;255m");
|
|
|
|
|
|
} else if (TTYQUANT()->alg == kTtyQuantAnsi) {
|
|
|
|
|
|
bg = g_ansi2rgb_[0];
|
|
|
|
|
|
fg = g_ansi2rgb_[7];
|
|
|
|
|
|
p +=
|
|
|
|
|
|
sprintf(p, "\e[%d;%dm", 30 + g_ansi2rgb_[0].xt, 40 + g_ansi2rgb_[7].xt);
|
|
|
|
|
|
} else {
|
|
|
|
|
|
bg = (struct TtyRgb){0, 0, 0, 16};
|
|
|
|
|
|
fg = (struct TtyRgb){0xff, 0xff, 0xff, 231};
|
|
|
|
|
|
p = stpcpy(p, "\e[48;5;16;38;5;231m");
|
|
|
|
|
|
}
|
2023-07-06 13:57:28 +00:00
|
|
|
|
p = ttyraster(p, xtcodes_, yn, xn, bg, fg);
|
2020-06-15 14:18:57 +00:00
|
|
|
|
if (ttymode_ && stats_) {
|
|
|
|
|
|
bpc = bpf = p - vt;
|
|
|
|
|
|
bpc /= wsize_.ws_row * wsize_.ws_col;
|
2020-08-25 11:23:25 +00:00
|
|
|
|
sprintf(status_[4], " %s/%s/%s %d×%d → %u×%u pixels ",
|
2020-06-15 14:18:57 +00:00
|
|
|
|
kPrimaries[primaries_].name, DescribeSwing(swing_),
|
|
|
|
|
|
kLightings[lighting_].name, plm_get_width(plm_),
|
|
|
|
|
|
plm_get_height(plm_), g2_->xn, g2_->yn);
|
|
|
|
|
|
sprintf(status_[5], " decode:%,8luµs | magikarp:%,8luµs ",
|
|
|
|
|
|
plmpegdecode_latency_, magikarp_latency_);
|
|
|
|
|
|
sprintf(status_[1], " ycbcr2rgb:%,8luµs | gyarados:%,8luµs ",
|
|
|
|
|
|
ycbcr2rgb_latency_, gyarados_latency_);
|
|
|
|
|
|
sprintf(status_[0], " fx:%,ldµs %.6fbpc %,ldbpf %.6ffps ",
|
|
|
|
|
|
lroundl(t6 / 1e3L), bpc, bpf, (size_t)(p - vt), MeasureFrameRate());
|
|
|
|
|
|
sprintf(status_[2], " gamma:%.1f %hu columns × %hu lines of text ", gamma_,
|
|
|
|
|
|
wsize_.ws_col, wsize_.ws_row);
|
|
|
|
|
|
DescribeAlgorithms(status_[3]);
|
|
|
|
|
|
p += sprintf(p, "\e[0m");
|
|
|
|
|
|
if (HasAdjustments()) {
|
|
|
|
|
|
DescribeAdjustments(status_[6]);
|
|
|
|
|
|
p += sprintf(p, "\e[%d;%dH%s", lastrow_ - 7,
|
2021-02-05 14:16:20 +00:00
|
|
|
|
HALF(xn) - strwidth(status_[6], 0), status_[6]);
|
2020-06-15 14:18:57 +00:00
|
|
|
|
}
|
|
|
|
|
|
p += sprintf(p, "\e[%d;%dH%s", lastrow_ - 6,
|
2021-02-05 14:16:20 +00:00
|
|
|
|
HALF(xn) - strwidth(status_[4], 0), status_[4]);
|
2020-06-15 14:18:57 +00:00
|
|
|
|
p += sprintf(p, "\e[%d;%dH%s", lastrow_ - 5,
|
2021-02-05 14:16:20 +00:00
|
|
|
|
HALF(xn) - strwidth(status_[5], 0), status_[5]);
|
2020-06-15 14:18:57 +00:00
|
|
|
|
p += sprintf(p, "\e[%d;%dH%s", lastrow_ - 4,
|
2021-02-05 14:16:20 +00:00
|
|
|
|
HALF(xn) - strwidth(status_[1], 0), status_[1]);
|
2020-06-15 14:18:57 +00:00
|
|
|
|
p += sprintf(p, "\e[%d;%dH%s", lastrow_ - 3,
|
2021-02-05 14:16:20 +00:00
|
|
|
|
HALF(xn) - strwidth(status_[0], 0), status_[0]);
|
2020-06-15 14:18:57 +00:00
|
|
|
|
p += sprintf(p, "\e[%d;%dH%30s", lastrow_ - 2,
|
2021-02-05 14:16:20 +00:00
|
|
|
|
HALF(xn) - strwidth(status_[2], 0), status_[2]);
|
2020-06-15 14:18:57 +00:00
|
|
|
|
p += sprintf(p, "\e[%d;%dH%s", lastrow_ - 1,
|
2021-02-05 14:16:20 +00:00
|
|
|
|
HALF(xn) - strwidth(status_[3], 0), status_[3]);
|
2020-06-15 14:18:57 +00:00
|
|
|
|
p += sprintf(p, "\e[%d;%dH %s %s ", lastrow_ - 2, 2,
|
|
|
|
|
|
program_invocation_name, "");
|
|
|
|
|
|
p += sprintf(p, "\e[%d;%dH %s ", lastrow_ - 1, 2,
|
|
|
|
|
|
"by justine tunney <jtunney@gmail.com>");
|
|
|
|
|
|
}
|
|
|
|
|
|
EndRender(p);
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static void RasterIt(void) {
|
|
|
|
|
|
static bool once;
|
|
|
|
|
|
static void *buf;
|
|
|
|
|
|
if (!once) {
|
2024-06-21 03:46:42 +00:00
|
|
|
|
buf = _mapanon(ROUNDUP(fb0_.size, __granularity()));
|
2020-06-15 14:18:57 +00:00
|
|
|
|
once = true;
|
|
|
|
|
|
}
|
|
|
|
|
|
WriteToFrameBuffer(fb0_.vscreen.yres_virtual, fb0_.vscreen.xres_virtual, buf,
|
2023-07-06 13:57:28 +00:00
|
|
|
|
g2_->yn, g2_->xn, g2_->b, fb0_.vscreen.yres,
|
2020-06-15 14:18:57 +00:00
|
|
|
|
fb0_.vscreen.xres);
|
|
|
|
|
|
memcpy(fb0_.map, buf, fb0_.size);
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static void TranscodeVideo(plm_frame_t *pf) {
|
|
|
|
|
|
CHECK_EQ(pf->cb.width, pf->cr.width);
|
|
|
|
|
|
CHECK_EQ(pf->cb.height, pf->cr.height);
|
Make improvements
- Every unit test now passes on Apple Silicon. The final piece of this
puzzle was porting our POSIX threads cancelation support, since that
works differently on ARM64 XNU vs. AMD64. Our semaphore support on
Apple Silicon is also superior now compared to AMD64, thanks to the
grand central dispatch library which lets *NSYNC locks go faster.
- The Cosmopolitan runtime is now more stable, particularly on Windows.
To do this, thread local storage is mandatory at all runtime levels,
and the innermost packages of the C library is no longer being built
using ASAN. TLS is being bootstrapped with a 128-byte TIB during the
process startup phase, and then later on the runtime re-allocates it
either statically or dynamically to support code using _Thread_local.
fork() and execve() now do a better job cooperating with threads. We
can now check how much stack memory is left in the process or thread
when functions like kprintf() / execve() etc. call alloca(), so that
ENOMEM can be raised, reduce a buffer size, or just print a warning.
- POSIX signal emulation is now implemented the same way kernels do it
with pthread_kill() and raise(). Any thread can interrupt any other
thread, regardless of what it's doing. If it's blocked on read/write
then the killer thread will cancel its i/o operation so that EINTR can
be returned in the mark thread immediately. If it's doing a tight CPU
bound operation, then that's also interrupted by the signal delivery.
Signal delivery works now by suspending a thread and pushing context
data structures onto its stack, and redirecting its execution to a
trampoline function, which calls SetThreadContext(GetCurrentThread())
when it's done.
- We're now doing a better job managing locks and handles. On NetBSD we
now close semaphore file descriptors in forked children. Semaphores on
Windows can now be canceled immediately, which means mutexes/condition
variables will now go faster. Apple Silicon semaphores can be canceled
too. We're now using Apple's pthread_yield() funciton. Apple _nocancel
syscalls are now used on XNU when appropriate to ensure pthread_cancel
requests aren't lost. The MbedTLS library has been updated to support
POSIX thread cancelations. See tool/build/runitd.c for an example of
how it can be used for production multi-threaded tls servers. Handles
on Windows now leak less often across processes. All i/o operations on
Windows are now overlapped, which means file pointers can no longer be
inherited across dup() and fork() for the time being.
- We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4()
which means, for example, that posix_spawn() now goes 3x faster. POSIX
spawn is also now more correct. Like Musl, it's now able to report the
failure code of execve() via a pipe although our approach favors using
shared memory to do that on systems that have a true vfork() function.
- We now spawn a thread to deliver SIGALRM to threads when setitimer()
is used. This enables the most precise wakeups the OS makes possible.
- The Cosmopolitan runtime now uses less memory. On NetBSD for example,
it turned out the kernel would actually commit the PT_GNU_STACK size
which caused RSS to be 6mb for every process. Now it's down to ~4kb.
On Apple Silicon, we reduce the mandatory upstream thread size to the
smallest possible size to reduce the memory overhead of Cosmo threads.
The examples directory has a program called greenbean which can spawn
a web server on Linux with 10,000 worker threads and have the memory
usage of the process be ~77mb. The 1024 byte overhead of POSIX-style
thread-local storage is now optional; it won't be allocated until the
pthread_setspecific/getspecific functions are called. On Windows, the
threads that get spawned which are internal to the libc implementation
use reserve rather than commit memory, which shaves a few hundred kb.
- sigaltstack() is now supported on Windows, however it's currently not
able to be used to handle stack overflows, since crash signals are
still generated by WIN32. However the crash handler will still switch
to the alt stack, which is helpful in environments with tiny threads.
- Test binaries are now smaller. Many of the mandatory dependencies of
the test runner have been removed. This ensures many programs can do a
better job only linking the the thing they're testing. This caused the
test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb
- long double is no longer used in the implementation details of libc,
except in the APIs that define it. The old code that used long double
for time (instead of struct timespec) has now been thoroughly removed.
- ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing
backtraces itself, it'll just print a command you can run on the shell
using our new `cosmoaddr2line` program to view the backtrace.
- Crash report signal handling now works in a much better way. Instead
of terminating the process, it now relies on SA_RESETHAND so that the
default SIG_IGN behavior can terminate the process if necessary.
- Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
|
|
|
|
DEBUGF("TranscodeVideo() [grace=%,ldns]", timespec_tonanos(GetGraceTime()));
|
2020-06-15 14:18:57 +00:00
|
|
|
|
g2_ = &graphic_[1];
|
|
|
|
|
|
t5 = 0;
|
|
|
|
|
|
|
|
|
|
|
|
TIMEIT(t1, {
|
|
|
|
|
|
pary_ = 2;
|
Apply clang-format update to repo (#1154)
Commit bc6c183 introduced a bunch of discrepancies between what files
look like in the repo and what clang-format says they should look like.
However, there were already a few discrepancies prior to that. Most of
these discrepancies seemed to be unintentional, but a few of them were
load-bearing (e.g., a #include that violated header ordering needing
something to have been #defined by a 'later' #include.)
I opted to take what I hope is a relatively smooth-brained approach: I
reverted the .clang-format change, ran clang-format on the whole repo,
reapplied the .clang-format change, reran clang-format again, and then
reverted the commit that contained the first run. Thus the full effect
of this PR should only be to apply the changed formatting rules to the
repo, and from skimming the results, this seems to be the case.
My work can be checked by applying the short, manual commits, and then
rerunning the command listed in the autogenerated commits (those whose
messages I have prefixed auto:) and seeing if your results agree.
It might be that the other diffs should be fixed at some point but I'm
leaving that aside for now.
fd '\.c(c|pp)?$' --print0| xargs -0 clang-format -i
2024-04-25 17:38:00 +00:00
|
|
|
|
if (pf1_)
|
|
|
|
|
|
pary_ = 1.;
|
|
|
|
|
|
if (pf2_)
|
|
|
|
|
|
pary_ = (266 / 64.) * (900 / 1600.);
|
2020-06-15 14:18:57 +00:00
|
|
|
|
pary_ *= plm_get_pixel_aspect_ratio(plm_);
|
2023-07-06 13:57:28 +00:00
|
|
|
|
YCbCr2RgbScale(g2_->yn, g2_->xn, g2_->b, pf->y.height, pf->y.width,
|
2020-06-15 14:18:57 +00:00
|
|
|
|
(void *)pf->y.data, pf->cr.height, pf->cr.width,
|
|
|
|
|
|
(void *)pf->cb.data, (void *)pf->cr.data, pf->y.height,
|
|
|
|
|
|
pf->y.width, pf->cr.height, pf->cr.width, pf->height,
|
|
|
|
|
|
pf->width, pary_, parx_, &ycbcr_);
|
|
|
|
|
|
});
|
|
|
|
|
|
t2 = 0;
|
|
|
|
|
|
t8 = 0;
|
|
|
|
|
|
|
|
|
|
|
|
TIMEIT(t6, {
|
|
|
|
|
|
switch (blur_) {
|
|
|
|
|
|
case kBlurBox:
|
|
|
|
|
|
boxblur(g2_);
|
|
|
|
|
|
break;
|
|
|
|
|
|
case kBlurGaussian:
|
2023-07-06 13:57:28 +00:00
|
|
|
|
gaussian(g2_->yn, g2_->xn, g2_->b);
|
2020-06-15 14:18:57 +00:00
|
|
|
|
break;
|
|
|
|
|
|
default:
|
|
|
|
|
|
break;
|
|
|
|
|
|
}
|
Apply clang-format update to repo (#1154)
Commit bc6c183 introduced a bunch of discrepancies between what files
look like in the repo and what clang-format says they should look like.
However, there were already a few discrepancies prior to that. Most of
these discrepancies seemed to be unintentional, but a few of them were
load-bearing (e.g., a #include that violated header ordering needing
something to have been #defined by a 'later' #include.)
I opted to take what I hope is a relatively smooth-brained approach: I
reverted the .clang-format change, ran clang-format on the whole repo,
reapplied the .clang-format change, reran clang-format again, and then
reverted the commit that contained the first run. Thus the full effect
of this PR should only be to apply the changed formatting rules to the
repo, and from skimming the results, this seems to be the case.
My work can be checked by applying the short, manual commits, and then
rerunning the command listed in the autogenerated commits (those whose
messages I have prefixed auto:) and seeing if your results agree.
