Improve mkdeps

Our makefile generator now accepts badly formatted include lines. It's
now more hermetic with better error checking in the cosmo repo, and it
can be configured to not be hermetic at all.

Makefile

@@ -329,7 +329,7 @@ o/$(MODE)/hdrs.txt: o/$(MODE)/.x $(MAKEFILES) $(call uniq,$(foreach x,$(HDRS) $(
 o/$(MODE)/incs.txt: o/$(MODE)/.x $(MAKEFILES) $(call uniq,$(foreach x,$(INCS) $(INCS),$(dir $(x)))) $(INCS) $(INCS)
 	$(file >$@,$(INCS))
 o/$(MODE)/depend: o/$(MODE)/.x o/$(MODE)/srcs.txt o/$(MODE)/hdrs.txt o/$(MODE)/incs.txt $(SRCS) $(HDRS) $(INCS)
-	$(COMPILE) -AMKDEPS -L320 $(MKDEPS) -o $@ -r o/$(MODE)/ @o/$(MODE)/srcs.txt @o/$(MODE)/hdrs.txt @o/$(MODE)/incs.txt
+	$(COMPILE) -AMKDEPS -L320 $(MKDEPS) -o $@ -s -r o/$(MODE)/ @o/$(MODE)/srcs.txt @o/$(MODE)/hdrs.txt @o/$(MODE)/incs.txt
 
 o/$(MODE)/srcs-old.txt: o/$(MODE)/.x $(MAKEFILES) $(call uniq,$(foreach x,$(SRCS),$(dir $(x))))
 	$(file >$@) $(foreach x,$(SRCS),$(file >>$@,$(x)))

bin/fatcosmocc

@@ -240,6 +240,8 @@ NEED_JOIN=
 NEED_EQUAL=
 NEED_OUTPUT=
 APELINKFLAGS=
+FLAGS_X86_64=
+FLAGS_AARCH64=
 INPUT_FILE_COUNT=0
 for x; do
   if [ x"$x" != x"${x#* }" ]; then
@@ -335,6 +337,14 @@ for x; do
     continue
   elif [ x"$x" = x"-march=native" ]; then
     fatal_error "-march=native can't be used when building fat binaries"
+  elif [ x"$x" != x"${x#-Xx86_64}" ]; then
+    x=${x#-Xx86_64}  # e.g. -Xx86_64,-msse3,-mavx,-mavx2,-mf16c,-mfma
+    FLAGS_X86_64="$FLAGS_X86_64 ${x//,/ }"
+    continue
+  elif [ x"$x" != x"${x#-Xaarch64}" ]; then
+    x=${x#-Xaarch64}
+    FLAGS_aarch64="$FLAGS_aarch64 ${x//,/ }"
+    continue
   elif [ x"$x" = x"-dumpversion" ]; then
     echo $GCC_VERSION
     Exit 0
@@ -473,6 +483,7 @@ build_object() {
         $CFLAGS_X86_64 \
         $CPPFLAGS_X86_64 \
         "$@" \
+        $FLAGS_X86_64 \
         $PRECIOUS
     log_command "$@"
     "$@" || exit
@@ -492,6 +503,7 @@ build_object() {
         $CFLAGS_AARCH64 \
         $CPPFLAGS_AARCH64 \
         "$@" \
+        $FLAGS_AARCH64 \
         $PRECIOUS &&
     log_command "$@"
     "$@" || exit

libc/calls/getrlimit.c

@@ -23,6 +23,7 @@
 #include "libc/intrin/asan.internal.h"
 #include "libc/intrin/describeflags.internal.h"
 #include "libc/intrin/strace.internal.h"
+#include "libc/runtime/runtime.h"
 #include "libc/runtime/stack.h"
 #include "libc/runtime/syslib.internal.h"
 #include "libc/sysv/consts/rlimit.h"

libc/calls/internal.h

@@ -31,10 +31,6 @@ int64_t GetConsoleOutputHandle(void);
 int IsWindowsExecutable(int64_t, const char16_t *);
 void InterceptTerminalCommands(const char *, size_t);
 
