Make improvements

- Introduce ualarm() function
- Make rename() report EISEMPTY on Windows
- Always raise EINVAL upon open(O_RDONLY|O_TRUNC)
- Add macro so ./configure will detect SOCK_CLOEXEC
- Fix O_TRUNC without O_CREAT not working on Windows
- Let fcntl(F_SETFL) change O_APPEND status on Windows
- Make sure pwrite() / pread() report ESPIPE on sockets
- Raise ESPIPE on Windows when pwrite() is used on pipe
- Properly compute O_APPEND CreateFile() flags on Windows
- Don't require O_DIRECTORY to open directories on Windows
- Fix more instances of Windows reporting EISDIR and ENOTDIR
- Normalize EFTYPE and EMLINK to ELOOP on NetBSD and FreeBSD
- Make unlink() / rmdir() work on read-only files on Windows
- Validate UTF-8 on Windows paths to fix bug with overlong NUL
- Always print signal name to stderr when crashing due to SIG_DFL
- Fix Windows bug where denormalized paths >260 chars didn't work
- Block signals on BSDs when thread exits before trashing its own stack

libc/str/istext.c

@@ -21,7 +21,7 @@
 /**
  * Returns true if buffer is most likely plaintext.
  */
-bool _istext(const void *data, size_t size) {
+bool istext(const void *data, size_t size) {
   const unsigned char *p, *pe;
   for (p = data, pe = p + size; p < pe; ++p) {
     if (*p <= 3) {

libc/str/isutf8.c

@@ -21,8 +21,6 @@
 #include "libc/intrin/likely.h"
 #include "libc/str/str.h"
 
-typedef char xmm_t __attribute__((__vector_size__(16), __aligned__(16)));
-
 static const char kUtf8Dispatch[] = {
     0, 0, 1, 1, 1, 1, 1, 1,  // 0300 utf8-2
     1, 1, 1, 1, 1, 1, 1, 1,  // 0310
@@ -37,10 +35,10 @@ static const char kUtf8Dispatch[] = {
 /**
  * Returns true if text is utf-8.
  *
- *     _isutf8 n=0                  1 nanoseconds
- *     _isutf8 n=5                661 ps/byte          1,476 mb/s
- *     _isutf8 ascii n=22851       26 ps/byte             35 GB/s
- *     _isutf8 unicode n=3193     543 ps/byte          1,795 mb/s
+ *     isutf8 n=0                  1 nanoseconds
+ *     isutf8 n=5                661 ps/byte          1,476 mb/s
+ *     isutf8 ascii n=22851       26 ps/byte             35 GB/s
+ *     isutf8 unicode n=3193     543 ps/byte          1,795 mb/s
  *
  * This function considers all ASCII characters including NUL to be
  * valid UTF-8. The conditions for something not being valid are:
@@ -51,7 +49,7 @@ static const char kUtf8Dispatch[] = {
  *
  * @param size if -1 implies strlen
  */
-dontasan bool _isutf8(const void *data, size_t size) {
+dontasan bool isutf8(const void *data, size_t size) {
   long c;
   unsigned m;
   const char *p, *e;
@@ -61,6 +59,7 @@ dontasan bool _isutf8(const void *data, size_t size) {
   e = p + size;
   while (p < e) {
 #if defined(__x86_64__) && !defined(__chibicc__)
+    typedef char xmm_t __attribute__((__vector_size__(16), __aligned__(16)));
     if (!((intptr_t)p & 15)) {
       for (;;) {
         if ((m = __builtin_ia32_pmovmskb128(*(xmm_t *)p >= (xmm_t){0}) ^

libc/str/str.h

@@ -11,6 +11,8 @@
 #define chomp16       _chomp16
 #define wchomp        _wchomp
 #define tpenc         _tpenc
+#define isutf8        _isutf8
+#define istext        _istext
 #define startswith    _startswith
 #define startswithi   _startswithi
 #define endswith      _endswith
@@ -187,9 +189,9 @@ wchar_t *wchomp(wchar_t *) libcesque;
 bool startswith(const char *, const char *) strlenesque;
 bool startswithi(const char *, const char *) strlenesque;
 bool endswith(const char *, const char *) strlenesque;
-bool _istext(const void *, size_t) libcesque;
-bool _isutf8(const void *, size_t) libcesque;
-char *strsignal_r(int, char[hasatleast 15]) returnsnonnull libcesque;
+bool istext(const void *, size_t) libcesque;
+bool isutf8(const void *, size_t) libcesque;
+char *strsignal_r(int, char[15]) returnsnonnull libcesque dontdiscard;
 int strerror_wr(int, uint32_t, char *, size_t)
 dontthrow nocallback;
 char16_t *chomp16(char16_t *) libcesque;