It might be that the other diffs should be fixed at some point but I'm
leaving that aside for now.
fd '\.c(c|pp)?$' --print0| xargs -0 clang-format -i
2024-04-25 17:38:00 +00:00
|
|
|
|
if (sobel_)
|
|
|
|
|
|
sobel(g2_);
|
|
|
|
|
|
if (emboss_)
|
|
|
|
|
|
emboss(g2_);
|
2020-06-15 14:18:57 +00:00
|
|
|
|
switch (sharp_) {
|
|
|
|
|
|
case kSharpSharp:
|
2023-07-06 13:57:28 +00:00
|
|
|
|
sharpen(3, g2_->yn, g2_->xn, g2_->b, g2_->yn, g2_->xn);
|
2020-06-15 14:18:57 +00:00
|
|
|
|
break;
|
|
|
|
|
|
case kSharpUnsharp:
|
2023-07-06 13:57:28 +00:00
|
|
|
|
unsharp(3, g2_->yn, g2_->xn, g2_->b, g2_->yn, g2_->xn);
|
2020-06-15 14:18:57 +00:00
|
|
|
|
break;
|
|
|
|
|
|
default:
|
|
|
|
|
|
break;
|
|
|
|
|
|
}
|
|
|
|
|
|
if (dither_ && TTYQUANT()->alg != kTtyQuantTrue) {
|
2023-07-06 13:57:28 +00:00
|
|
|
|
dither(g2_->yn, g2_->xn, g2_->b, g2_->yn, g2_->xn);
|
2020-06-15 14:18:57 +00:00
|
|
|
|
}
|
|
|
|
|
|
});
|
|
|
|
|
|
|
|
|
|
|
|
if (ShouldUseFrameBuffer()) {
|
|
|
|
|
|
t3 = 0;
|
|
|
|
|
|
TIMEIT(t4, RasterIt());
|
|
|
|
|
|
} else {
|
2023-07-06 13:57:28 +00:00
|
|
|
|
TIMEIT(t3, getxtermcodes(xtcodes_, g2_));
|
2020-06-15 14:18:57 +00:00
|
|
|
|
TIMEIT(t4, RenderIt());
|
|
|
|
|
|
}
|
|
|
|
|
|
|
2021-09-04 04:14:26 +00:00
|
|
|
|
INFOF("𝑓%zu(%u×%u) %,zub (%f BPP) "
|
|
|
|
|
|
"ycbcr=%,zuns "
|
|
|
|
|
|
"scale=%,zuns "
|
|
|
|
|
|
"lace=%,zuns "
|
|
|
|
|
|
"fx=%,zuns "
|
|
|
|
|
|
"quantize=%,zuns "
|
|
|
|
|
|
"render=%,zuns",
|
|
|
|
|
|
framecount_++, g2_->yn, g2_->xn, f2_->n,
|
|
|
|
|
|
(f1_->n / (double)(g2_->yn * g2_->xn)), t1, t2, t8, t6, t3, t4);
|
2020-06-15 14:18:57 +00:00
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static void OnVideo(plm_t *mpeg, plm_frame_t *pf, void *user) {
|
|
|
|
|
|
gotvideo_ = true;
|
|
|
|
|
|
if (f2_->n) {
|
|
|
|
|
|
WARNF("video frame dropped");
|
|
|
|
|
|
} else {
|
|
|
|
|
|
TranscodeVideo(pf);
|
|
|
|
|
|
if (!f1_->n) {
|
2023-07-30 01:44:15 +00:00
|
|
|
|
struct VtFrame *t = f1_;
|
|
|
|
|
|
f1_ = f2_, f2_ = t;
|
2020-06-15 14:18:57 +00:00
|
|
|
|
f1_start_ = decode_start_;
|
|
|
|
|
|
} else {
|
|
|
|
|
|
f2_start_ = decode_start_;
|
|
|
|
|
|
}
|
|
|
|
|
|
}
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static void OpenVideo(void) {
|
|
|
|
|
|
size_t yn, xn;
|
|
|
|
|
|
playfd_ = -1;
|
2021-09-04 04:14:26 +00:00
|
|
|
|
INFOF("%s(%`'s)", "OpenVideo", patharg_);
|
2020-12-25 00:11:38 +00:00
|
|
|
|
CHECK_NOTNULL((plm_ = plm_create_with_filename(patharg_)));
|
2020-06-15 14:18:57 +00:00
|
|
|
|
swing_ = 219;
|
|
|
|
|
|
xn = plm_get_width(plm_);
|
|
|
|
|
|
yn = plm_get_height(plm_);
|
|
|
|
|
|
lighting_ = GetLighting("D65");
|
|
|
|
|
|
primaries_ = IsHighDefinition(yn, xn) ? GetPrimaries("BT.709")
|
|
|
|
|
|
: GetPrimaries("BT.601");
|
|
|
|
|
|
plm_set_video_decode_callback(plm_, OnVideo, NULL);
|
|
|
|
|
|
plm_set_audio_decode_callback(plm_, OnAudio, NULL);
|
|
|
|
|
|
plm_set_loop(plm_, false);
|
2022-04-23 01:55:28 +00:00
|
|
|
|
FormatInt64(chansstr_, (chans_ = 2));
|
|
|
|
|
|
FormatInt64(sratestr_, (srate_ = plm_get_samplerate(plm_)));
|
2020-06-15 14:18:57 +00:00
|
|
|
|
if (plm_get_num_audio_streams(plm_) && OpenSpeaker()) {
|
|
|
|
|
|
plm_set_audio_enabled(plm_, true, 0);
|
|
|
|
|
|
} else {
|
|
|
|
|
|
plm_set_audio_enabled(plm_, false, 0);
|
|
|
|
|
|
}
|
|
|
|
|
|
g2_ = g1_ = resizegraphic(&graphic_[0], yn, xn);
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static ssize_t WriteVideoCall(void) {
|
|
|
|
|
|
size_t amt;
|
|
|
|
|
|
ssize_t rc;
|
2023-07-30 01:44:15 +00:00
|
|
|
|
amt = min(4096 * 4, f1_->n - f1_->i);
|
2020-06-15 14:18:57 +00:00
|
|
|
|
if ((rc = write(outfd_, f1_->bytes + f1_->i, amt)) != -1) {
|
|
|
|
|
|
if ((f1_->i += rc) == f1_->n) {
|
|
|
|
|
|
if (plm_get_audio_enabled(plm_)) {
|
|
|
|
|
|
plm_set_audio_lead_time(
|
|
|
|
|
|
plm_,
|
Make improvements
- Every unit test now passes on Apple Silicon. The final piece of this
puzzle was porting our POSIX threads cancelation support, since that
works differently on ARM64 XNU vs. AMD64. Our semaphore support on
Apple Silicon is also superior now compared to AMD64, thanks to the
grand central dispatch library which lets *NSYNC locks go faster.
- The Cosmopolitan runtime is now more stable, particularly on Windows.
To do this, thread local storage is mandatory at all runtime levels,
and the innermost packages of the C library is no longer being built
using ASAN. TLS is being bootstrapped with a 128-byte TIB during the
process startup phase, and then later on the runtime re-allocates it
either statically or dynamically to support code using _Thread_local.
fork() and execve() now do a better job cooperating with threads. We
can now check how much stack memory is left in the process or thread
when functions like kprintf() / execve() etc. call alloca(), so that
ENOMEM can be raised, reduce a buffer size, or just print a warning.
- POSIX signal emulation is now implemented the same way kernels do it
with pthread_kill() and raise(). Any thread can interrupt any other
thread, regardless of what it's doing. If it's blocked on read/write
then the killer thread will cancel its i/o operation so that EINTR can
be returned in the mark thread immediately. If it's doing a tight CPU
bound operation, then that's also interrupted by the signal delivery.
Signal delivery works now by suspending a thread and pushing context
data structures onto its stack, and redirecting its execution to a
trampoline function, which calls SetThreadContext(GetCurrentThread())
when it's done.
- We're now doing a better job managing locks and handles. On NetBSD we
now close semaphore file descriptors in forked children. Semaphores on
Windows can now be canceled immediately, which means mutexes/condition
variables will now go faster. Apple Silicon semaphores can be canceled
too. We're now using Apple's pthread_yield() funciton. Apple _nocancel
syscalls are now used on XNU when appropriate to ensure pthread_cancel
requests aren't lost. The MbedTLS library has been updated to support
POSIX thread cancelations. See tool/build/runitd.c for an example of
how it can be used for production multi-threaded tls servers. Handles
on Windows now leak less often across processes. All i/o operations on
Windows are now overlapped, which means file pointers can no longer be
inherited across dup() and fork() for the time being.
- We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4()
which means, for example, that posix_spawn() now goes 3x faster. POSIX
spawn is also now more correct. Like Musl, it's now able to report the
failure code of execve() via a pipe although our approach favors using
shared memory to do that on systems that have a true vfork() function.
- We now spawn a thread to deliver SIGALRM to threads when setitimer()
is used. This enables the most precise wakeups the OS makes possible.
- The Cosmopolitan runtime now uses less memory. On NetBSD for example,
it turned out the kernel would actually commit the PT_GNU_STACK size
which caused RSS to be 6mb for every process. Now it's down to ~4kb.
On Apple Silicon, we reduce the mandatory upstream thread size to the
smallest possible size to reduce the memory overhead of Cosmo threads.
The examples directory has a program called greenbean which can spawn
a web server on Linux with 10,000 worker threads and have the memory
usage of the process be ~77mb. The 1024 byte overhead of POSIX-style
thread-local storage is now optional; it won't be allocated until the
pthread_setspecific/getspecific functions are called. On Windows, the
threads that get spawned which are internal to the libc implementation
use reserve rather than commit memory, which shaves a few hundred kb.
- sigaltstack() is now supported on Windows, however it's currently not
able to be used to handle stack overflows, since crash signals are
still generated by WIN32. However the crash handler will still switch
to the alt stack, which is helpful in environments with tiny threads.
- Test binaries are now smaller. Many of the mandatory dependencies of
the test runner have been removed. This ensures many programs can do a
better job only linking the the thing they're testing. This caused the
test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb
- long double is no longer used in the implementation details of libc,
except in the APIs that define it. The old code that used long double
for time (instead of struct timespec) has now been thoroughly removed.
- ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing
backtraces itself, it'll just print a command you can run on the shell
using our new `cosmoaddr2line` program to view the backtrace.
- Crash report signal handling now works in a much better way. Instead
of terminating the process, it now relies on SA_RESETHAND so that the
default SIG_IGN behavior can terminate the process if necessary.
- Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
|
|
|
|
max(0,
|
|
|
|
|
|
min(timespec_tofloat(timespec_sub(timespec_real(), f1_start_)),
|
|
|
|
|
|
plm_get_samplerate(plm_) / PLM_AUDIO_SAMPLES_PER_FRAME)));
|
2020-06-15 14:18:57 +00:00
|
|
|
|
}
|
|
|
|
|
|
f1_start_ = f2_start_;
|
|
|
|
|
|
f1_->i = f1_->n = 0;
|
2023-07-30 01:44:15 +00:00
|
|
|
|
struct VtFrame *t = f1_;
|
|
|
|
|
|
f1_ = f2_, f2_ = t;
|
2020-06-15 14:18:57 +00:00
|
|
|
|
RecordFactThatFrameWasFullyRendered();
|
|
|
|
|
|
}
|
|
|
|
|
|
}
|
|
|
|
|
|
return rc;
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static void DrainVideo(void) {
|
|
|
|
|
|
if (f1_ && f1_->n) {
|
|
|
|
|
|
ttywrite(outfd_, f1_->bytes + f1_->i, f1_->n - f1_->i);
|
|
|
|
|
|
f1_->i = f1_->n = 0;
|
|
|
|
|
|
}
|
|
|
|
|
|
if (f2_ && f2_->n) {
|
|
|
|
|
|
f2_->i = f2_->n = 0;
|
|
|
|
|
|
}
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static void WriteVideo(void) {
|
|
|
|
|
|
ssize_t rc;
|
Make improvements
- Every unit test now passes on Apple Silicon. The final piece of this
puzzle was porting our POSIX threads cancelation support, since that
works differently on ARM64 XNU vs. AMD64. Our semaphore support on
Apple Silicon is also superior now compared to AMD64, thanks to the
grand central dispatch library which lets *NSYNC locks go faster.
- The Cosmopolitan runtime is now more stable, particularly on Windows.
To do this, thread local storage is mandatory at all runtime levels,
and the innermost packages of the C library is no longer being built
using ASAN. TLS is being bootstrapped with a 128-byte TIB during the
process startup phase, and then later on the runtime re-allocates it
either statically or dynamically to support code using _Thread_local.
fork() and execve() now do a better job cooperating with threads. We
can now check how much stack memory is left in the process or thread
when functions like kprintf() / execve() etc. call alloca(), so that
ENOMEM can be raised, reduce a buffer size, or just print a warning.
- POSIX signal emulation is now implemented the same way kernels do it
with pthread_kill() and raise(). Any thread can interrupt any other
thread, regardless of what it's doing. If it's blocked on read/write
then the killer thread will cancel its i/o operation so that EINTR can
be returned in the mark thread immediately. If it's doing a tight CPU
bound operation, then that's also interrupted by the signal delivery.
Signal delivery works now by suspending a thread and pushing context
data structures onto its stack, and redirecting its execution to a
trampoline function, which calls SetThreadContext(GetCurrentThread())
when it's done.
- We're now doing a better job managing locks and handles. On NetBSD we
now close semaphore file descriptors in forked children. Semaphores on
Windows can now be canceled immediately, which means mutexes/condition
variables will now go faster. Apple Silicon semaphores can be canceled
too. We're now using Apple's pthread_yield() funciton. Apple _nocancel
syscalls are now used on XNU when appropriate to ensure pthread_cancel
requests aren't lost. The MbedTLS library has been updated to support
POSIX thread cancelations. See tool/build/runitd.c for an example of
how it can be used for production multi-threaded tls servers. Handles
on Windows now leak less often across processes. All i/o operations on
Windows are now overlapped, which means file pointers can no longer be
inherited across dup() and fork() for the time being.
- We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4()
which means, for example, that posix_spawn() now goes 3x faster. POSIX
spawn is also now more correct. Like Musl, it's now able to report the
failure code of execve() via a pipe although our approach favors using
shared memory to do that on systems that have a true vfork() function.
- We now spawn a thread to deliver SIGALRM to threads when setitimer()
is used. This enables the most precise wakeups the OS makes possible.
- The Cosmopolitan runtime now uses less memory. On NetBSD for example,
it turned out the kernel would actually commit the PT_GNU_STACK size
which caused RSS to be 6mb for every process. Now it's down to ~4kb.
On Apple Silicon, we reduce the mandatory upstream thread size to the
smallest possible size to reduce the memory overhead of Cosmo threads.
The examples directory has a program called greenbean which can spawn
a web server on Linux with 10,000 worker threads and have the memory
usage of the process be ~77mb. The 1024 byte overhead of POSIX-style
thread-local storage is now optional; it won't be allocated until the
pthread_setspecific/getspecific functions are called. On Windows, the
threads that get spawned which are internal to the libc implementation
use reserve rather than commit memory, which shaves a few hundred kb.
- sigaltstack() is now supported on Windows, however it's currently not
able to be used to handle stack overflows, since crash signals are
still generated by WIN32. However the crash handler will still switch
to the alt stack, which is helpful in environments with tiny threads.
- Test binaries are now smaller. Many of the mandatory dependencies of
the test runner have been removed. This ensures many programs can do a
better job only linking the the thing they're testing. This caused the
test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb
- long double is no longer used in the implementation details of libc,
except in the APIs that define it. The old code that used long double
for time (instead of struct timespec) has now been thoroughly removed.
- ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing
backtraces itself, it'll just print a command you can run on the shell
using our new `cosmoaddr2line` program to view the backtrace.
- Crash report signal handling now works in a much better way. Instead
of terminating the process, it now relies on SA_RESETHAND so that the
default SIG_IGN behavior can terminate the process if necessary.
- Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
|
|
|
|
DEBUGF("write(tty) grace=%,ldns", timespec_tonanos(GetGraceTime()));
|
2020-06-15 14:18:57 +00:00
|
|
|
|
if ((rc = WriteVideoCall()) != -1) {
|
|
|
|
|
|
DEBUGF("write(tty) → %zd [grace=%,ldns]", rc,
|
Make improvements
- Every unit test now passes on Apple Silicon. The final piece of this
puzzle was porting our POSIX threads cancelation support, since that
works differently on ARM64 XNU vs. AMD64. Our semaphore support on
Apple Silicon is also superior now compared to AMD64, thanks to the
grand central dispatch library which lets *NSYNC locks go faster.
- The Cosmopolitan runtime is now more stable, particularly on Windows.
To do this, thread local storage is mandatory at all runtime levels,
and the innermost packages of the C library is no longer being built
using ASAN. TLS is being bootstrapped with a 128-byte TIB during the
process startup phase, and then later on the runtime re-allocates it
either statically or dynamically to support code using _Thread_local.
fork() and execve() now do a better job cooperating with threads. We
can now check how much stack memory is left in the process or thread
when functions like kprintf() / execve() etc. call alloca(), so that
ENOMEM can be raised, reduce a buffer size, or just print a warning.
- POSIX signal emulation is now implemented the same way kernels do it
with pthread_kill() and raise(). Any thread can interrupt any other
thread, regardless of what it's doing. If it's blocked on read/write
then the killer thread will cancel its i/o operation so that EINTR can
be returned in the mark thread immediately. If it's doing a tight CPU
bound operation, then that's also interrupted by the signal delivery.
Signal delivery works now by suspending a thread and pushing context
data structures onto its stack, and redirecting its execution to a
trampoline function, which calls SetThreadContext(GetCurrentThread())
when it's done.
- We're now doing a better job managing locks and handles. On NetBSD we
now close semaphore file descriptors in forked children. Semaphores on
Windows can now be canceled immediately, which means mutexes/condition
variables will now go faster. Apple Silicon semaphores can be canceled
too. We're now using Apple's pthread_yield() funciton. Apple _nocancel
syscalls are now used on XNU when appropriate to ensure pthread_cancel
requests aren't lost. The MbedTLS library has been updated to support
POSIX thread cancelations. See tool/build/runitd.c for an example of
how it can be used for production multi-threaded tls servers. Handles
on Windows now leak less often across processes. All i/o operations on
Windows are now overlapped, which means file pointers can no longer be
inherited across dup() and fork() for the time being.
- We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4()
which means, for example, that posix_spawn() now goes 3x faster. POSIX
spawn is also now more correct. Like Musl, it's now able to report the
failure code of execve() via a pipe although our approach favors using
shared memory to do that on systems that have a true vfork() function.
- We now spawn a thread to deliver SIGALRM to threads when setitimer()
is used. This enables the most precise wakeups the OS makes possible.
- The Cosmopolitan runtime now uses less memory. On NetBSD for example,
it turned out the kernel would actually commit the PT_GNU_STACK size
which caused RSS to be 6mb for every process. Now it's down to ~4kb.