-forceinline int64_t __getfdhandleactual(int fd) {
-  return g_fds.p[fd].handle;
-}
-
 forceinline bool __isfdopen(int fd) {
   return 0 <= fd && fd < g_fds.n && g_fds.p[fd].kind != kFdEmpty;
 }

libc/calls/ioctl.c

@@ -75,7 +75,7 @@ static int ioctl_default(int fd, unsigned long request, void *arg) {
     return sys_ioctl(fd, request, arg);
   } else if (__isfdopen(fd)) {
     if (g_fds.p[fd].kind == kFdSocket) {
-      handle = __getfdhandleactual(fd);
+      handle = g_fds.p[fd].handle;
       if ((rc = _weaken(__sys_ioctlsocket_nt)(handle, request, arg)) != -1) {
         return rc;
       } else {

libc/calls/ischardev.c

@@ -56,6 +56,6 @@ bool32 ischardev(int fd) {
   } else {
     return __isfdkind(fd, kFdConsole) || __isfdkind(fd, kFdDevNull) ||
            (__isfdkind(fd, kFdFile) &&
-            GetFileType(__getfdhandleactual(fd)) == kNtFileTypeChar);
+            GetFileType(g_fds.p[fd].handle) == kNtFileTypeChar);
   }
 }

libc/calls/pthread_yield.c

@@ -18,7 +18,6 @@
 ╚─────────────────────────────────────────────────────────────────────────────*/
 #include "libc/calls/calls.h"
 #include "libc/dce.h"
-#include "libc/nexgen32e/yield.h"
 #include "libc/runtime/runtime.h"
 #include "libc/runtime/syslib.internal.h"
 #include "libc/thread/thread.h"
@@ -32,7 +31,7 @@ int pthread_yield(void) {
   if (IsXnuSilicon()) {
     __syslib->__pthread_yield_np();
   } else if (IsOpenbsd()) {
-    spin_yield();  // sched_yield() is punishingly slow on OpenBSD
+    pthread_pause_np();  // sched_yield() is punishingly slow on OpenBSD
   } else {
     sched_yield();
   }

libc/calls/read-nt.c

@@ -142,7 +142,7 @@ struct Keystrokes {
 
 static struct Keystrokes __keystroke;
 
-textwindows void __keystroke_wipe(void) {
+textwindows void WipeKeystrokes(void) {
   bzero(&__keystroke, sizeof(__keystroke));
 }
 
@@ -754,8 +754,8 @@ static textwindows ssize_t ReadFromConsole(struct Fd *f, void *data,
   return rc;
 }
 
-textwindows ssize_t sys_read_nt_impl(int fd, void *data, size_t size,
-                                     int64_t offset, sigset_t waitmask) {
+textwindows ssize_t ReadBuffer(int fd, void *data, size_t size, int64_t offset,
+                               sigset_t waitmask) {
 
   // switch to terminal polyfill if reading from win32 console
   struct Fd *f = g_fds.p + fd;
@@ -786,18 +786,18 @@ textwindows ssize_t sys_read_nt_impl(int fd, void *data, size_t size,
   }
 }
 
-static textwindows ssize_t sys_read_nt2(int fd, const struct iovec *iov,
-                                        size_t iovlen, int64_t opt_offset,
-                                        sigset_t waitmask) {
+static textwindows ssize_t ReadIovecs(int fd, const struct iovec *iov,
+                                      size_t iovlen, int64_t opt_offset,
+                                      sigset_t waitmask) {
   ssize_t rc;
   size_t i, total;
   if (opt_offset < -1) return einval();
   while (iovlen && !iov[0].iov_len) iov++, iovlen--;
   if (iovlen) {
     for (total = i = 0; i < iovlen; ++i) {
       if (!iov[i].iov_len) continue;
-      rc = sys_read_nt_impl(fd, iov[i].iov_base, iov[i].iov_len, opt_offset,
-                            waitmask);
+      rc =
+          ReadBuffer(fd, iov[i].iov_base, iov[i].iov_len, opt_offset, waitmask);
       if (rc == -1) {
         if (total && errno != ECANCELED) {
           return total;
@@ -811,15 +811,15 @@ static textwindows ssize_t sys_read_nt2(int fd, const struct iovec *iov,
     }
     return total;
   } else {
-    return sys_read_nt_impl(fd, NULL, 0, opt_offset, waitmask);
+    return ReadBuffer(fd, NULL, 0, opt_offset, waitmask);
   }
 }
 
