macOS: work around freeze: don't suspend 'sleep' on interactive

Steps to reproduce:

 1. sleep 3
 2. press Ctrl+Z before it finishes (sleep builtin is forked into
    its own process and suspended)
 3. fg

sleep should resume and terminate, but instead, the forked sleep
builtin freezes, consuming 100% CPU. Only SIGKILL (kill -9) will
terminate it, although it can still be suspended again.

This is reproducible on (at least) macOS 12.7.6 arm64. I can also
reproduce this on 93u+ 2012-08-01 and on ksh2020.

It looks like this is a bug in macOS. When I compile ksh with ASan,
and the forked/resumed ksh sleep process hangs, and I simulate a
segfault by doing kill -s SEGV (the PID of the hanging process),
then the stack trace shows that it was hanging in the nanosleep
function of libsystem_c.dylib.

src/cmd/ksh93/bltins/sleep.c: sh_delay():
- Make it impossible to suspend the sleep builtin on macOS by
  ignoring SIGTSTP. We only need to do that if we've invoked
  'sleep' from the interactive prompt, i.e., if the SH_INTERACTIVE
  state bit is active. (See 48ba696, d11d4c7)

Resolves: #814

Author: McDutchie

src/cmd/ksh93/bltins/sleep.c

@@ -159,10 +159,23 @@ void sh_delay(double t, int sflag)
 	n = (uint32_t)t;
 	ts.tv_sec = n;
 	ts.tv_nsec = 1000000000 * (t - (double)n);
+#if __APPLE__ && __MACH__
+	/*
+	 * Bug in macOS: if sleep is invoked from the interactive command line and then suspended
+	 * (^Z), the forked ksh process freezes in the nanosleep(2) function in libsystem_c.dylib.
+	 * As a workaround, make it impossible to suspend sleep in that case, by ignoring SIGTSTP.
+	 */
+	if (sh_isstate(SH_INTERACTIVE))
+		signal(SIGTSTP,SIG_IGN);
+#endif
 	while(tvsleep(&ts, &tx) < 0)
 	{
 		if ((sh.trapnote & (SH_SIGSET | SH_SIGTRAP)) || sflag)
-			return;
+			break;
 		ts = tx;
 	}
+#if __APPLE__ && __MACH__
+	if (sh_isstate(SH_INTERACTIVE))
+		signal(SIGTSTP,SIG_DFL);
+#endif
 }

src/cmd/ksh93/include/version.h

@@ -18,7 +18,7 @@
 #include <ast_release.h>
 #include "git.h"
 
-#define SH_RELEASE_DATE	"2025-03-21"	/* must be in this format for $((.sh.version)) */
+#define SH_RELEASE_DATE	"2025-03-22"	/* must be in this format for $((.sh.version)) */
 /*
  * This comment keeps SH_RELEASE_DATE a few lines away from SH_RELEASE_SVER to avoid
  * merge conflicts when cherry-picking dev branch commits onto a release branch.