On Apple Silicon, we reduce the mandatory upstream thread size to the
smallest possible size to reduce the memory overhead of Cosmo threads.
The examples directory has a program called greenbean which can spawn
a web server on Linux with 10,000 worker threads and have the memory
usage of the process be ~77mb. The 1024 byte overhead of POSIX-style
thread-local storage is now optional; it won't be allocated until the
pthread_setspecific/getspecific functions are called. On Windows, the
threads that get spawned which are internal to the libc implementation
use reserve rather than commit memory, which shaves a few hundred kb.
- sigaltstack() is now supported on Windows, however it's currently not
able to be used to handle stack overflows, since crash signals are
still generated by WIN32. However the crash handler will still switch
to the alt stack, which is helpful in environments with tiny threads.
- Test binaries are now smaller. Many of the mandatory dependencies of
the test runner have been removed. This ensures many programs can do a
better job only linking the the thing they're testing. This caused the
test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb
- long double is no longer used in the implementation details of libc,
except in the APIs that define it. The old code that used long double
for time (instead of struct timespec) has now been thoroughly removed.
- ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing
backtraces itself, it'll just print a command you can run on the shell
using our new `cosmoaddr2line` program to view the backtrace.
- Crash report signal handling now works in a much better way. Instead
of terminating the process, it now relies on SA_RESETHAND so that the
default SIG_IGN behavior can terminate the process if necessary.
- Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
|
|
|
|
timespec_tonanos(GetGraceTime()));
|
2020-06-15 14:18:57 +00:00
|
|
|
|
} else if (errno == EAGAIN || errno == EINTR) {
|
|
|
|
|
|
DEBUGF("write(tty) → EINTR");
|
|
|
|
|
|
longjmp(jbi_, 1);
|
|
|
|
|
|
} else if (errno == EPIPE) {
|
|
|
|
|
|
DEBUGF("write(tty) → EPIPE");
|
|
|
|
|
|
longjmp(jb_, 1);
|
|
|
|
|
|
} else {
|
|
|
|
|
|
FATALF("write(tty) → %s", strerror(errno));
|
|
|
|
|
|
}
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static void RefreshDisplay(void) {
|
Apply clang-format update to repo (#1154)
Commit bc6c183 introduced a bunch of discrepancies between what files
look like in the repo and what clang-format says they should look like.
However, there were already a few discrepancies prior to that. Most of
these discrepancies seemed to be unintentional, but a few of them were
load-bearing (e.g., a #include that violated header ordering needing
something to have been #defined by a 'later' #include.)
I opted to take what I hope is a relatively smooth-brained approach: I
reverted the .clang-format change, ran clang-format on the whole repo,
reapplied the .clang-format change, reran clang-format again, and then
reverted the commit that contained the first run. Thus the full effect
of this PR should only be to apply the changed formatting rules to the
repo, and from skimming the results, this seems to be the case.
My work can be checked by applying the short, manual commits, and then
rerunning the command listed in the autogenerated commits (those whose
messages I have prefixed auto:) and seeing if your results agree.
It might be that the other diffs should be fixed at some point but I'm
leaving that aside for now.
fd '\.c(c|pp)?$' --print0| xargs -0 clang-format -i
2024-04-25 17:38:00 +00:00
|
|
|
|
if (f1_ && f1_->n)
|
|
|
|
|
|
f1_->i = 0;
|
2020-06-15 14:18:57 +00:00
|
|
|
|
DimensionDisplay();
|
|
|
|
|
|
resized_ = false;
|
|
|
|
|
|
historyclear_ = true;
|
|
|
|
|
|
ttysend(outfd_, "\e[0m\e[H\e[3J");
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static void SetQuant(enum TtyQuantizationAlgorithm alg,
|
|
|
|
|
|
enum TtyQuantizationChannels chans,
|
|
|
|
|
|
enum TtyBlocksSelection blocks) {
|
|
|
|
|
|
tuned_ = true;
|
2020-08-25 11:23:25 +00:00
|
|
|
|
ttyquantsetup(alg, chans, blocks);
|
2020-06-15 14:18:57 +00:00
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static void SetQuantizationAlgorithm(enum TtyQuantizationAlgorithm alg) {
|
|
|
|
|
|
SetQuant(alg, TTYQUANT()->chans, TTYQUANT()->blocks);
|
|
|
|
|
|
/* TODO(jart): autotune */
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static void SetDithering(bool dither) {
|
|
|
|
|
|
tuned_ = true;
|
|
|
|
|
|
dither_ = dither;
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static optimizesize bool ProcessOptKey(int opt) {
|
|
|
|
|
|
switch (opt) {
|
|
|
|
|
|
case 's':
|
|
|
|
|
|
stats_ = !stats_;
|
|
|
|
|
|
return true;
|
|
|
|
|
|
case '3':
|
|
|
|
|
|
TTYQUANT()->blocks = kTtyBlocksCp437;
|
|
|
|
|
|
return true;
|
|
|
|
|
|
case '4':
|
|
|
|
|
|
TTYQUANT()->blocks = kTtyBlocksUnicode;
|
|
|
|
|
|
return true;
|
|
|
|
|
|
case 'd':
|
|
|
|
|
|
SetDithering(!dither_);
|
|
|
|
|
|
return true;
|
|
|
|
|
|
case 't':
|
|
|
|
|
|
SetQuantizationAlgorithm(kTtyQuantTrue);
|
|
|
|
|
|
return true;
|
|
|
|
|
|
case 'a':
|
|
|
|
|
|
SetQuantizationAlgorithm(kTtyQuantAnsi);
|
|
|
|
|
|
return true;
|
|
|
|
|
|
case 'x':
|
|
|
|
|
|
SetQuantizationAlgorithm(kTtyQuantXterm256);
|
|
|
|
|
|
return true;
|
|
|
|
|
|
case 'A':
|
|
|
|
|
|
istango_ = false;
|
|
|
|
|
|
memcpy(g_ansi2rgb_, &kCgaPalette, sizeof(kCgaPalette));
|
|
|
|
|
|
return true;
|
|
|
|
|
|
case 'T':
|
|
|
|
|
|
istango_ = true;
|
|
|
|
|
|
memcpy(g_ansi2rgb_, &kTangoPalette, sizeof(kTangoPalette));
|
|
|
|
|
|
return true;
|
|
|
|
|
|
default:
|
|
|
|
|
|
return false;
|
|
|
|
|
|
}
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static optimizesize void ReadKeyboard(void) {
|
|
|
|
|
|
char b[64];
|
|
|
|
|
|
int c, i, n, sgn;
|
|
|
|
|
|
memset(b, -1, sizeof(b));
|
|
|
|
|
|
b[0] = CTRL('B'); /* for eof case */
|
|
|
|
|
|
if ((n = read(infd_, &b, sizeof(b))) != -1) {
|
|
|
|
|
|
for (;;) {
|
|
|
|
|
|
i = 0;
|
|
|
|
|
|
c = b[i++];
|
|
|
|
|
|
if (!ProcessOptKey(c)) {
|
|
|
|
|
|
sgn = isupper(c) ? -1 : 1;
|
|
|
|
|
|
switch (c) {
|
|
|
|
|
|
case 'Y':
|
|
|
|
|
|
yonly_ = !yonly_;
|
|
|
|
|
|
ComputeColoringSolution();
|
|
|
|
|
|
break;
|
|
|
|
|
|
case 'S':
|
|
|
|
|
|
swing_ = swing_ == 219 ? 255 : 219;
|
|
|
|
|
|
ComputeColoringSolution();
|
|
|
|
|
|
break;
|
|
|
|
|
|
case 'p':
|
|
|
|
|
|
case 'P':
|
|
|
|
|
|
primaries_ = MOD(sgn + primaries_, ARRAYLEN(kPrimaries));
|
|
|
|
|
|
ComputeColoringSolution();
|
|
|
|
|
|
break;
|
|
|
|
|
|
case 'l':
|
|
|
|
|
|
case 'L':
|
|
|
|
|
|
lighting_ = MOD(sgn + lighting_, ARRAYLEN(kLightings));
|
|
|
|
|
|
ComputeColoringSolution();
|
|
|
|
|
|
break;
|
|
|
|
|
|
case 'g':
|
|
|
|
|
|
case 'G':
|
|
|
|
|
|
gamma_ += sgn * GAMMADELTA;
|
|
|
|
|
|
g_xterm256_gamma += sgn * GAMMADELTA;
|
|
|
|
|
|
ComputeColoringSolution();
|
|
|
|
|
|
break;
|
|
|
|
|
|
case 'k':
|
|
|
|
|
|
case 'K':
|
|
|
|
|
|
lumakernel_ = MOD(sgn + lumakernel_, ARRAYLEN(kMagikarp));
|
|
|
|
|
|
memcpy(g_magikarp, kMagikarp[lumakernel_], sizeof(kMagikarp[0]));
|
|
|
|
|
|
break;
|
|
|
|
|
|
case 'j':
|
|
|
|
|
|
case 'J':
|
|
|
|
|
|
chromakernel_ = MOD(sgn + chromakernel_, ARRAYLEN(kMagkern));
|
|
|
|
|
|
memcpy(g_magkern, kMagkern[chromakernel_], sizeof(kMagkern[0]));
|
|
|
|
|
|
break;
|
2021-02-03 23:53:33 +00:00
|
|
|
|
case 'q':
|
2020-06-15 14:18:57 +00:00
|
|
|
|
case CTRL('C'):
|
|
|
|
|
|
longjmp(jb_, 1);
|
|
|
|
|
|
break;
|
|
|
|
|
|
case CTRL('Z'):
|
|
|
|
|
|
ttyshowcursor(outfd_);
|
|
|
|
|
|
raise(SIGSTOP);
|
|
|
|
|
|
break;
|
|
|
|
|
|
case CTRL('G'):
|
|
|
|
|
|
sharp_ = (sharp_ + 1) % kSharpMAX;
|
|
|
|
|
|
break;
|
|
|
|
|
|
case CTRL('\\'):
|
|
|
|
|
|
raise(SIGQUIT);
|
|
|
|
|
|
break;
|
|
|
|
|
|
case CTRL('L'):
|
|
|
|
|
|
RefreshDisplay();
|
|
|
|
|
|
break;
|
|
|
|
|
|
case '\e':
|
|
|
|
|
|
if (n == 1) {
|
|
|
|
|
|
longjmp(jb_, 1); /* \e <𝟷𝟶𝟶ms*VTIME> is ESC */
|
|
|
|
|
|
}
|
|
|
|
|
|
switch (b[i++]) {
|
|
|
|
|
|
case '[':
|
|
|
|
|
|
switch (b[i++]) {
|
|
|
|
|
|
case 'A': /* "\e[A" is up arrow */
|
|
|
|
|
|
++volscale_;
|
|
|
|
|
|
break;
|
|
|
|
|
|
case 'B': /* "\e[B" is down arrow */
|
|
|
|
|
|
--volscale_;
|
|
|
|
|
|
break;
|
|
|
|
|
|
case 'C': /* "\e[C" is right arrow */
|
|
|
|
|
|
break;
|
|
|
|
|
|
case 'D': /* "\e[D" is left arrow */
|
|
|
|
|
|
break;
|
|
|
|
|
|
case '1':
|
|
|
|
|
|
switch (b[i++]) {
|
2021-03-14 03:39:17 +00:00
|
|
|
|
case '1':
|
|
|
|
|
|
switch (b[i++]) {
|
|
|
|
|
|
case '~': /* \e[11~ is F1 */
|
|
|
|
|
|
pf1_ = !pf1_;
|
|
|
|
|
|
break;
|
|
|
|
|
|
default:
|
|
|
|
|
|
break;
|
|
|
|
|
|
}
|
|
|
|
|
|
break;
|
|
|
|
|
|
case '2':
|
|
|
|
|
|
switch (b[i++]) {
|
|
|
|
|
|
case '~': /* \e[12~ is F2 */
|
|
|
|
|
|
pf2_ = !pf2_;
|
|
|
|
|
|
break;
|
|
|
|
|
|
default:
|
|
|
|
|
|
break;
|
|
|
|
|
|
}
|
|
|
|
|
|
break;
|
|
|
|
|
|
case '3':
|
|
|
|
|
|
switch (b[i++]) {
|
|
|
|
|
|
case '~': /* \e[13~ is F3 */
|
|
|
|
|
|
pf3_ = !pf3_;
|
|
|
|
|
|
break;
|
|
|
|
|
|
default:
|
|
|
|
|
|
break;
|
|
|
|
|
|
}
|
|
|
|
|
|
break;
|
|
|
|
|
|
case '4':
|
|
|
|
|
|
switch (b[i++]) {
|
|
|
|
|
|
case '~': /* \e[14~ is F4 */
|
|
|
|
|
|
pf4_ = !pf4_;
|
|
|
|
|
|
break;
|
|
|
|
|
|
default:
|
|
|
|
|
|
break;
|
|
|
|
|
|
}
|
|
|
|
|
|
break;
|
2020-06-15 14:18:57 +00:00
|
|
|
|
case '5':
|
|
|
|
|
|
switch (b[i++]) {
|
2021-03-14 03:39:17 +00:00
|
|
|
|
case '~': /* \e[15~ is F5 */
|
2020-06-15 14:18:57 +00:00
|
|
|
|
pf5_ = !pf5_;
|
|
|
|
|
|
break;
|
|
|
|
|
|
default:
|
|
|
|
|
|
break;
|
|
|
|
|
|
}
|
|
|
|
|
|
break;
|
|
|
|
|
|
case '7':
|
|
|
|
|
|
switch (b[i++]) {
|
2021-03-14 03:39:17 +00:00
|
|
|
|
case '~': /* \e[17~ is F6 */
|
2020-06-15 14:18:57 +00:00
|
|
|
|
pf6_ = !pf6_;
|
|
|
|
|
|
break;
|
|
|
|
|
|
default:
|
|
|
|
|
|
break;
|
|
|
|
|
|
}
|
|
|
|
|
|
break;
|
|
|
|
|
|
case '8':
|
|
|
|
|
|
switch (b[i++]) {
|
2021-03-14 03:39:17 +00:00
|
|
|
|
case '~': /* \e[18~ is F7 */
|
2020-06-15 14:18:57 +00:00
|
|
|
|
pf7_ = !pf7_;
|
|
|
|
|
|
break;
|
|
|
|
|
|
default:
|
|
|
|
|
|
break;
|
|
|
|
|
|
}
|
|
|
|
|
|
break;
|
|
|
|
|
|
case '9':
|
|
|
|
|
|
switch (b[i++]) {
|
2021-03-14 03:39:17 +00:00
|
|
|
|
case '~': /* \e[19~ is F8 */
|
2020-06-15 14:18:57 +00:00
|
|
|
|
pf8_ = !pf8_;
|
|
|
|
|
|
break;
|
|
|
|
|
|
default:
|
|
|
|
|
|
break;
|
|
|
|
|
|
}
|
|
|
|
|
|
break;
|
|
|
|
|
|
default:
|
|
|
|
|
|
break;
|
|
|
|
|
|
}
|
|
|
|
|
|
break;
|
|
|
|
|
|
case '2':
|
|
|
|
|
|
switch (b[i++]) {
|
|
|
|
|
|
case '0':
|
|
|
|
|
|
switch (b[i++]) {
|
2021-03-14 03:39:17 +00:00
|
|
|
|
case '~': /* \e[20~ is F9 */
|
2020-06-15 14:18:57 +00:00
|
|
|
|
pf9_ = !pf9_;
|
|
|
|
|
|
break;
|
|
|
|
|
|
default:
|
|
|
|
|
|
break;
|
|
|
|
|
|
}
|
|
|
|
|
|
break;
|
|
|
|
|
|
case '1':
|
|
|
|
|
|
switch (b[i++]) {
|
|
|
|
|
|
case '~': /* \e[21~ is F10 */
|
|
|
|
|
|
pf10_ = !pf10_;
|
|
|
|
|
|
break;
|
|
|
|
|
|
default:
|
|
|
|
|
|
break;
|
|
|
|
|
|
}
|
|
|
|
|
|
break;
|
|
|
|
|
|
case '3':
|
|
|
|
|
|
switch (b[i++]) {
|
|
|
|
|
|
case '~': /* \e[23~ is F11 */
|
|
|
|
|
|
pf11_ = !pf11_;
|
|
|
|
|
|
break;
|
|
|
|
|
|
default:
|
|
|
|
|
|
break;
|
|
|
|
|
|
}
|
|
|
|
|
|
break;
|
|
|
|
|
|
case '4':
|
|
|
|
|
|
switch (b[i++]) {
|
|
|
|
|
|
case '~': /* \e[24~ is F12 */
|
|
|
|
|
|
pf12_ = !pf12_;
|
|
|
|
|
|
break;
|
|
|
|
|
|
default:
|
|
|
|
|
|
break;
|
|
|
|
|
|
}
|
|
|
|
|
|
break;
|
|
|
|
|
|
default:
|
|
|
|
|
|
break;
|
|
|
|
|
|
}
|
|
|
|
|
|
break;
|
|
|
|
|
|
case '[':
|
|
|
|
|
|
switch (b[i++]) {
|
|
|
|
|
|
case 'A': /* \e[[A is F1 */
|
|
|
|
|
|
pf1_ = !pf1_;
|
|
|
|
|
|
break;
|
|
|
|
|
|
case 'B': /* \e[[B is F2 */
|
|
|
|
|
|
pf2_ = !pf2_;
|
|
|
|
|
|
break;
|
|
|
|
|
|
case 'C': /* \e[[C is F3 */
|
|
|
|
|
|
pf3_ = !pf3_;
|
|
|
|
|
|
break;
|
|
|
|
|
|
case 'D': /* \e[[D is F4 */
|
|
|
|
|
|
pf4_ = !pf4_;
|
|
|
|
|
|
break;
|
|
|
|
|
|
case 'E': /* \e[[E is F5 */
|
|
|
|
|
|
pf5_ = !pf5_;
|
|
|
|
|
|
break;
|
|
|
|
|
|
default:
|
|
|
|
|
|
break;
|
|
|
|
|
|
}
|
|
|
|
|
|
break;
|
|
|
|
|
|
default:
|
|
|
|
|
|
break;
|
|
|
|
|
|
}
|
|
|
|
|
|
break;
|
|
|
|
|
|
case 'O':
|
|
|
|
|
|
switch (b[i++]) {
|
|
|
|
|
|
case 'P': /* \eOP is F1 */
|
|
|
|
|
|
pf1_ = !pf1_;
|
|
|
|
|
|
break;
|
|
|
|
|
|
case 'Q': /* \eOQ is F2 */
|
|
|
|
|
|
pf2_ = !pf2_;
|
|
|
|
|
|
break;
|
|
|
|
|
|
case 'R': /* \eOR is F3 */
|
|
|
|
|
|
pf3_ = !pf3_;
|
|
|
|
|
|
break;
|
|
|
|
|
|
case 'S': /* \eOS is F4 */
|
|
|
|
|
|
pf4_ = !pf4_;
|
|
|
|
|
|
break;
|
2021-08-15 07:05:27 +00:00
|
|
|
|
case 'T': /* \eOT is F5 */
|
|
|
|
|
|
pf5_ = !pf5_;
|
|
|
|
|
|
break;
|
|
|
|
|
|
case 'U': /* \eOU is F6 */
|
|
|
|
|
|
pf6_ = !pf6_;
|
|
|
|
|
|
break;
|
|
|
|
|
|
case 'V': /* \eOV is F7 */
|
|
|
|
|
|
pf7_ = !pf7_;
|
|
|
|
|
|
break;
|
|
|
|
|
|
case 'W': /* \eOW is F8 */
|
|
|
|
|
|
pf8_ = !pf8_;
|
|
|
|
|
|
break;
|
|
|
|
|
|
case 'Y': /* \eOY is F10 */
|
|
|
|
|
|
pf10_ = !pf10_;
|
|
|
|
|
|
break;
|
|
|
|
|
|
case 'Z': /* \eOZ is F11 */
|
|
|
|
|
|
pf11_ = !pf11_;
|
|
|
|
|
|
break;
|
|
|
|
|
|
case '[': /* \eO[ is F12 */
|
|
|
|
|
|
pf12_ = !pf12_;
|
|
|
|
|
|
break;
|
2020-06-15 14:18:57 +00:00
|
|
|
|
default:
|
|
|
|
|
|
break;
|
|
|
|
|
|
}
|
|
|
|
|
|
break;
|
|
|
|
|
|
default:
|
|
|
|
|
|
break;
|
|
|
|
|
|
}
|
|
|
|
|
|
break;
|
|
|
|
|
|
default:
|
|
|
|
|
|
break;
|
|
|
|
|
|
}
|
|
|
|
|
|
}
|
|
|
|
|
|
if ((n -= i) <= 0) {
|
|
|
|
|
|
break;
|
|
|
|
|
|
} else {
|
|
|
|
|
|
memmove(b, b + i, sizeof(b) - i);
|
|
|
|
|
|
}
|
|
|
|
|
|
}
|
|
|
|
|
|
} else if (errno == EINTR) {
|
|
|
|
|
|
longjmp(jbi_, 1);
|
|
|
|
|
|
}
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static void PerformBestEffortIo(void) {
|
|
|
|
|
|
int toto, pollms;
|
|
|
|
|
|
struct pollfd fds[] = {
|
|
|
|
|
|
{infd_, POLLIN},
|
|
|
|
|
|
{outfd_, f1_ && f1_->n ? POLLOUT : 0},
|
|
|
|
|
|
};
|
Make improvements
- Every unit test now passes on Apple Silicon. The final piece of this
puzzle was porting our POSIX threads cancelation support, since that
works differently on ARM64 XNU vs. AMD64. Our semaphore support on
Apple Silicon is also superior now compared to AMD64, thanks to the
