Various assorted improvements to spawnveg

[JohnoKing]

Alterations:

• Bugfix: Don't set the terminal signals to their defaults in the parent prior to calling spawnveg. This is the primary cause of a lockup in the pty tests occurring on Linux distributions using glibc 2.35+. The correct way to go about dealing with SIGT* is to set those to SIG_DFL in the child process; see _sh_fork() for a working example that doesn't lock up or produce segfaults. In this commit I duplicate _sh_fork's behavior in spawnveg for the fork fallback and with POSIX_SPAWN_SETSIGDEF for the posix_spawn version. The lockup can be reproduced as follows:

$ exec ksh
$ bin/shtests pty 2>/dev/null
^C ^C ^C
# (SIGINT doesn't work anymore and may segfault)

• Added support for tcsetpgrp to the fork fallback in spawnveg. Some form of this appears to have already been attempted in AT&T olden times, but that old version was broken and needed bugfixes desperately.
  • If the child needs tcsetpgrp, block the terminal signals in the parent process via sigcritical(). The posix_spawn version doesn't need this because posix_spawn will usually block signals automatically and therefore doesn't need sigcritical.
• Now that the fork fallback for spawnveg works correctly in interactive terminals, prefer that to the sh_fork() codepath on operating systems without posix_spawn_file_actions_addtcsetpgrp_np. Even though the underlying system call is still ultimately fork, the sh_ntfork() codepath is faster than the traditional sh_fork() codepath. Benchmark (FreeBSD 14.2):

$ time /tmp/ksh-devbranch -ic "for i in {1..10000}; do $(whence -p true); done"
real       0m03.302s
user       0m00.988s
sys        0m02.320s
$ time /tmp/ksh-newspawnveg -ic "for i in {1..10000}; do $(whence -p true); done"
real       0m03.160s
user       0m01.187s
sys        0m01.977s

• To that end, split up the spawnveg versions into spawnveg_fast and spawnveg_slow. Choose the appropriate one when spawnveg is called; this removes the need for the xec.c ifdef hackery.
  • Removed the ntfork_tcpgrp ifdefs from xec.c; spawnveg can handle it by itself now.
• Bugfix: With the spawnveg_fast and spawnveg_slow innovation, spawnveg now always has support for setsid. It'll fallback to fork if POSIX_SPAWN_SETSID isn't available.
• Bugfix: For the posix_spawn version of spawnveg, the flags should be of the short type pursuant to the POSIX specification.
• Optimization: Use pipe2 in the fork fallback for spawnveg when it's available to avoid two fcntl syscalls.
• Updated the spawnveg documentation to reflect the new changes.

[JohnoKing]

This pull request has been superseded by #888.