 textwindows ssize_t sys_read_nt(int fd, const struct iovec *iov, size_t iovlen,
                                 int64_t opt_offset) {
   ssize_t rc;
   sigset_t m = __sig_block();
-  rc = sys_read_nt2(fd, iov, iovlen, opt_offset, m);
+  rc = ReadIovecs(fd, iov, iovlen, opt_offset, m);
   __sig_unblock(m);
   return rc;
 }

libc/calls/setrlimit.c

@@ -25,6 +25,7 @@
 #include "libc/intrin/describeflags.internal.h"
 #include "libc/intrin/strace.internal.h"
 #include "libc/macros.internal.h"
+#include "libc/runtime/runtime.h"
 #include "libc/runtime/syslib.internal.h"
 #include "libc/sysv/consts/rlimit.h"
 #include "libc/sysv/errfuns.h"

libc/calls/sig.c

@@ -201,7 +201,6 @@ textwindows int __sig_raise(volatile int sig, int sic) {
     STRACE("__sig_raise(%G, %t) mask %s", sig, __sig_handler(rva),
            (DescribeSigset)(ssbuf, 0, (sigset_t *)&pt->tib->tib_sigmask));
     __sig_handler(rva)(sig, &si, &ctx);
-    (void)ssbuf;
 
     // record this handler
     if (flags & SA_RESTART) {
@@ -271,7 +270,6 @@ static textwindows wontreturn void __sig_tramp(struct SignalFrame *sf) {
            (DescribeSigset)(ssbuf[0], 0, &sf->ctx.uc_sigmask),
            (DescribeSigset)(ssbuf[1], 0, (sigset_t *)&tib->tib_sigmask));
     __sig_handler(sf->rva)(sig, &sf->si, &sf->ctx);
-    (void)ssbuf;
 
     // restore the signal mask that was used by the interrupted code
     // this may have been modified by the signal handler in the callback

libc/calls/sigtimedwait.c

@@ -51,7 +51,6 @@ int sigtimedwait(const sigset_t *set, siginfo_t *info,
   struct timespec ts;
   union siginfo_meta si = {0};
   BEGIN_CANCELATION_POINT;
-  (void)strsig;
 
   if (IsAsan() && (!__asan_is_valid(set, sizeof(*set)) ||
                    (info && !__asan_is_valid(info, sizeof(*info))) ||

libc/calls/sync.c

@@ -21,6 +21,7 @@
 #include "libc/calls/syscall-sysv.internal.h"
 #include "libc/dce.h"
 #include "libc/intrin/strace.internal.h"
+#include "libc/runtime/runtime.h"
 
 /**
  * Flushes file system changes to disk by any means necessary.

libc/calls/sys_ptrace.c

@@ -22,6 +22,7 @@
 #include "libc/intrin/describeflags.internal.h"
 #include "libc/intrin/likely.h"
 #include "libc/intrin/strace.internal.h"
+#include "libc/runtime/runtime.h"
 
 #define IsPeek(request) (IsLinux() && (request)-1u < 3)
 

libc/calls/winexec.c

@@ -28,9 +28,10 @@
 #include "libc/str/tab.internal.h"
 #include "third_party/linenoise/linenoise.h"
 
-#define EXT(s) READ32LE(s "\0\0")
+#define Read32(s) (s[3] << 24 | s[2] << 16 | s[1] << 8 | s[0])
+#define EXT(s)    Read32(s "\0\0")
 
-static bool IsGraph(wint_t c) {
+static inline bool IsGraph(wint_t c) {
   return 0x21 <= c && c <= 0x7E;
 }
 

libc/intrin/__getenv.c

@@ -18,6 +18,7 @@
 ╚─────────────────────────────────────────────────────────────────────────────*/
 #include "libc/dce.h"
 #include "libc/intrin/getenv.internal.h"
+#include "libc/intrin/kprintf.h"
 
 privileged struct Env __getenv(char **p, const char *k) {
   char *t;

libc/intrin/createfile.c

@@ -49,13 +49,9 @@ CreateFile(const char16_t *lpFileName,                         //
   int64_t hHandle;
   uint32_t micros = 1;
   char buf_accessflags[512];
-  (void)buf_accessflags;
   char buf_shareflags[64];
-  (void)buf_shareflags;
   char buf_secattr[32];
-  (void)buf_secattr;
   char buf_flagattr[256];
-  (void)buf_flagattr;
 TryAgain:
   hHandle = __imp_CreateFileW(lpFileName, dwDesiredAccess, dwShareMode,
                               opt_lpSecurity, dwCreationDisposition,

libc/intrin/cxalock.c

@@ -33,7 +33,10 @@ void __cxa_unlock(void) {
   pthread_mutex_unlock(&__cxa_lock_obj);
 }
 