grand central dispatch library which lets *NSYNC locks go faster.
- The Cosmopolitan runtime is now more stable, particularly on Windows.
To do this, thread local storage is mandatory at all runtime levels,
and the innermost packages of the C library is no longer being built
using ASAN. TLS is being bootstrapped with a 128-byte TIB during the
process startup phase, and then later on the runtime re-allocates it
either statically or dynamically to support code using _Thread_local.
fork() and execve() now do a better job cooperating with threads. We
can now check how much stack memory is left in the process or thread
when functions like kprintf() / execve() etc. call alloca(), so that
ENOMEM can be raised, reduce a buffer size, or just print a warning.
- POSIX signal emulation is now implemented the same way kernels do it
with pthread_kill() and raise(). Any thread can interrupt any other
thread, regardless of what it's doing. If it's blocked on read/write
then the killer thread will cancel its i/o operation so that EINTR can
be returned in the mark thread immediately. If it's doing a tight CPU
bound operation, then that's also interrupted by the signal delivery.
Signal delivery works now by suspending a thread and pushing context
data structures onto its stack, and redirecting its execution to a
trampoline function, which calls SetThreadContext(GetCurrentThread())
when it's done.
- We're now doing a better job managing locks and handles. On NetBSD we
now close semaphore file descriptors in forked children. Semaphores on
Windows can now be canceled immediately, which means mutexes/condition
variables will now go faster. Apple Silicon semaphores can be canceled
too. We're now using Apple's pthread_yield() funciton. Apple _nocancel
syscalls are now used on XNU when appropriate to ensure pthread_cancel
requests aren't lost. The MbedTLS library has been updated to support
POSIX thread cancelations. See tool/build/runitd.c for an example of
how it can be used for production multi-threaded tls servers. Handles
on Windows now leak less often across processes. All i/o operations on
Windows are now overlapped, which means file pointers can no longer be
inherited across dup() and fork() for the time being.
- We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4()
which means, for example, that posix_spawn() now goes 3x faster. POSIX
spawn is also now more correct. Like Musl, it's now able to report the
failure code of execve() via a pipe although our approach favors using
shared memory to do that on systems that have a true vfork() function.
- We now spawn a thread to deliver SIGALRM to threads when setitimer()
is used. This enables the most precise wakeups the OS makes possible.
- The Cosmopolitan runtime now uses less memory. On NetBSD for example,
it turned out the kernel would actually commit the PT_GNU_STACK size
which caused RSS to be 6mb for every process. Now it's down to ~4kb.
On Apple Silicon, we reduce the mandatory upstream thread size to the
smallest possible size to reduce the memory overhead of Cosmo threads.
The examples directory has a program called greenbean which can spawn
a web server on Linux with 10,000 worker threads and have the memory
usage of the process be ~77mb. The 1024 byte overhead of POSIX-style
thread-local storage is now optional; it won't be allocated until the
pthread_setspecific/getspecific functions are called. On Windows, the
threads that get spawned which are internal to the libc implementation
use reserve rather than commit memory, which shaves a few hundred kb.
- sigaltstack() is now supported on Windows, however it's currently not
able to be used to handle stack overflows, since crash signals are
still generated by WIN32. However the crash handler will still switch
to the alt stack, which is helpful in environments with tiny threads.
- Test binaries are now smaller. Many of the mandatory dependencies of
the test runner have been removed. This ensures many programs can do a
better job only linking the the thing they're testing. This caused the
test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb
- long double is no longer used in the implementation details of libc,
except in the APIs that define it. The old code that used long double
for time (instead of struct timespec) has now been thoroughly removed.
- ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing
backtraces itself, it'll just print a command you can run on the shell
using our new `cosmoaddr2line` program to view the backtrace.
- Crash report signal handling now works in a much better way. Instead
of terminating the process, it now relies on SA_RESETHAND so that the
default SIG_IGN behavior can terminate the process if necessary.
- Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
|
|
|
|
pollms = MAX(0, timespec_tomillis(GetGraceTime()));
|
2020-06-15 14:18:57 +00:00
|
|
|
|
DEBUGF("poll() ms=%,d", pollms);
|
|
|
|
|
|
if ((toto = poll(fds, ARRAYLEN(fds), pollms)) != -1) {
|
|
|
|
|
|
DEBUGF("poll() toto=%d [grace=%,ldns]", toto,
|
Make improvements
- Every unit test now passes on Apple Silicon. The final piece of this
puzzle was porting our POSIX threads cancelation support, since that
works differently on ARM64 XNU vs. AMD64. Our semaphore support on
Apple Silicon is also superior now compared to AMD64, thanks to the
grand central dispatch library which lets *NSYNC locks go faster.
- The Cosmopolitan runtime is now more stable, particularly on Windows.
To do this, thread local storage is mandatory at all runtime levels,
and the innermost packages of the C library is no longer being built
using ASAN. TLS is being bootstrapped with a 128-byte TIB during the
process startup phase, and then later on the runtime re-allocates it
either statically or dynamically to support code using _Thread_local.
fork() and execve() now do a better job cooperating with threads. We
can now check how much stack memory is left in the process or thread
when functions like kprintf() / execve() etc. call alloca(), so that
ENOMEM can be raised, reduce a buffer size, or just print a warning.
- POSIX signal emulation is now implemented the same way kernels do it
with pthread_kill() and raise(). Any thread can interrupt any other
thread, regardless of what it's doing. If it's blocked on read/write
then the killer thread will cancel its i/o operation so that EINTR can
be returned in the mark thread immediately. If it's doing a tight CPU
bound operation, then that's also interrupted by the signal delivery.
Signal delivery works now by suspending a thread and pushing context
data structures onto its stack, and redirecting its execution to a
trampoline function, which calls SetThreadContext(GetCurrentThread())
when it's done.
- We're now doing a better job managing locks and handles. On NetBSD we
now close semaphore file descriptors in forked children. Semaphores on
Windows can now be canceled immediately, which means mutexes/condition
variables will now go faster. Apple Silicon semaphores can be canceled
too. We're now using Apple's pthread_yield() funciton. Apple _nocancel
syscalls are now used on XNU when appropriate to ensure pthread_cancel
requests aren't lost. The MbedTLS library has been updated to support
POSIX thread cancelations. See tool/build/runitd.c for an example of
how it can be used for production multi-threaded tls servers. Handles
on Windows now leak less often across processes. All i/o operations on
Windows are now overlapped, which means file pointers can no longer be
inherited across dup() and fork() for the time being.
- We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4()
which means, for example, that posix_spawn() now goes 3x faster. POSIX
spawn is also now more correct. Like Musl, it's now able to report the
failure code of execve() via a pipe although our approach favors using
shared memory to do that on systems that have a true vfork() function.
- We now spawn a thread to deliver SIGALRM to threads when setitimer()
is used. This enables the most precise wakeups the OS makes possible.
- The Cosmopolitan runtime now uses less memory. On NetBSD for example,
it turned out the kernel would actually commit the PT_GNU_STACK size
which caused RSS to be 6mb for every process. Now it's down to ~4kb.
On Apple Silicon, we reduce the mandatory upstream thread size to the
smallest possible size to reduce the memory overhead of Cosmo threads.
The examples directory has a program called greenbean which can spawn
a web server on Linux with 10,000 worker threads and have the memory
usage of the process be ~77mb. The 1024 byte overhead of POSIX-style
thread-local storage is now optional; it won't be allocated until the
pthread_setspecific/getspecific functions are called. On Windows, the
threads that get spawned which are internal to the libc implementation
use reserve rather than commit memory, which shaves a few hundred kb.
- sigaltstack() is now supported on Windows, however it's currently not
able to be used to handle stack overflows, since crash signals are
still generated by WIN32. However the crash handler will still switch
to the alt stack, which is helpful in environments with tiny threads.
- Test binaries are now smaller. Many of the mandatory dependencies of
the test runner have been removed. This ensures many programs can do a
better job only linking the the thing they're testing. This caused the
test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb
- long double is no longer used in the implementation details of libc,
except in the APIs that define it. The old code that used long double
for time (instead of struct timespec) has now been thoroughly removed.
- ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing
backtraces itself, it'll just print a command you can run on the shell
using our new `cosmoaddr2line` program to view the backtrace.
- Crash report signal handling now works in a much better way. Instead
of terminating the process, it now relies on SA_RESETHAND so that the
default SIG_IGN behavior can terminate the process if necessary.
- Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
|
|
|
|
timespec_tonanos(GetGraceTime()));
|
2020-06-15 14:18:57 +00:00
|
|
|
|
if (toto) {
|
Apply clang-format update to repo (#1154)
Commit bc6c183 introduced a bunch of discrepancies between what files
look like in the repo and what clang-format says they should look like.
However, there were already a few discrepancies prior to that. Most of
these discrepancies seemed to be unintentional, but a few of them were
load-bearing (e.g., a #include that violated header ordering needing
something to have been #defined by a 'later' #include.)
I opted to take what I hope is a relatively smooth-brained approach: I
reverted the .clang-format change, ran clang-format on the whole repo,
reapplied the .clang-format change, reran clang-format again, and then
reverted the commit that contained the first run. Thus the full effect
of this PR should only be to apply the changed formatting rules to the
repo, and from skimming the results, this seems to be the case.
My work can be checked by applying the short, manual commits, and then
rerunning the command listed in the autogenerated commits (those whose
messages I have prefixed auto:) and seeing if your results agree.
It might be that the other diffs should be fixed at some point but I'm
leaving that aside for now.
fd '\.c(c|pp)?$' --print0| xargs -0 clang-format -i
2024-04-25 17:38:00 +00:00
|
|
|
|
if (fds[0].revents & (POLLIN | POLLERR))
|
|
|
|
|
|
ReadKeyboard();
|
|
|
|
|
|
if (fds[1].revents & (POLLOUT | POLLERR))
|
|
|
|
|
|
WriteVideo();
|
2020-06-15 14:18:57 +00:00
|
|
|
|
}
|
|
|
|
|
|
} else if (errno == EINTR) {
|
|
|
|
|
|
DEBUGF("poll() → EINTR");
|
|
|
|
|
|
return;
|
|
|
|
|
|
} else {
|
|
|
|
|
|
FATALF("poll() → %s", strerror(errno));
|
|
|
|
|
|
}
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static void RestoreTty(void) {
|
|
|
|
|
|
DrainVideo();
|
Apply clang-format update to repo (#1154)
Commit bc6c183 introduced a bunch of discrepancies between what files
look like in the repo and what clang-format says they should look like.
However, there were already a few discrepancies prior to that. Most of
these discrepancies seemed to be unintentional, but a few of them were
load-bearing (e.g., a #include that violated header ordering needing
something to have been #defined by a 'later' #include.)
I opted to take what I hope is a relatively smooth-brained approach: I
reverted the .clang-format change, ran clang-format on the whole repo,
reapplied the .clang-format change, reran clang-format again, and then
reverted the commit that contained the first run. Thus the full effect
of this PR should only be to apply the changed formatting rules to the
repo, and from skimming the results, this seems to be the case.
My work can be checked by applying the short, manual commits, and then
rerunning the command listed in the autogenerated commits (those whose
messages I have prefixed auto:) and seeing if your results agree.
It might be that the other diffs should be fixed at some point but I'm
leaving that aside for now.
fd '\.c(c|pp)?$' --print0| xargs -0 clang-format -i
2024-04-25 17:38:00 +00:00
|
|
|
|
if (ttymode_)
|
|
|
|
|
|
ttysend(outfd_, "\r\n\e[J");
|
2020-06-15 14:18:57 +00:00
|
|
|
|
ttymode_ = false;
|
|
|
|
|
|
ttyraw(-1);
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static void HandleSignals(void) {
|
|
|
|
|
|
if (piped_) {
|
|
|
|
|
|
WARNF("SIGPIPE");
|
|
|
|
|
|
CloseSpeaker();
|
|
|
|
|
|
piped_ = false;
|
|
|
|
|
|
}
|
|
|
|
|
|
if (resized_) {
|
|
|
|
|
|
RefreshDisplay();
|
|
|
|
|
|
}
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static void PrintVideo(void) {
|
Make improvements
- Every unit test now passes on Apple Silicon. The final piece of this
puzzle was porting our POSIX threads cancelation support, since that
works differently on ARM64 XNU vs. AMD64. Our semaphore support on
Apple Silicon is also superior now compared to AMD64, thanks to the
grand central dispatch library which lets *NSYNC locks go faster.
