Re: Python issues, mostly

[Taylor R Campbell <campbell+netbsd-tech-pkg%mumble.net@localhost>]

> Date: Tue, 17 Sep 2024 15:59:52 +0000 (UTC)
> From: John Klos <john%klos.com@localhost>
> 
> Also, python310 still hangs on non-interactive build.

Are you running in the background with stdin as a tty but with stdout
to a log file?  Is this reproducible if you run with `</dev/null',
i.e., with stdin not being a tty?

If the answers are `yes' and `no', respectively, then this is probably
still the unfixed PR pkg/57886, <https://gnats.NetBSD.org/57886>
(please put followups on the PR so it's easier to track):

   >Synopsis:       lang/pythonN build suspends itself with SIGTTOU
   >Description:
   Now that readline is unconditionally included in the lang/pythonN packages, Python will unconditionally do silly things when stdin is a tty but stdout is not (e.g., redirect output to a file but don't redirect input, in a background job) -- under these circumstances it decides to initialize readline, and initializing readline makes it do TIOCGWINSZ, which is fine, and then TIOCSWINSZ, which suspends the process with SIGTTOU.

   static int
   get_tty_settings (int tty, TIOTYPE *tiop)
   {
     set_winsize (tty);
   ...
   }

   /* Dummy call to force a backgrounded readline to stop before it tries
      to get the tty settings. */
   static void
   set_winsize (tty)
        int tty;
   {
   #if defined (TIOCGWINSZ)
     struct winsize w;

     if (ioctl (tty, TIOCGWINSZ, &w) == 0)
         (void) ioctl (tty, TIOCSWINSZ, &w);
   #endif /* TIOCGWINSZ */
   }

   See https://mail-index.netbsd.org/tech-pkg/2024/01/03/msg028751.html and the thread leading into that for more detail.
   >How-To-Repeat:
   build python or things that use python at build-time in a background job with output redirected to a file and input from the tty
   >Fix:
   Workaround: run the build with input redirected from /dev/null.