-__attribute__((__constructor__)) static void __cxa_init(void) {
-  __cxa_wipe();
+static textstartup void __cxa_init() {
   pthread_atfork(__cxa_lock, __cxa_unlock, __cxa_wipe);
 }
+
+const void *const __cxa_ctor[] initarray = {
+    __cxa_init,
+};

libc/intrin/ftrace_enabled.c

@@ -25,7 +25,7 @@
  * @param delta is added to enabled state
  * @return enabled state before `delta` was applied
  */
-dontinstrument int ftrace_enabled(int delta) {
+dontinstrument int(ftrace_enabled)(int delta) {
   int res;
   struct CosmoTib *tib;
   if (__tls_enabled) {

libc/intrin/isdebuggerpresent.c

@@ -50,7 +50,7 @@ int IsDebuggerPresent(bool force) {
   int e, fd, res;
   char *p, buf[1024];
   if (!force && IsGenuineBlink()) return 0;
-  if (!force && __getenv(environ, "HEISENDEBUG").s) return 0;
+  if (!force && environ && __getenv(environ, "HEISENDEBUG").s) return 0;
   if (IsWindows()) return IsBeingDebugged();
   if (__isworker) return false;
   if (!PLEDGED(RPATH)) return false;

libc/intrin/kprintf.greg.c

@@ -345,7 +345,7 @@ privileged long kloghandle(void) {
         // it's too early in the initialization process for kprintf
         return -1;
       }
-      path = __getenv(__envp, "KPRINTF_LOG").s;
+      path = environ ? __getenv(environ, "KPRINTF_LOG").s : 0;
       closefd = false;
       if (!path) {
         fd = STDERR_FILENO;

libc/intrin/pthread_mutex_lock.c

@@ -22,7 +22,6 @@
 #include "libc/intrin/atomic.h"
 #include "libc/intrin/strace.internal.h"
 #include "libc/intrin/weaken.h"
-#include "libc/nexgen32e/yield.h"
 #include "libc/runtime/internal.h"
 #include "libc/thread/thread.h"
 #include "libc/thread/tls.h"
@@ -83,7 +82,7 @@ errno_t pthread_mutex_lock(pthread_mutex_t *mutex) {
 
   if (mutex->_type == PTHREAD_MUTEX_NORMAL) {
     while (atomic_exchange_explicit(&mutex->_lock, 1, memory_order_acquire)) {
-      spin_yield();
+      pthread_pause_np();
     }
     return 0;
   }
@@ -103,7 +102,7 @@ errno_t pthread_mutex_lock(pthread_mutex_t *mutex) {
   }
 
   while (atomic_exchange_explicit(&mutex->_lock, 1, memory_order_acquire)) {
-    spin_yield();
+    pthread_pause_np();
   }
 
   mutex->_depth = 0;

libc/intrin/pthread_next.c

@@ -19,15 +19,15 @@
 #include "libc/assert.h"
 #include "libc/dce.h"
 #include "libc/intrin/atomic.h"
-#include "libc/nexgen32e/yield.h"
 #include "libc/thread/posixthread.internal.h"
+#include "libc/thread/thread.h"
 
 intptr_t _pthread_syshand(struct PosixThread *pt) {
   intptr_t syshand;
   unassert(IsWindows() || IsXnuSilicon());
   for (;;) {
     syshand = atomic_load_explicit(&pt->tib->tib_syshand, memory_order_acquire);
     if (syshand) return syshand;
-    spin_yield();
+    pthread_pause_np();
   }
 }

libc/intrin/pthread_tid.c

@@ -17,13 +17,13 @@
 │ PERFORMANCE OF THIS SOFTWARE.                                                │
 ╚─────────────────────────────────────────────────────────────────────────────*/
 #include "libc/intrin/atomic.h"
-#include "libc/nexgen32e/yield.h"
 #include "libc/thread/posixthread.internal.h"
+#include "libc/thread/thread.h"
 
 int _pthread_tid(struct PosixThread *pt) {
   int tid = 0;
   while (pt && !(tid = atomic_load_explicit(&pt->ptid, memory_order_acquire))) {
-    spin_yield();
+    pthread_pause_np();
   }
   return tid;
 }