- The Cosmopolitan runtime is now more stable, particularly on Windows.
To do this, thread local storage is mandatory at all runtime levels,
and the innermost packages of the C library is no longer being built
using ASAN. TLS is being bootstrapped with a 128-byte TIB during the
process startup phase, and then later on the runtime re-allocates it
either statically or dynamically to support code using _Thread_local.
fork() and execve() now do a better job cooperating with threads. We
can now check how much stack memory is left in the process or thread
when functions like kprintf() / execve() etc. call alloca(), so that
ENOMEM can be raised, reduce a buffer size, or just print a warning.
- POSIX signal emulation is now implemented the same way kernels do it
with pthread_kill() and raise(). Any thread can interrupt any other
thread, regardless of what it's doing. If it's blocked on read/write
then the killer thread will cancel its i/o operation so that EINTR can
be returned in the mark thread immediately. If it's doing a tight CPU
bound operation, then that's also interrupted by the signal delivery.
Signal delivery works now by suspending a thread and pushing context
data structures onto its stack, and redirecting its execution to a
trampoline function, which calls SetThreadContext(GetCurrentThread())
when it's done.
- We're now doing a better job managing locks and handles. On NetBSD we
now close semaphore file descriptors in forked children. Semaphores on
Windows can now be canceled immediately, which means mutexes/condition
variables will now go faster. Apple Silicon semaphores can be canceled
too. We're now using Apple's pthread_yield() funciton. Apple _nocancel
syscalls are now used on XNU when appropriate to ensure pthread_cancel
requests aren't lost. The MbedTLS library has been updated to support
POSIX thread cancelations. See tool/build/runitd.c for an example of
how it can be used for production multi-threaded tls servers. Handles
on Windows now leak less often across processes. All i/o operations on
Windows are now overlapped, which means file pointers can no longer be
inherited across dup() and fork() for the time being.
- We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4()
which means, for example, that posix_spawn() now goes 3x faster. POSIX
spawn is also now more correct. Like Musl, it's now able to report the
failure code of execve() via a pipe although our approach favors using
shared memory to do that on systems that have a true vfork() function.
- We now spawn a thread to deliver SIGALRM to threads when setitimer()
is used. This enables the most precise wakeups the OS makes possible.
- The Cosmopolitan runtime now uses less memory. On NetBSD for example,
it turned out the kernel would actually commit the PT_GNU_STACK size
which caused RSS to be 6mb for every process. Now it's down to ~4kb.
On Apple Silicon, we reduce the mandatory upstream thread size to the
smallest possible size to reduce the memory overhead of Cosmo threads.
The examples directory has a program called greenbean which can spawn
a web server on Linux with 10,000 worker threads and have the memory
usage of the process be ~77mb. The 1024 byte overhead of POSIX-style
thread-local storage is now optional; it won't be allocated until the
pthread_setspecific/getspecific functions are called. On Windows, the
threads that get spawned which are internal to the libc implementation
use reserve rather than commit memory, which shaves a few hundred kb.
- sigaltstack() is now supported on Windows, however it's currently not
able to be used to handle stack overflows, since crash signals are
still generated by WIN32. However the crash handler will still switch
to the alt stack, which is helpful in environments with tiny threads.
- Test binaries are now smaller. Many of the mandatory dependencies of
the test runner have been removed. This ensures many programs can do a
better job only linking the the thing they're testing. This caused the
test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb
- long double is no longer used in the implementation details of libc,
except in the APIs that define it. The old code that used long double
for time (instead of struct timespec) has now been thoroughly removed.
- ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing
backtraces itself, it'll just print a command you can run on the shell
using our new `cosmoaddr2line` program to view the backtrace.
- Crash report signal handling now works in a much better way. Instead
of terminating the process, it now relies on SA_RESETHAND so that the
default SIG_IGN behavior can terminate the process if necessary.
- Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
|
|
|
|
struct timespec decode_last, decode_end, next_tick, lag;
|
|
|
|
|
|
dura_ = timespec_frommicros(min(MAX_FRAMERATE, 1 / plm_get_framerate(plm_)) *
|
|
|
|
|
|
1e6);
|
2021-09-04 04:14:26 +00:00
|
|
|
|
INFOF("framerate=%f dura=%f", plm_get_framerate(plm_), dura_);
|
Make improvements
- Every unit test now passes on Apple Silicon. The final piece of this
puzzle was porting our POSIX threads cancelation support, since that
works differently on ARM64 XNU vs. AMD64. Our semaphore support on
Apple Silicon is also superior now compared to AMD64, thanks to the
grand central dispatch library which lets *NSYNC locks go faster.
- The Cosmopolitan runtime is now more stable, particularly on Windows.
To do this, thread local storage is mandatory at all runtime levels,
and the innermost packages of the C library is no longer being built
using ASAN. TLS is being bootstrapped with a 128-byte TIB during the
process startup phase, and then later on the runtime re-allocates it
either statically or dynamically to support code using _Thread_local.
fork() and execve() now do a better job cooperating with threads. We
can now check how much stack memory is left in the process or thread
when functions like kprintf() / execve() etc. call alloca(), so that
ENOMEM can be raised, reduce a buffer size, or just print a warning.
- POSIX signal emulation is now implemented the same way kernels do it
with pthread_kill() and raise(). Any thread can interrupt any other
thread, regardless of what it's doing. If it's blocked on read/write
then the killer thread will cancel its i/o operation so that EINTR can
be returned in the mark thread immediately. If it's doing a tight CPU
bound operation, then that's also interrupted by the signal delivery.
Signal delivery works now by suspending a thread and pushing context
data structures onto its stack, and redirecting its execution to a
trampoline function, which calls SetThreadContext(GetCurrentThread())
when it's done.
- We're now doing a better job managing locks and handles. On NetBSD we
now close semaphore file descriptors in forked children. Semaphores on
Windows can now be canceled immediately, which means mutexes/condition
variables will now go faster. Apple Silicon semaphores can be canceled
too. We're now using Apple's pthread_yield() funciton. Apple _nocancel
syscalls are now used on XNU when appropriate to ensure pthread_cancel
requests aren't lost. The MbedTLS library has been updated to support
POSIX thread cancelations. See tool/build/runitd.c for an example of
how it can be used for production multi-threaded tls servers. Handles
on Windows now leak less often across processes. All i/o operations on
Windows are now overlapped, which means file pointers can no longer be
inherited across dup() and fork() for the time being.
- We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4()
which means, for example, that posix_spawn() now goes 3x faster. POSIX
spawn is also now more correct. Like Musl, it's now able to report the
failure code of execve() via a pipe although our approach favors using
shared memory to do that on systems that have a true vfork() function.
- We now spawn a thread to deliver SIGALRM to threads when setitimer()
is used. This enables the most precise wakeups the OS makes possible.
- The Cosmopolitan runtime now uses less memory. On NetBSD for example,
it turned out the kernel would actually commit the PT_GNU_STACK size
which caused RSS to be 6mb for every process. Now it's down to ~4kb.
On Apple Silicon, we reduce the mandatory upstream thread size to the
smallest possible size to reduce the memory overhead of Cosmo threads.
The examples directory has a program called greenbean which can spawn
a web server on Linux with 10,000 worker threads and have the memory
usage of the process be ~77mb. The 1024 byte overhead of POSIX-style
thread-local storage is now optional; it won't be allocated until the
pthread_setspecific/getspecific functions are called. On Windows, the
threads that get spawned which are internal to the libc implementation
use reserve rather than commit memory, which shaves a few hundred kb.
- sigaltstack() is now supported on Windows, however it's currently not
able to be used to handle stack overflows, since crash signals are
still generated by WIN32. However the crash handler will still switch
to the alt stack, which is helpful in environments with tiny threads.
- Test binaries are now smaller. Many of the mandatory dependencies of
the test runner have been removed. This ensures many programs can do a
better job only linking the the thing they're testing. This caused the
test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb
- long double is no longer used in the implementation details of libc,
except in the APIs that define it. The old code that used long double
for time (instead of struct timespec) has now been thoroughly removed.
- ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing
backtraces itself, it'll just print a command you can run on the shell
using our new `cosmoaddr2line` program to view the backtrace.
- Crash report signal handling now works in a much better way. Instead
of terminating the process, it now relies on SA_RESETHAND so that the
default SIG_IGN behavior can terminate the process if necessary.
- Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
|
|
|
|
next_tick = deadline_ = decode_last = timespec_real();
|
|
|
|
|
|
next_tick = timespec_add(next_tick, dura_);
|
|
|
|
|
|
deadline_ = timespec_add(deadline_, dura_);
|
2020-06-15 14:18:57 +00:00
|
|
|
|
do {
|
Make improvements
- Every unit test now passes on Apple Silicon. The final piece of this
puzzle was porting our POSIX threads cancelation support, since that
works differently on ARM64 XNU vs. AMD64. Our semaphore support on
Apple Silicon is also superior now compared to AMD64, thanks to the
grand central dispatch library which lets *NSYNC locks go faster.
- The Cosmopolitan runtime is now more stable, particularly on Windows.
To do this, thread local storage is mandatory at all runtime levels,
and the innermost packages of the C library is no longer being built
using ASAN. TLS is being bootstrapped with a 128-byte TIB during the
process startup phase, and then later on the runtime re-allocates it
either statically or dynamically to support code using _Thread_local.
fork() and execve() now do a better job cooperating with threads. We
can now check how much stack memory is left in the process or thread
when functions like kprintf() / execve() etc. call alloca(), so that
ENOMEM can be raised, reduce a buffer size, or just print a warning.
- POSIX signal emulation is now implemented the same way kernels do it
with pthread_kill() and raise(). Any thread can interrupt any other
thread, regardless of what it's doing. If it's blocked on read/write
then the killer thread will cancel its i/o operation so that EINTR can
be returned in the mark thread immediately. If it's doing a tight CPU
bound operation, then that's also interrupted by the signal delivery.
Signal delivery works now by suspending a thread and pushing context
data structures onto its stack, and redirecting its execution to a
trampoline function, which calls SetThreadContext(GetCurrentThread())
when it's done.
- We're now doing a better job managing locks and handles. On NetBSD we
now close semaphore file descriptors in forked children. Semaphores on
Windows can now be canceled immediately, which means mutexes/condition
variables will now go faster. Apple Silicon semaphores can be canceled
too. We're now using Apple's pthread_yield() funciton. Apple _nocancel
syscalls are now used on XNU when appropriate to ensure pthread_cancel
requests aren't lost. The MbedTLS library has been updated to support
POSIX thread cancelations. See tool/build/runitd.c for an example of
how it can be used for production multi-threaded tls servers. Handles
on Windows now leak less often across processes. All i/o operations on
Windows are now overlapped, which means file pointers can no longer be
inherited across dup() and fork() for the time being.
- We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4()
which means, for example, that posix_spawn() now goes 3x faster. POSIX
spawn is also now more correct. Like Musl, it's now able to report the
failure code of execve() via a pipe although our approach favors using
shared memory to do that on systems that have a true vfork() function.
- We now spawn a thread to deliver SIGALRM to threads when setitimer()
is used. This enables the most precise wakeups the OS makes possible.
- The Cosmopolitan runtime now uses less memory. On NetBSD for example,
it turned out the kernel would actually commit the PT_GNU_STACK size
which caused RSS to be 6mb for every process. Now it's down to ~4kb.
On Apple Silicon, we reduce the mandatory upstream thread size to the
smallest possible size to reduce the memory overhead of Cosmo threads.
The examples directory has a program called greenbean which can spawn
a web server on Linux with 10,000 worker threads and have the memory
usage of the process be ~77mb. The 1024 byte overhead of POSIX-style
thread-local storage is now optional; it won't be allocated until the
pthread_setspecific/getspecific functions are called. On Windows, the
threads that get spawned which are internal to the libc implementation
use reserve rather than commit memory, which shaves a few hundred kb.
- sigaltstack() is now supported on Windows, however it's currently not
able to be used to handle stack overflows, since crash signals are
still generated by WIN32. However the crash handler will still switch
to the alt stack, which is helpful in environments with tiny threads.
- Test binaries are now smaller. Many of the mandatory dependencies of
the test runner have been removed. This ensures many programs can do a
better job only linking the the thing they're testing. This caused the
test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb
- long double is no longer used in the implementation details of libc,
except in the APIs that define it. The old code that used long double
for time (instead of struct timespec) has now been thoroughly removed.
- ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing
backtraces itself, it'll just print a command you can run on the shell
using our new `cosmoaddr2line` program to view the backtrace.
- Crash report signal handling now works in a much better way. Instead
of terminating the process, it now relies on SA_RESETHAND so that the
default SIG_IGN behavior can terminate the process if necessary.
- Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
|
|
|
|
DEBUGF("plm_decode [grace=%,ldns]", timespec_tonanos(GetGraceTime()));
|
|
|
|
|
|
decode_start_ = timespec_real();
|
|
|
|
|
|
plm_decode(plm_,
|
|
|
|
|
|
timespec_tofloat(timespec_sub(decode_start_, decode_last)));
|
2020-06-15 14:18:57 +00:00
|
|
|
|
decode_last = decode_start_;
|
Make improvements
- Every unit test now passes on Apple Silicon. The final piece of this
puzzle was porting our POSIX threads cancelation support, since that
works differently on ARM64 XNU vs. AMD64. Our semaphore support on
Apple Silicon is also superior now compared to AMD64, thanks to the
grand central dispatch library which lets *NSYNC locks go faster.
- The Cosmopolitan runtime is now more stable, particularly on Windows.
To do this, thread local storage is mandatory at all runtime levels,
and the innermost packages of the C library is no longer being built
using ASAN. TLS is being bootstrapped with a 128-byte TIB during the
process startup phase, and then later on the runtime re-allocates it
either statically or dynamically to support code using _Thread_local.
fork() and execve() now do a better job cooperating with threads. We
can now check how much stack memory is left in the process or thread
when functions like kprintf() / execve() etc. call alloca(), so that
ENOMEM can be raised, reduce a buffer size, or just print a warning.
- POSIX signal emulation is now implemented the same way kernels do it
with pthread_kill() and raise(). Any thread can interrupt any other
thread, regardless of what it's doing. If it's blocked on read/write
then the killer thread will cancel its i/o operation so that EINTR can
be returned in the mark thread immediately. If it's doing a tight CPU
bound operation, then that's also interrupted by the signal delivery.
Signal delivery works now by suspending a thread and pushing context
data structures onto its stack, and redirecting its execution to a
trampoline function, which calls SetThreadContext(GetCurrentThread())
when it's done.
- We're now doing a better job managing locks and handles. On NetBSD we
now close semaphore file descriptors in forked children. Semaphores on
Windows can now be canceled immediately, which means mutexes/condition
variables will now go faster. Apple Silicon semaphores can be canceled
too. We're now using Apple's pthread_yield() funciton. Apple _nocancel
syscalls are now used on XNU when appropriate to ensure pthread_cancel
requests aren't lost. The MbedTLS library has been updated to support
POSIX thread cancelations. See tool/build/runitd.c for an example of
how it can be used for production multi-threaded tls servers. Handles
on Windows now leak less often across processes. All i/o operations on
Windows are now overlapped, which means file pointers can no longer be
inherited across dup() and fork() for the time being.
- We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4()
which means, for example, that posix_spawn() now goes 3x faster. POSIX
spawn is also now more correct. Like Musl, it's now able to report the
failure code of execve() via a pipe although our approach favors using
shared memory to do that on systems that have a true vfork() function.
- We now spawn a thread to deliver SIGALRM to threads when setitimer()
is used. This enables the most precise wakeups the OS makes possible.
- The Cosmopolitan runtime now uses less memory. On NetBSD for example,
it turned out the kernel would actually commit the PT_GNU_STACK size
which caused RSS to be 6mb for every process. Now it's down to ~4kb.
On Apple Silicon, we reduce the mandatory upstream thread size to the
smallest possible size to reduce the memory overhead of Cosmo threads.
The examples directory has a program called greenbean which can spawn
a web server on Linux with 10,000 worker threads and have the memory
usage of the process be ~77mb. The 1024 byte overhead of POSIX-style
thread-local storage is now optional; it won't be allocated until the
pthread_setspecific/getspecific functions are called. On Windows, the
threads that get spawned which are internal to the libc implementation
use reserve rather than commit memory, which shaves a few hundred kb.
