There were a few errors in how capacity and memory was being handled for
small strings. The capacity errors meant that small strings would become
big strings too soon, and the memory error introduced undefined behavior
that was caught by CheckMemoryLeaks in our test file but only sometimes.
The crucial change is in reserve: we only copy n bytes into p2, and then
we manually set the null terminator instead of expecting it to have been
there already. (E.g. it might not be there for an empty small string.)
We also fix one other doozy in append when we were exactly at the small-
to-big string boundary: we set the last byte (i.e., the remainder field)
to 0, then decremented it, giving us size_t max. Whoops. We boneheadedly
fix this by setting the 0 byte after we've fixed up the remainder, so it
is at worst a no-op.
Otherwise, capacity now works the same for small strings as it does with
big strings: it's the amount of space available including the null byte.
We test all of this with a new test that only gets included if our class
under test is not std::string (presumably meaning it's ctl::string.) The
test manually verifies that the small string optimization behaves how we
expect.
Since this test checks against std::string, we go ahead and include that
other header from the STL.
Also modifies the new test we introduced to also run on std::string, but
it just does the append without expecting anything about how its data is
stored. We also check that the string has the right value afterwards.
A small-string optimization is a way of reusing inline storage space for
sufficiently small strings, rather than allocating them on the heap. The
current approach takes after an old Facebook string class: it reuses the
highest-order byte for flags and small-string size, in such a way that a
maximally-sized small string will have its last byte zeroed, making it a
null terminator for the C string.
The only flag we have is in the highest-order bit, that says whether the
string is big (set) or small (cleared.) Most of the logic switches based
on the value of this bit; e.g. data() returns big()->p if it's set, else
small()->buf if it's cleared. For a small string, the capacity is always
fixed at sizeof(string) - 1 bytes; we store the length in the last byte,
but we store it as the number of remaining bytes of capacity, so that at
max size, the last byte will read zero and serve as our null terminator.
Morally speaking, our class's storage is a union over two POD C structs.
For now I gravitated towards a slightly more obtuse approach: the string
class itself contains a blob of the right size, and we alias that blob's
pointer for the two structs, taking some care not to run afoul of object
lifetime rules in C++. If anyone wants to improve on this, contributions
are welcome.
This commit also introduces the `ctl::__` namespace. It can't be legally
spelled by library users, and serves as our version of boost's "detail".
We introduced a string::swap function, and we now use that in operator=.
operator= now takes its argument by value, so we never need to check for
the case where the pointers are equal and can just swap the entire store
of the argument with our own, leaving the C++ destructor to free our old
storage afterwards.
There are probably still a few places where our capacity is slightly off
and we grow too fast, although there don't appear to be any where we are
too slow. I will leave these to be fixed in future changes.
If pthread_create() is linked into the binary, then the cosmo runtime
will create an independent dlmalloc arena for each core. Whenever the
malloc() function is used it will index `g_heaps[sched_getcpu() / 2]`
to find the arena with the greatest hyperthread / numa locality. This
may be configured via an environment variable. For example if you say
`export COSMOPOLITAN_HEAP_COUNT=1` then you can restore the old ways.
Your process may be configured to have anywhere between 1 - 128 heaps
We need this revision because it makes multithreaded C++ applications
faster. For example, an HTTP server I'm working on that makes extreme
use of the STL went from 16k to 2000k requests per second, after this
change was made. To understand why, try out the malloc_test benchmark
which calls malloc() + realloc() in a loop across many threads, which
sees a a 250x improvement in process clock time and 200x on wall time
The tradeoff is this adds ~25ns of latency to individual malloc calls
compared to MODE=tiny, once the cosmo runtime has transitioned into a
fully multi-threaded state. If you don't need malloc() to be scalable
then cosmo provides many options for you. For starters the heap count
variable above can be set to put the process back in single heap mode
plus you can go even faster still, if you include tinymalloc.inc like
many of the programs in tool/build/.. are already doing since that'll
shave tens of kb off your binary footprint too. Theres also MODE=tiny
which is configured to use just 1 plain old dlmalloc arena by default
Another tradeoff is we need more memory now (except in MODE=tiny), to
track the provenance of memory allocation. This is so allocations can
be freely shared across threads, and because OSes can reschedule code
to different CPUs at any time.
It hasn't been maintained in years. I'm tired of the root level of our
project having an advertisement for Microsoft Visual Studio Code. Your
preferred editor should be Emacs or Vim.
Cosmo will now print C++ symbols correctly in --ftrace logs and
backtraces. Doing this required reducing the memory requirement
of the __demangle() function by 3x. This was accomplished using
16-bit indices and 16-bit malloc granularity. That puts a limit
on the longest symbol we can successfully decode, which I think
would be around 6553 characters long, given a 65536-byte buffer
If you follow the directions in that file then git blame will ignore the
listed commits. A commit should only go in that file if its only changes
were to formatting, particularly on a large part of the codebase (like a
change to .clang-format getting applied to the repo.)
Cribbed from here:
https://www.stefanjudis.com/today-i-learned/how-to-exclude-commits-from-git-blame/
The V8 behavior of encoding infinity as null doesn't make sense to me.
Using ±1e5000 is better, because JSON.parse decodes it as INFINITY and
the information is preserved. This could be a breaking change for some
We're now able to rewind the instruction pointer in x86 backtraces. This
helps ensure addr2line cannot print information about unrelated adjacent
code. I've restored -fno-schedule-insns2 in most cases because it really
does cause unpredictable breakage for backtraces.
This change fixes a bug where exiting a crash signal handler on Windows
after adding the signal to uc_sigmask, but not correcting the CPU state
would cause the signal handler to loop infinitely, causing process hang
Another issue is that very tiny programs, that don't link posix signals
would not have their SIGILL / SIGSEGV / etc. status reported to Cosmo's
bash shell when terminating on crash. That's fixed by a tiny handler in
WinMain() that knows how to map WIN32 crash codes to the POSIX flavors.
Microsoft caused some very gentle breakages for Cosmopolitan. They
removed the version information from the PEB which caused uname to
report WINDOWS 0.0.0. We should have called GetVersionExW but that
doesn't really exist anymore either. Windows policy is now to give
whatever version we used in ape/ape.S. Windows8 has been EOL since
2023-01-10 so lets avoid our modern executables being relegated to
legacy infrastructure. Requiring Windows 10+ going forward lets us
remove runtime compatibility bloat from the codebase. Further note
Cosmopolitan maintains a Windows Vista branch on GitHub, so anyone
preferring the older versions, can still have a future with Cosmo.
Another neat thing this fixes is UTF-8 support in the console. The
changes Microsoft made broke the if statement that enabled UTF8 in
terminals. This explains why bug reports had broken arrows. In the
future this should be less of an issue, since the PEB code is gone
which means we more strictly conform to only Microsoft's WIN32 API
Cosmo's _Cz_crc32() function now goes 73 GiB/s on Threadripper. This
will significantly improve the performance of the PKZIP file format.
This algorithm is also used by apelink, to create deterministic ids.
The normal getopt() function is bloated because it links printf(). This
change exports the original authentic bsd getopt function, that cosmo's
always used internally so cosmocc users don't need to include internals
This change adds a TLS freelist for small dynamic memory allocations.
Cosmopolitan's TIB is now 512 bytes in size. Single-threaded malloc()
performance isn't impacted by this, until pthread_create() is called.
Single-threaded programs may also want to consider using:
#include "libc/mem/tinymalloc.inc"
Which will shave 30k off the executable size and sometimes go faster.