- sigaltstack() is now supported on Windows, however it's currently not
able to be used to handle stack overflows, since crash signals are
still generated by WIN32. However the crash handler will still switch
to the alt stack, which is helpful in environments with tiny threads.
- Test binaries are now smaller. Many of the mandatory dependencies of
the test runner have been removed. This ensures many programs can do a
better job only linking the the thing they're testing. This caused the
test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb
- long double is no longer used in the implementation details of libc,
except in the APIs that define it. The old code that used long double
for time (instead of struct timespec) has now been thoroughly removed.
- ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing
backtraces itself, it'll just print a command you can run on the shell
using our new `cosmoaddr2line` program to view the backtrace.
- Crash report signal handling now works in a much better way. Instead
of terminating the process, it now relies on SA_RESETHAND so that the
default SIG_IGN behavior can terminate the process if necessary.
- Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
|
|
|
|
decode_end = timespec_real();
|
|
|
|
|
|
lag = timespec_sub(decode_end, decode_start_);
|
|
|
|
|
|
while (timespec_cmp(timespec_add(decode_end, lag), next_tick) > 0) {
|
|
|
|
|
|
next_tick = timespec_add(next_tick, dura_);
|
|
|
|
|
|
}
|
|
|
|
|
|
deadline_ = timespec_sub(next_tick, lag);
|
2020-06-15 14:18:57 +00:00
|
|
|
|
if (gotvideo_ || !plm_get_video_enabled(plm_)) {
|
|
|
|
|
|
gotvideo_ = false;
|
2021-09-04 04:14:26 +00:00
|
|
|
|
INFOF("entering printvideo event loop (lag=%,ldns, grace=%,ldns)",
|
Make improvements
- Every unit test now passes on Apple Silicon. The final piece of this
puzzle was porting our POSIX threads cancelation support, since that
works differently on ARM64 XNU vs. AMD64. Our semaphore support on
Apple Silicon is also superior now compared to AMD64, thanks to the
grand central dispatch library which lets *NSYNC locks go faster.
- The Cosmopolitan runtime is now more stable, particularly on Windows.
To do this, thread local storage is mandatory at all runtime levels,
and the innermost packages of the C library is no longer being built
using ASAN. TLS is being bootstrapped with a 128-byte TIB during the
process startup phase, and then later on the runtime re-allocates it
either statically or dynamically to support code using _Thread_local.
fork() and execve() now do a better job cooperating with threads. We
can now check how much stack memory is left in the process or thread
when functions like kprintf() / execve() etc. call alloca(), so that
ENOMEM can be raised, reduce a buffer size, or just print a warning.
- POSIX signal emulation is now implemented the same way kernels do it
with pthread_kill() and raise(). Any thread can interrupt any other
thread, regardless of what it's doing. If it's blocked on read/write
then the killer thread will cancel its i/o operation so that EINTR can
be returned in the mark thread immediately. If it's doing a tight CPU
bound operation, then that's also interrupted by the signal delivery.
Signal delivery works now by suspending a thread and pushing context
data structures onto its stack, and redirecting its execution to a
trampoline function, which calls SetThreadContext(GetCurrentThread())
when it's done.
- We're now doing a better job managing locks and handles. On NetBSD we
now close semaphore file descriptors in forked children. Semaphores on
Windows can now be canceled immediately, which means mutexes/condition
variables will now go faster. Apple Silicon semaphores can be canceled
too. We're now using Apple's pthread_yield() funciton. Apple _nocancel
syscalls are now used on XNU when appropriate to ensure pthread_cancel
requests aren't lost. The MbedTLS library has been updated to support
POSIX thread cancelations. See tool/build/runitd.c for an example of
how it can be used for production multi-threaded tls servers. Handles
on Windows now leak less often across processes. All i/o operations on
Windows are now overlapped, which means file pointers can no longer be
inherited across dup() and fork() for the time being.
- We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4()
which means, for example, that posix_spawn() now goes 3x faster. POSIX
spawn is also now more correct. Like Musl, it's now able to report the
failure code of execve() via a pipe although our approach favors using
shared memory to do that on systems that have a true vfork() function.
- We now spawn a thread to deliver SIGALRM to threads when setitimer()
is used. This enables the most precise wakeups the OS makes possible.
- The Cosmopolitan runtime now uses less memory. On NetBSD for example,
it turned out the kernel would actually commit the PT_GNU_STACK size
which caused RSS to be 6mb for every process. Now it's down to ~4kb.
On Apple Silicon, we reduce the mandatory upstream thread size to the
smallest possible size to reduce the memory overhead of Cosmo threads.
The examples directory has a program called greenbean which can spawn
a web server on Linux with 10,000 worker threads and have the memory
usage of the process be ~77mb. The 1024 byte overhead of POSIX-style
thread-local storage is now optional; it won't be allocated until the
pthread_setspecific/getspecific functions are called. On Windows, the
threads that get spawned which are internal to the libc implementation
use reserve rather than commit memory, which shaves a few hundred kb.
- sigaltstack() is now supported on Windows, however it's currently not
able to be used to handle stack overflows, since crash signals are
still generated by WIN32. However the crash handler will still switch
to the alt stack, which is helpful in environments with tiny threads.
- Test binaries are now smaller. Many of the mandatory dependencies of
the test runner have been removed. This ensures many programs can do a
better job only linking the the thing they're testing. This caused the
test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb
- long double is no longer used in the implementation details of libc,
except in the APIs that define it. The old code that used long double
for time (instead of struct timespec) has now been thoroughly removed.
- ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing
backtraces itself, it'll just print a command you can run on the shell
using our new `cosmoaddr2line` program to view the backtrace.
- Crash report signal handling now works in a much better way. Instead
of terminating the process, it now relies on SA_RESETHAND so that the
default SIG_IGN behavior can terminate the process if necessary.
- Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
|
|
|
|
timespec_tonanos(lag), timespec_tonanos(GetGraceTime()));
|
2020-06-15 14:18:57 +00:00
|
|
|
|
}
|
|
|
|
|
|
do {
|
|
|
|
|
|
if (!setjmp(jbi_)) {
|
|
|
|
|
|
PerformBestEffortIo();
|
|
|
|
|
|
}
|
|
|
|
|
|
HandleSignals();
|
Make improvements
- Every unit test now passes on Apple Silicon. The final piece of this
puzzle was porting our POSIX threads cancelation support, since that
works differently on ARM64 XNU vs. AMD64. Our semaphore support on
Apple Silicon is also superior now compared to AMD64, thanks to the
grand central dispatch library which lets *NSYNC locks go faster.
- The Cosmopolitan runtime is now more stable, particularly on Windows.
To do this, thread local storage is mandatory at all runtime levels,
and the innermost packages of the C library is no longer being built
using ASAN. TLS is being bootstrapped with a 128-byte TIB during the
process startup phase, and then later on the runtime re-allocates it
either statically or dynamically to support code using _Thread_local.
fork() and execve() now do a better job cooperating with threads. We
can now check how much stack memory is left in the process or thread
when functions like kprintf() / execve() etc. call alloca(), so that
ENOMEM can be raised, reduce a buffer size, or just print a warning.
- POSIX signal emulation is now implemented the same way kernels do it
with pthread_kill() and raise(). Any thread can interrupt any other
thread, regardless of what it's doing. If it's blocked on read/write
then the killer thread will cancel its i/o operation so that EINTR can
be returned in the mark thread immediately. If it's doing a tight CPU
bound operation, then that's also interrupted by the signal delivery.
Signal delivery works now by suspending a thread and pushing context
data structures onto its stack, and redirecting its execution to a
trampoline function, which calls SetThreadContext(GetCurrentThread())
when it's done.
- We're now doing a better job managing locks and handles. On NetBSD we
now close semaphore file descriptors in forked children. Semaphores on
Windows can now be canceled immediately, which means mutexes/condition
variables will now go faster. Apple Silicon semaphores can be canceled
too. We're now using Apple's pthread_yield() funciton. Apple _nocancel
syscalls are now used on XNU when appropriate to ensure pthread_cancel
requests aren't lost. The MbedTLS library has been updated to support
POSIX thread cancelations. See tool/build/runitd.c for an example of
how it can be used for production multi-threaded tls servers. Handles
on Windows now leak less often across processes. All i/o operations on
Windows are now overlapped, which means file pointers can no longer be
inherited across dup() and fork() for the time being.
- We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4()
which means, for example, that posix_spawn() now goes 3x faster. POSIX
spawn is also now more correct. Like Musl, it's now able to report the
failure code of execve() via a pipe although our approach favors using
shared memory to do that on systems that have a true vfork() function.
- We now spawn a thread to deliver SIGALRM to threads when setitimer()
is used. This enables the most precise wakeups the OS makes possible.
- The Cosmopolitan runtime now uses less memory. On NetBSD for example,
it turned out the kernel would actually commit the PT_GNU_STACK size
which caused RSS to be 6mb for every process. Now it's down to ~4kb.
On Apple Silicon, we reduce the mandatory upstream thread size to the
smallest possible size to reduce the memory overhead of Cosmo threads.
The examples directory has a program called greenbean which can spawn
a web server on Linux with 10,000 worker threads and have the memory
usage of the process be ~77mb. The 1024 byte overhead of POSIX-style
thread-local storage is now optional; it won't be allocated until the
pthread_setspecific/getspecific functions are called. On Windows, the
threads that get spawned which are internal to the libc implementation
use reserve rather than commit memory, which shaves a few hundred kb.
- sigaltstack() is now supported on Windows, however it's currently not
able to be used to handle stack overflows, since crash signals are
still generated by WIN32. However the crash handler will still switch
to the alt stack, which is helpful in environments with tiny threads.
- Test binaries are now smaller. Many of the mandatory dependencies of
the test runner have been removed. This ensures many programs can do a
better job only linking the the thing they're testing. This caused the
test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb
- long double is no longer used in the implementation details of libc,
except in the APIs that define it. The old code that used long double
for time (instead of struct timespec) has now been thoroughly removed.
- ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing
backtraces itself, it'll just print a command you can run on the shell
using our new `cosmoaddr2line` program to view the backtrace.
- Crash report signal handling now works in a much better way. Instead
of terminating the process, it now relies on SA_RESETHAND so that the
default SIG_IGN behavior can terminate the process if necessary.
- Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
|
|
|
|
} while (timespec_tomillis(GetGraceTime()) > 0);
|
2020-06-15 14:18:57 +00:00
|
|
|
|
} while (plm_ && !plm_has_ended(plm_));
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static bool AskUserYesOrNoQuestion(const char *prompt) {
|
|
|
|
|
|
char c;
|
Apply clang-format update to repo (#1154)
Commit bc6c183 introduced a bunch of discrepancies between what files
look like in the repo and what clang-format says they should look like.
However, there were already a few discrepancies prior to that. Most of
these discrepancies seemed to be unintentional, but a few of them were
load-bearing (e.g., a #include that violated header ordering needing
something to have been #defined by a 'later' #include.)
I opted to take what I hope is a relatively smooth-brained approach: I
reverted the .clang-format change, ran clang-format on the whole repo,
reapplied the .clang-format change, reran clang-format again, and then
reverted the commit that contained the first run. Thus the full effect
of this PR should only be to apply the changed formatting rules to the
repo, and from skimming the results, this seems to be the case.
My work can be checked by applying the short, manual commits, and then
rerunning the command listed in the autogenerated commits (those whose
messages I have prefixed auto:) and seeing if your results agree.
It might be that the other diffs should be fixed at some point but I'm
leaving that aside for now.
fd '\.c(c|pp)?$' --print0| xargs -0 clang-format -i
2024-04-25 17:38:00 +00:00
|
|
|
|
if (yes_ || !ttymode_)
|
|
|
|
|
|
return true;
|
2020-06-15 14:18:57 +00:00
|
|
|
|
ttysend(outfd_, "\r\e[K");
|
|
|
|
|
|
ttysend(outfd_, prompt);
|
|
|
|
|
|
ttysend(outfd_, " [yn] ");
|
|
|
|
|
|
poll((struct pollfd[]){{infd_, POLLIN}}, 1, -1);
|
|
|
|
|
|
c = 0, read(infd_, &c, 1);
|
|
|
|
|
|
ttysend(infd_, "\r\e[K");
|
|
|
|
|
|
return c == 'y' || c == 'Y';
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static bool CanPlayAudio(void) {
|
|
|
|
|
|
if (ffplay_ || sox_) {
|
|
|
|
|
|
return true;
|
|
|
|
|
|
} else if (AskUserYesOrNoQuestion(
|
|
|
|
|
|
"ffplay not found; continue without audio?")) {
|
|
|
|
|
|
return false;
|
|
|
|
|
|
} else {
|
|
|
|
|
|
longjmp(jb_, 1);
|
|
|
|
|
|
}
|
|
|
|
|
|
}
|
|
|
|
|
|
|
2023-12-23 05:39:27 +00:00
|
|
|
|
static void PrintUsage(int rc, int fd) {
|
|
|
|
|
|
tinyprint(fd, "Usage: ", program_invocation_name, USAGE, NULL);
|
2020-06-15 14:18:57 +00:00
|
|
|
|
exit(rc);
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static void GetOpts(int argc, char *argv[]) {
|
|
|
|
|
|
int opt;
|
Make improvements
- Every unit test now passes on Apple Silicon. The final piece of this
puzzle was porting our POSIX threads cancelation support, since that
works differently on ARM64 XNU vs. AMD64. Our semaphore support on
Apple Silicon is also superior now compared to AMD64, thanks to the
grand central dispatch library which lets *NSYNC locks go faster.
- The Cosmopolitan runtime is now more stable, particularly on Windows.
To do this, thread local storage is mandatory at all runtime levels,
and the innermost packages of the C library is no longer being built
using ASAN. TLS is being bootstrapped with a 128-byte TIB during the
process startup phase, and then later on the runtime re-allocates it
either statically or dynamically to support code using _Thread_local.
fork() and execve() now do a better job cooperating with threads. We
can now check how much stack memory is left in the process or thread
when functions like kprintf() / execve() etc. call alloca(), so that
ENOMEM can be raised, reduce a buffer size, or just print a warning.
- POSIX signal emulation is now implemented the same way kernels do it
with pthread_kill() and raise(). Any thread can interrupt any other
thread, regardless of what it's doing. If it's blocked on read/write
then the killer thread will cancel its i/o operation so that EINTR can
be returned in the mark thread immediately. If it's doing a tight CPU
bound operation, then that's also interrupted by the signal delivery.
Signal delivery works now by suspending a thread and pushing context
data structures onto its stack, and redirecting its execution to a
trampoline function, which calls SetThreadContext(GetCurrentThread())
when it's done.
- We're now doing a better job managing locks and handles. On NetBSD we
now close semaphore file descriptors in forked children. Semaphores on
Windows can now be canceled immediately, which means mutexes/condition
variables will now go faster. Apple Silicon semaphores can be canceled
too. We're now using Apple's pthread_yield() funciton. Apple _nocancel
syscalls are now used on XNU when appropriate to ensure pthread_cancel
requests aren't lost. The MbedTLS library has been updated to support
POSIX thread cancelations. See tool/build/runitd.c for an example of
how it can be used for production multi-threaded tls servers. Handles
on Windows now leak less often across processes. All i/o operations on
Windows are now overlapped, which means file pointers can no longer be
inherited across dup() and fork() for the time being.
- We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4()
which means, for example, that posix_spawn() now goes 3x faster. POSIX
spawn is also now more correct. Like Musl, it's now able to report the
failure code of execve() via a pipe although our approach favors using
shared memory to do that on systems that have a true vfork() function.
- We now spawn a thread to deliver SIGALRM to threads when setitimer()
is used. This enables the most precise wakeups the OS makes possible.
- The Cosmopolitan runtime now uses less memory. On NetBSD for example,
it turned out the kernel would actually commit the PT_GNU_STACK size
which caused RSS to be 6mb for every process. Now it's down to ~4kb.
On Apple Silicon, we reduce the mandatory upstream thread size to the
smallest possible size to reduce the memory overhead of Cosmo threads.
The examples directory has a program called greenbean which can spawn
a web server on Linux with 10,000 worker threads and have the memory
usage of the process be ~77mb. The 1024 byte overhead of POSIX-style
thread-local storage is now optional; it won't be allocated until the
pthread_setspecific/getspecific functions are called. On Windows, the
threads that get spawned which are internal to the libc implementation
use reserve rather than commit memory, which shaves a few hundred kb.
- sigaltstack() is now supported on Windows, however it's currently not
able to be used to handle stack overflows, since crash signals are
still generated by WIN32. However the crash handler will still switch
to the alt stack, which is helpful in environments with tiny threads.
- Test binaries are now smaller. Many of the mandatory dependencies of
the test runner have been removed. This ensures many programs can do a
better job only linking the the thing they're testing. This caused the
test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb
- long double is no longer used in the implementation details of libc,
except in the APIs that define it. The old code that used long double
for time (instead of struct timespec) has now been thoroughly removed.
- ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing
backtraces itself, it'll just print a command you can run on the shell
using our new `cosmoaddr2line` program to view the backtrace.
- Crash report signal handling now works in a much better way. Instead
of terminating the process, it now relies on SA_RESETHAND so that the
default SIG_IGN behavior can terminate the process if necessary.
- Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
|
|
|
|
snprintf(logpath_, sizeof(logpath_), "%s%s.log", __get_tmpdir(),
|
2022-05-13 20:31:21 +00:00
|
|
|
|
firstnonnull(program_invocation_short_name, "unknown"));
|
2020-06-15 14:18:57 +00:00
|
|
|
|
while ((opt = getopt(argc, argv, "?34AGSTVYabdfhnpstxyzvL:")) != -1) {
|
|
|
|
|
|
switch (opt) {
|
|
|
|
|
|
case 'y':
|
|
|
|
|
|
yes_ = true;
|
|
|
|
|
|
break;
|
|
|
|
|
|
case 'v':
|
2021-03-01 07:42:35 +00:00
|
|
|
|
++__log_level;
|
2020-06-15 14:18:57 +00:00
|
|
|
|
break;
|
|
|
|
|
|
case 'L':
|
|
|
|
|
|
snprintf(logpath_, sizeof(logpath_), "%s", optarg);
|
|
|
|
|
|
break;
|
|
|
|
|
|
case 'Y':
|
|
|
|
|
|
yonly_ = true;
|
|
|
|
|
|
break;
|
|
|
|
|
|
case 'h':
|
2023-12-23 05:39:27 +00:00
|
|
|
|
case '?':
|
2020-06-15 14:18:57 +00:00
|
|
|
|
default:
|
|
|
|
|
|
if (!ProcessOptKey(opt)) {
|
2023-12-23 05:39:27 +00:00
|
|
|
|
if (opt == optopt) {
|
|
|
|
|
|
PrintUsage(EXIT_SUCCESS, STDOUT_FILENO);
|
|
|
|
|
|
} else {
|
|
|
|
|
|
PrintUsage(EX_USAGE, STDERR_FILENO);
|
|
|
|
|
|
}
|
2020-06-15 14:18:57 +00:00
|
|
|
|
}
|
|
|
|
|
|
}
|
|
|
|
|
|
}
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static void OnExit(void) {
|
Apply clang-format update to repo (#1154)
Commit bc6c183 introduced a bunch of discrepancies between what files
look like in the repo and what clang-format says they should look like.
However, there were already a few discrepancies prior to that. Most of
these discrepancies seemed to be unintentional, but a few of them were
load-bearing (e.g., a #include that violated header ordering needing
something to have been #defined by a 'later' #include.)
I opted to take what I hope is a relatively smooth-brained approach: I
reverted the .clang-format change, ran clang-format on the whole repo,
reapplied the .clang-format change, reran clang-format again, and then
reverted the commit that contained the first run. Thus the full effect
of this PR should only be to apply the changed formatting rules to the
repo, and from skimming the results, this seems to be the case.
My work can be checked by applying the short, manual commits, and then
rerunning the command listed in the autogenerated commits (those whose
messages I have prefixed auto:) and seeing if your results agree.
It might be that the other diffs should be fixed at some point but I'm
leaving that aside for now.
fd '\.c(c|pp)?$' --print0| xargs -0 clang-format -i
2024-04-25 17:38:00 +00:00
|
|
|
|
if (playpid_)
|
|
|
|
|
|
kill(playpid_, SIGTERM), sched_yield();
|
|
|
|
|
|
if (plm_)
|
|
|
|
|
|
plm_destroy(plm_), plm_ = NULL;
|
2020-06-15 14:18:57 +00:00
|
|
|
|
YCbCrFree(&ycbcr_);
|
|
|
|
|
|
RestoreTty();
|
|
|
|
|
|
ttyidentclear(&ti_);
|
2022-09-13 06:10:38 +00:00
|
|
|
|
close(infd_), infd_ = -1;
|
|
|
|
|
|
close(outfd_), outfd_ = -1;
|
2023-07-06 13:57:28 +00:00
|
|
|
|
free(graphic_[0].b);
|
|
|
|
|
|
free(graphic_[1].b);
|
|
|
|
|
|
free(vtframe_[0].b);
|
|
|
|
|
|
free(vtframe_[1].b);
|
|
|
|
|
|
free(xtcodes_);
|
|
|
|
|
|
free(audio_);
|
2020-06-15 14:18:57 +00:00
|
|
|
|
CloseSpeaker();
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static void MakeLatencyLittleLessBad(void) {
|
2022-09-13 18:20:35 +00:00
|
|
|
|
LOGIFNEG1(sys_mlockall(MCL_CURRENT));
|
2020-06-15 14:18:57 +00:00
|
|
|
|
LOGIFNEG1(nice(-5));
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static void PickDefaults(void) {
|
|
|
|
|
|
/*
|
|
|
|
|
|
* Direct color ain't true color -- it just means xterm does the
|
|
|
|
|
|
* xterm256 rgb quantization for you. we're better at xterm256
|
|
|
|
|
|
* than xterm is, so we don't need the training wheels.
|
|
|
|
|
|
*
|
|
|
|
|
|
* strcmp(nulltoempty(getenv("TERM")), "xterm-direct") == 0
|
|
|
|
|
|
*/
|
|
|
|
|
|
if (strcmp(nulltoempty(getenv("TERM")), "xterm-kitty") == 0) {
|
2020-08-25 11:23:25 +00:00
|
|
|
|
ttyquantsetup(kTtyQuantTrue, TTYQUANT()->chans, kTtyBlocksUnicode);
|
2020-06-15 14:18:57 +00:00
|
|
|
|
}
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static void RenounceSpecialPrivileges(void) {
|
2022-07-25 02:40:32 +00:00
|
|
|
|
if (issetugid()) {
|
2020-06-15 14:18:57 +00:00
|
|
|
|
setegid(getgid());
|
|
|
|
|
|
seteuid(getuid());
|
|
|
|
|
|
}
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
#define FBIOGET_VSCREENINFO 0x4600
|
|
|
|
|
|
#define FBIOGET_FSCREENINFO 0x4602
|
|
|
|
|
|
|
|
|
|
|
|
static void TryToOpenFrameBuffer(void) {
|
|
|
|
|
|
/*
|
|
|
|
|
|
* Linux (i.e. without some X or Wayland thing running on top of it)
|
|
|
|
|
|
* is barely able to display any non-ascii characters, so things look
|
|
|
|
|
|
* much better if we can access the framebuffer.
|
|
|
|
|
|
*/
|
|
|
|
|
|
int rc;
|
|
|
|
|
|
fb0_.fd = -1;
|
|
|
|
|
|
fb0_.path = NULL;
|
|
|
|
|
|
if (!isempty(getenv("FRAMEBUFFER"))) {
|
|
|
|
|
|
fb0_.path = strdup(getenv("FRAMEBUFFER"));
|
|
|
|
|
|
} else if (strcmp(nulltoempty(getenv("TERM")), "linux") == 0) {
|
|
|
|
|
|
fb0_.path = strdup("/dev/fb0");
|
|
|
|
|
|
}
|
|
|
|
|
|
if ((fb0_.fd = open(fb0_.path, O_RDWR)) != -1) {
|
|
|
|
|
|
CHECK_NE(-1, (rc = ioctl(fb0_.fd, FBIOGET_FSCREENINFO, &fb0_.fscreen)));
|
2021-09-04 04:14:26 +00:00
|
|
|
|
INFOF("ioctl(%s) → %d", "FBIOGET_FSCREENINFO", rc);
|
|
|
|
|
|
INFOF("%s.%s=%.*s", "fb0_.fscreen", "id", sizeof(fb0_.fscreen.id),
|
|
|
|
|
|
fb0_.fscreen.id);
|
|
|
|
|
|
INFOF("%s.%s=%p", "fb0_.fscreen", "smem_start", fb0_.fscreen.smem_start);
|
|
|
|
|
|
INFOF("%s.%s=%u", "fb0_.fscreen", "smem_len", fb0_.fscreen.smem_len);
|
|
|
|
|
|
INFOF("%s.%s=%u", "fb0_.fscreen", "type", fb0_.fscreen.type);
|
|
|
|
|
|
INFOF("%s.%s=%u", "fb0_.fscreen", "type_aux", fb0_.fscreen.type_aux);
|
|
|
|
|
|
INFOF("%s.%s=%u", "fb0_.fscreen", "visual", fb0_.fscreen.visual);
|
|
|
|
|
|
INFOF("%s.%s=%hu", "fb0_.fscreen", "xpanstep", fb0_.fscreen.xpanstep);
|
|
|
|
|
|
INFOF("%s.%s=%hu", "fb0_.fscreen", "ypanstep", fb0_.fscreen.ypanstep);
|
|
|
|
|
|
INFOF("%s.%s=%hu", "fb0_.fscreen", "ywrapstep", fb0_.fscreen.ywrapstep);
|
|
|
|
|
|
INFOF("%s.%s=%u", "fb0_.fscreen", "line_length", fb0_.fscreen.line_length);
|
|
|
|
|
|
INFOF("%s.%s=%p", "fb0_.fscreen", "mmio_start", fb0_.fscreen.mmio_start);
|
|
|
|
|
|
INFOF("%s.%s=%u", "fb0_.fscreen", "mmio_len", fb0_.fscreen.mmio_len);
|
|
|
|
|
|
INFOF("%s.%s=%u", "fb0_.fscreen", "accel", fb0_.fscreen.accel);
|
|
|
|
|
|
INFOF("%s.%s=%#b", "fb0_.fscreen", "capabilities",
|
|
|
|
|
|
fb0_.fscreen.capabilities);
|
2020-06-15 14:18:57 +00:00
|
|
|
|
CHECK_NE(-1, (rc = ioctl(fb0_.fd, FBIOGET_VSCREENINFO, &fb0_.vscreen)));
|
2021-09-04 04:14:26 +00:00
|
|
|
|
INFOF("ioctl(%s) → %d", "FBIOGET_VSCREENINFO", rc);
|
|
|
|
|
|
INFOF("%s.%s=%u", "fb0_.vscreen", "xres", fb0_.vscreen.xres);
|
|
|
|
|
|
INFOF("%s.%s=%u", "fb0_.vscreen", "yres", fb0_.vscreen.yres);
|
2022-03-18 11:20:10 +00:00
|
|
|
|
INFOF("%s.%s=%u", "fb0_.vscreen", "xres_virtual",
|
|
|
|
|
|
fb0_.vscreen.xres_virtual);
|
|
|
|
|
|
INFOF("%s.%s=%u", "fb0_.vscreen", "yres_virtual",
|
|
|
|
|
|
fb0_.vscreen.yres_virtual);
|
2021-09-04 04:14:26 +00:00
|
|
|
|
INFOF("%s.%s=%u", "fb0_.vscreen", "xoffset", fb0_.vscreen.xoffset);
|
|
|
|
|
|
INFOF("%s.%s=%u", "fb0_.vscreen", "yoffset", fb0_.vscreen.yoffset);
|
|
|
|
|
|
INFOF("%s.%s=%u", "fb0_.vscreen", "bits_per_pixel",
|
|
|
|
|
|
fb0_.vscreen.bits_per_pixel);
|
|
|
|
|
|
INFOF("%s.%s=%u", "fb0_.vscreen", "grayscale", fb0_.vscreen.grayscale);
|
|
|
|
|
|
INFOF("%s.%s=%u", "fb0_.vscreen.red", "offset", fb0_.vscreen.red.offset);
|
|
|
|
|
|
INFOF("%s.%s=%u", "fb0_.vscreen.red", "length", fb0_.vscreen.red.length);
|
|
|
|
|
|
INFOF("%s.%s=%u", "fb0_.vscreen.red", "msb_right",
|
|
|
|
|
|
fb0_.vscreen.red.msb_right);
|
2022-03-18 11:20:10 +00:00
|
|
|
|
INFOF("%s.%s=%u", "fb0_.vscreen.green", "offset",
|
|
|
|
|
|
fb0_.vscreen.green.offset);
|
|
|
|
|
|
INFOF("%s.%s=%u", "fb0_.vscreen.green", "length",
|
|
|
|
|
|
fb0_.vscreen.green.length);
|
2021-09-04 04:14:26 +00:00
|
|
|
|
INFOF("%s.%s=%u", "fb0_.vscreen.green", "msb_right",
|
|
|
|
|
|
fb0_.vscreen.green.msb_right);
|
|
|
|
|
|
INFOF("%s.%s=%u", "fb0_.vscreen.blue", "offset", fb0_.vscreen.blue.offset);
|
|
|
|
|
|
INFOF("%s.%s=%u", "fb0_.vscreen.blue", "length", fb0_.vscreen.blue.length);
|
|
|
|
|
|
INFOF("%s.%s=%u", "fb0_.vscreen.blue", "msb_right",
|
|
|
|
|
|
fb0_.vscreen.blue.msb_right);
|
|
|
|
|
|
INFOF("%s.%s=%u", "fb0_.vscreen.transp", "offset",
|
|
|
|
|
|
fb0_.vscreen.transp.offset);
|
|
|
|
|
|
INFOF("%s.%s=%u", "fb0_.vscreen.transp", "length",
|
|
|
|
|
|
fb0_.vscreen.transp.length);
|
|
|
|
|
|
INFOF("%s.%s=%u", "fb0_.vscreen.transp", "msb_right",
|
|
|
|
|
|
fb0_.vscreen.transp.msb_right);
|
|
|
|
|
|
INFOF("%s.%s=%u", "fb0_.vscreen", "nonstd", fb0_.vscreen.nonstd);
|
|
|
|
|
|
INFOF("%s.%s=%u", "fb0_.vscreen", "activate", fb0_.vscreen.activate);
|
|
|
|
|
|
INFOF("%s.%s=%u", "fb0_.vscreen", "height", fb0_.vscreen.height);
|
|
|
|
|
|
INFOF("%s.%s=%u", "fb0_.vscreen", "width", fb0_.vscreen.width);
|
|
|
|
|
|
INFOF("%s.%s=%u", "fb0_.vscreen", "accel_flags", fb0_.vscreen.accel_flags);
|
|
|
|
|
|
INFOF("%s.%s=%u", "fb0_.vscreen", "pixclock", fb0_.vscreen.pixclock);
|
|
|
|
|
|
INFOF("%s.%s=%u", "fb0_.vscreen", "left_margin", fb0_.vscreen.left_margin);
|
2022-03-18 11:20:10 +00:00
|
|
|
|
INFOF("%s.%s=%u", "fb0_.vscreen", "right_margin",
|
|
|
|
|
|
fb0_.vscreen.right_margin);
|
|
|
|
|
|
INFOF("%s.%s=%u", "fb0_.vscreen", "upper_margin",
|
|
|
|
|
|
fb0_.vscreen.upper_margin);
|
|
|
|
|
|
INFOF("%s.%s=%u", "fb0_.vscreen", "lower_margin",
|
|
|
|
|
|
fb0_.vscreen.lower_margin);
|
2021-09-04 04:14:26 +00:00
|
|
|
|
INFOF("%s.%s=%u", "fb0_.vscreen", "hsync_len", fb0_.vscreen.hsync_len);
|
|
|
|
|
|
INFOF("%s.%s=%u", "fb0_.vscreen", "vsync_len", fb0_.vscreen.vsync_len);
|
|
|
|
|
|
INFOF("%s.%s=%u", "fb0_.vscreen", "sync", fb0_.vscreen.sync);
|
|
|
|
|
|
INFOF("%s.%s=%u", "fb0_.vscreen", "vmode", fb0_.vscreen.vmode);
|
|
|
|
|
|
INFOF("%s.%s=%u", "fb0_.vscreen", "rotate", fb0_.vscreen.rotate);
|
|
|
|
|
|
INFOF("%s.%s=%u", "fb0_.vscreen", "colorspace", fb0_.vscreen.colorspace);
|
2020-06-15 14:18:57 +00:00
|
|
|
|
fb0_.size = fb0_.fscreen.smem_len;
|
|
|
|
|
|
CHECK_NE(MAP_FAILED,
|
|
|
|
|
|
(fb0_.map = mmap(NULL, fb0_.size, PROT_READ | PROT_WRITE,
|
|
|
|
|
|
MAP_SHARED, fb0_.fd, 0)));
|
|
|
|
|
|
}
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
int main(int argc, char *argv[]) {
|
|
|
|
|
|
sigset_t wut;
|
2023-06-06 16:12:30 +00:00
|
|
|
|
const char *s;
|
Make improvements
- Every unit test now passes on Apple Silicon. The final piece of this
puzzle was porting our POSIX threads cancelation support, since that
works differently on ARM64 XNU vs. AMD64. Our semaphore support on
Apple Silicon is also superior now compared to AMD64, thanks to the
grand central dispatch library which lets *NSYNC locks go faster.
- The Cosmopolitan runtime is now more stable, particularly on Windows.
To do this, thread local storage is mandatory at all runtime levels,
and the innermost packages of the C library is no longer being built
using ASAN. TLS is being bootstrapped with a 128-byte TIB during the
process startup phase, and then later on the runtime re-allocates it
either statically or dynamically to support code using _Thread_local.
fork() and execve() now do a better job cooperating with threads. We
can now check how much stack memory is left in the process or thread
when functions like kprintf() / execve() etc. call alloca(), so that
ENOMEM can be raised, reduce a buffer size, or just print a warning.
- POSIX signal emulation is now implemented the same way kernels do it
with pthread_kill() and raise(). Any thread can interrupt any other
thread, regardless of what it's doing. If it's blocked on read/write
then the killer thread will cancel its i/o operation so that EINTR can
be returned in the mark thread immediately. If it's doing a tight CPU
bound operation, then that's also interrupted by the signal delivery.
Signal delivery works now by suspending a thread and pushing context
data structures onto its stack, and redirecting its execution to a
trampoline function, which calls SetThreadContext(GetCurrentThread())
when it's done.
- We're now doing a better job managing locks and handles. On NetBSD we
now close semaphore file descriptors in forked children. Semaphores on
Windows can now be canceled immediately, which means mutexes/condition
variables will now go faster. Apple Silicon semaphores can be canceled
too. We're now using Apple's pthread_yield() funciton. Apple _nocancel
syscalls are now used on XNU when appropriate to ensure pthread_cancel
requests aren't lost. The MbedTLS library has been updated to support
POSIX thread cancelations. See tool/build/runitd.c for an example of
how it can be used for production multi-threaded tls servers. Handles
on Windows now leak less often across processes. All i/o operations on
Windows are now overlapped, which means file pointers can no longer be
inherited across dup() and fork() for the time being.
- We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4()
which means, for example, that posix_spawn() now goes 3x faster. POSIX
spawn is also now more correct. Like Musl, it's now able to report the
failure code of execve() via a pipe although our approach favors using
shared memory to do that on systems that have a true vfork() function.
- We now spawn a thread to deliver SIGALRM to threads when setitimer()
is used. This enables the most precise wakeups the OS makes possible.
- The Cosmopolitan runtime now uses less memory. On NetBSD for example,
it turned out the kernel would actually commit the PT_GNU_STACK size
which caused RSS to be 6mb for every process. Now it's down to ~4kb.
On Apple Silicon, we reduce the mandatory upstream thread size to the
smallest possible size to reduce the memory overhead of Cosmo threads.
The examples directory has a program called greenbean which can spawn
a web server on Linux with 10,000 worker threads and have the memory
usage of the process be ~77mb. The 1024 byte overhead of POSIX-style
thread-local storage is now optional; it won't be allocated until the
pthread_setspecific/getspecific functions are called. On Windows, the
threads that get spawned which are internal to the libc implementation
use reserve rather than commit memory, which shaves a few hundred kb.
- sigaltstack() is now supported on Windows, however it's currently not
able to be used to handle stack overflows, since crash signals are
still generated by WIN32. However the crash handler will still switch
to the alt stack, which is helpful in environments with tiny threads.
- Test binaries are now smaller. Many of the mandatory dependencies of
the test runner have been removed. This ensures many programs can do a
better job only linking the the thing they're testing. This caused the
test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb
- long double is no longer used in the implementation details of libc,
except in the APIs that define it. The old code that used long double
for time (instead of struct timespec) has now been thoroughly removed.
- ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing
backtraces itself, it'll just print a command you can run on the shell
using our new `cosmoaddr2line` program to view the backtrace.
- Crash report signal handling now works in a much better way. Instead
of terminating the process, it now relies on SA_RESETHAND so that the
default SIG_IGN behavior can terminate the process if necessary.
- Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
|
|
|
|
ShowCrashReports();
|
2020-06-15 14:18:57 +00:00
|
|
|
|
gamma_ = 2.4;
|
|
|
|
|
|
volscale_ -= 2;
|
|
|
|
|
|
dither_ = true;
|
|
|
|
|
|
sigemptyset(&wut);
|
|
|
|
|
|
sigaddset(&wut, SIGCHLD);
|
|
|
|
|
|
sigaddset(&wut, SIGPIPE);
|
|
|
|
|
|
sigprocmask(SIG_SETMASK, &wut, NULL);
|
2022-03-21 10:46:16 +00:00
|
|
|
|
ShowCrashReports();
|
2020-06-15 14:18:57 +00:00
|
|
|
|
fullclear_ = true;
|
|
|
|
|
|
GetOpts(argc, argv);
|
Apply clang-format update to repo (#1154)
Commit bc6c183 introduced a bunch of discrepancies between what files
look like in the repo and what clang-format says they should look like.
However, there were already a few discrepancies prior to that. Most of
these discrepancies seemed to be unintentional, but a few of them were
load-bearing (e.g., a #include that violated header ordering needing
something to have been #defined by a 'later' #include.)
I opted to take what I hope is a relatively smooth-brained approach: I
reverted the .clang-format change, ran clang-format on the whole repo,
reapplied the .clang-format change, reran clang-format again, and then
reverted the commit that contained the first run. Thus the full effect
of this PR should only be to apply the changed formatting rules to the
repo, and from skimming the results, this seems to be the case.
My work can be checked by applying the short, manual commits, and then
rerunning the command listed in the autogenerated commits (those whose
messages I have prefixed auto:) and seeing if your results agree.
It might be that the other diffs should be fixed at some point but I'm
leaving that aside for now.
fd '\.c(c|pp)?$' --print0| xargs -0 clang-format -i
2024-04-25 17:38:00 +00:00
|
|
|
|
if (!tuned_)
|
|
|
|
|
|
PickDefaults();
|
|
|
|
|
|
if (optind == argc)
|
|
|
|
|
|
PrintUsage(EX_USAGE, STDERR_FILENO);
|
2020-12-25 00:11:38 +00:00
|
|
|
|
patharg_ = argv[optind];
|
2023-06-06 16:12:30 +00:00
|
|
|
|
s = commandvenv("SOX", "sox");
|
|
|
|
|
|
sox_ = s ? strdup(s) : 0;
|
|
|
|
|
|
s = commandvenv("FFPLAY", "ffplay");
|
|
|
|
|
|
ffplay_ = s ? strdup(s) : 0;
|
|
|
|
|
|
if (!sox_ && !ffplay_) {
|
|
|
|
|
|
fprintf(stderr, "please install either the "
|
|
|
|
|
|
"`play` (sox) or "
|
|
|
|
|
|
"`ffplay` (ffmpeg) "
|
2024-03-03 00:57:56 +00:00
|
|
|
|
"commands, so printvideo can play audio\n");
|
2023-06-06 16:12:30 +00:00
|
|
|
|
usleep(10000);
|
|
|
|
|
|
}
|
2020-06-15 14:18:57 +00:00
|
|
|
|
infd_ = STDIN_FILENO;
|
|
|
|
|
|
outfd_ = STDOUT_FILENO;
|
|
|
|
|
|
if (!setjmp(jb_)) {
|
|
|
|
|
|
xsigaction(SIGINT, OnCtrlC, 0, 0, NULL);
|
|
|
|
|
|
xsigaction(SIGHUP, OnCtrlC, 0, 0, NULL);
|
|
|
|
|
|
xsigaction(SIGTERM, OnCtrlC, 0, 0, NULL);
|
|
|
|
|
|
xsigaction(SIGWINCH, OnResize, 0, 0, NULL);
|
|
|
|
|
|
xsigaction(SIGCHLD, OnSigChld, 0, 0, NULL);
|
|
|
|
|
|
xsigaction(SIGPIPE, OnSigPipe, 0, 0, NULL);
|
Apply clang-format update to repo (#1154)
Commit bc6c183 introduced a bunch of discrepancies between what files
look like in the repo and what clang-format says they should look like.
However, there were already a few discrepancies prior to that. Most of
these discrepancies seemed to be unintentional, but a few of them were
load-bearing (e.g., a #include that violated header ordering needing
something to have been #defined by a 'later' #include.)
I opted to take what I hope is a relatively smooth-brained approach: I
reverted the .clang-format change, ran clang-format on the whole repo,
reapplied the .clang-format change, reran clang-format again, and then
reverted the commit that contained the first run. Thus the full effect
of this PR should only be to apply the changed formatting rules to the
repo, and from skimming the results, this seems to be the case.
My work can be checked by applying the short, manual commits, and then
rerunning the command listed in the autogenerated commits (those whose
messages I have prefixed auto:) and seeing if your results agree.
It might be that the other diffs should be fixed at some point but I'm
leaving that aside for now.
fd '\.c(c|pp)?$' --print0| xargs -0 clang-format -i
2024-04-25 17:38:00 +00:00
|
|
|
|
if (ttyraw(kTtyLfToCrLf) != -1)
|
|
|
|
|
|
ttymode_ = true;
|
2023-11-12 06:32:12 +00:00
|
|
|
|
__cxa_atexit((void *)OnExit, NULL, NULL);
|
2021-03-01 07:42:35 +00:00
|
|
|
|
__log_file = fopen(logpath_, "a");
|
2020-06-15 14:18:57 +00:00
|
|
|
|
if (ischardev(infd_) && ischardev(outfd_)) {
|
2020-10-27 10:39:46 +00:00
|
|
|
|
/* CHECK_NE(-1, fcntl(infd_, F_SETFL, O_NONBLOCK)); */
|
2020-06-15 14:18:57 +00:00
|
|
|
|
} else if (infd_ != outfd_) {
|
|
|
|
|
|
infd_ = -1;
|
|
|
|
|
|
}
|
2020-10-27 10:39:46 +00:00
|
|
|
|
/* CHECK_NE(-1, fcntl(outfd_, F_SETFL, O_NONBLOCK)); */
|
Apply clang-format update to repo (#1154)
Commit bc6c183 introduced a bunch of discrepancies between what files
look like in the repo and what clang-format says they should look like.
However, there were already a few discrepancies prior to that. Most of
these discrepancies seemed to be unintentional, but a few of them were
load-bearing (e.g., a #include that violated header ordering needing
something to have been #defined by a 'later' #include.)
I opted to take what I hope is a relatively smooth-brained approach: I
reverted the .clang-format change, ran clang-format on the whole repo,
reapplied the .clang-format change, reran clang-format again, and then
reverted the commit that contained the first run. Thus the full effect
of this PR should only be to apply the changed formatting rules to the
repo, and from skimming the results, this seems to be the case.
My work can be checked by applying the short, manual commits, and then
rerunning the command listed in the autogenerated commits (those whose
messages I have prefixed auto:) and seeing if your results agree.
It might be that the other diffs should be fixed at some point but I'm
leaving that aside for now.
fd '\.c(c|pp)?$' --print0| xargs -0 clang-format -i
2024-04-25 17:38:00 +00:00
|
|
|
|
if (CanPlayAudio())
|
|
|
|
|
|
MakeLatencyLittleLessBad();
|
2020-06-15 14:18:57 +00:00
|
|
|
|
TryToOpenFrameBuffer();
|
|
|
|
|
|
RenounceSpecialPrivileges();
|
Apply clang-format update to repo (#1154)
Commit bc6c183 introduced a bunch of discrepancies between what files
look like in the repo and what clang-format says they should look like.
However, there were already a few discrepancies prior to that. Most of
these discrepancies seemed to be unintentional, but a few of them were
load-bearing (e.g., a #include that violated header ordering needing
something to have been #defined by a 'later' #include.)
I opted to take what I hope is a relatively smooth-brained approach: I
reverted the .clang-format change, ran clang-format on the whole repo,
reapplied the .clang-format change, reran clang-format again, and then
reverted the commit that contained the first run. Thus the full effect
of this PR should only be to apply the changed formatting rules to the
repo, and from skimming the results, this seems to be the case.
My work can be checked by applying the short, manual commits, and then
rerunning the command listed in the autogenerated commits (those whose
messages I have prefixed auto:) and seeing if your results agree.
It might be that the other diffs should be fixed at some point but I'm
leaving that aside for now.
fd '\.c(c|pp)?$' --print0| xargs -0 clang-format -i
2024-04-25 17:38:00 +00:00
|
|
|
|
if (t2 > t1)
|
|
|
|
|
|
longjmp(jb_, 1);
|
2020-06-15 14:18:57 +00:00
|
|
|
|
OpenVideo();
|
|
|
|
|
|
DimensionDisplay();
|
Make improvements
- Every unit test now passes on Apple Silicon. The final piece of this
puzzle was porting our POSIX threads cancelation support, since that
works differently on ARM64 XNU vs. AMD64. Our semaphore support on
Apple Silicon is also superior now compared to AMD64, thanks to the
grand central dispatch library which lets *NSYNC locks go faster.
- The Cosmopolitan runtime is now more stable, particularly on Windows.
To do this, thread local storage is mandatory at all runtime levels,
and the innermost packages of the C library is no longer being built
using ASAN. TLS is being bootstrapped with a 128-byte TIB during the
process startup phase, and then later on the runtime re-allocates it
either statically or dynamically to support code using _Thread_local.
fork() and execve() now do a better job cooperating with threads. We
can now check how much stack memory is left in the process or thread
when functions like kprintf() / execve() etc. call alloca(), so that
ENOMEM can be raised, reduce a buffer size, or just print a warning.
- POSIX signal emulation is now implemented the same way kernels do it
with pthread_kill() and raise(). Any thread can interrupt any other
thread, regardless of what it's doing. If it's blocked on read/write
then the killer thread will cancel its i/o operation so that EINTR can
be returned in the mark thread immediately. If it's doing a tight CPU
bound operation, then that's also interrupted by the signal delivery.
Signal delivery works now by suspending a thread and pushing context
data structures onto its stack, and redirecting its execution to a
trampoline function, which calls SetThreadContext(GetCurrentThread())
when it's done.
- We're now doing a better job managing locks and handles. On NetBSD we
now close semaphore file descriptors in forked children. Semaphores on
Windows can now be canceled immediately, which means mutexes/condition
variables will now go faster. Apple Silicon semaphores can be canceled
too. We're now using Apple's pthread_yield() funciton. Apple _nocancel
syscalls are now used on XNU when appropriate to ensure pthread_cancel
requests aren't lost. The MbedTLS library has been updated to support
POSIX thread cancelations. See tool/build/runitd.c for an example of
how it can be used for production multi-threaded tls servers. Handles
on Windows now leak less often across processes. All i/o operations on
Windows are now overlapped, which means file pointers can no longer be
inherited across dup() and fork() for the time being.
- We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4()
which means, for example, that posix_spawn() now goes 3x faster. POSIX
spawn is also now more correct. Like Musl, it's now able to report the
failure code of execve() via a pipe although our approach favors using
shared memory to do that on systems that have a true vfork() function.
- We now spawn a thread to deliver SIGALRM to threads when setitimer()
is used. This enables the most precise wakeups the OS makes possible.
- The Cosmopolitan runtime now uses less memory. On NetBSD for example,
it turned out the kernel would actually commit the PT_GNU_STACK size
which caused RSS to be 6mb for every process. Now it's down to ~4kb.
On Apple Silicon, we reduce the mandatory upstream thread size to the
smallest possible size to reduce the memory overhead of Cosmo threads.
The examples directory has a program called greenbean which can spawn
a web server on Linux with 10,000 worker threads and have the memory
usage of the process be ~77mb. The 1024 byte overhead of POSIX-style
thread-local storage is now optional; it won't be allocated until the
pthread_setspecific/getspecific functions are called. On Windows, the
threads that get spawned which are internal to the libc implementation
use reserve rather than commit memory, which shaves a few hundred kb.
- sigaltstack() is now supported on Windows, however it's currently not
able to be used to handle stack overflows, since crash signals are
still generated by WIN32. However the crash handler will still switch
to the alt stack, which is helpful in environments with tiny threads.
- Test binaries are now smaller. Many of the mandatory dependencies of
the test runner have been removed. This ensures many programs can do a
better job only linking the the thing they're testing. This caused the
test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb
- long double is no longer used in the implementation details of libc,
except in the APIs that define it. The old code that used long double
for time (instead of struct timespec) has now been thoroughly removed.
- ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing
backtraces itself, it'll just print a command you can run on the shell
using our new `cosmoaddr2line` program to view the backtrace.
- Crash report signal handling now works in a much better way. Instead
of terminating the process, it now relies on SA_RESETHAND so that the
default SIG_IGN behavior can terminate the process if necessary.
- Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
|
|
|
|
starttime_ = timespec_real();
|
2020-06-15 14:18:57 +00:00
|
|
|
|
PrintVideo();
|
|
|
|
|
|
}
|
2021-09-04 04:14:26 +00:00
|
|
|
|
INFOF("jb_ triggered");
|
2020-06-15 14:18:57 +00:00
|
|
|
|
return 0;
|
|
|
|
|
|
}
|
