Re: [ANNOUNCE] Git v2.32.0-rc3 - t5300 Still Broken on NonStop ia64/x86

From: Jeff King <peff@peff.net>
Date: Jun 2, 2021, 19:49 UTC
Message-ID: <YLfgy94sbmStC0mR@coredump.intra.peff.net>
In-Reply-To: <YLfc2+Te7Y3UY+Sm@nand.local>

On Wed, Jun 02, 2021 at 03:32:43PM -0400, Taylor Blau wrote:

Show 7 quoted lines

> On Wed, Jun 02, 2021 at 01:52:44PM -0400, Randall S. Becker wrote:
> > I'm happy to help figure this out but need some direction. I don't
> > know the pack-object code.
> 
> Is the failure consistent, i.e., that it occurs every time you run the
> test? Not knowing much about your platform, it would be helpful to have
> a bisection showing where this breakage first occurs.

I suspect the symptom comes from the test Randall noted, but that the actual issue has been there all along. The test uses "--progress" explicitly, so we'll be sending SIGALRM (whereas most tests will disable the progress mechanism because their output isn't going to a tty).

And so when he gets this error:

  fatal: fsync error on '.git/objects/pack/tmp_pack_NkPgqN': Interrupted system call

presumably we were in fsync() when the signal arrived, and unlike most other platforms, the call needs to be restarted manually (even though we set up the signal with SA_RESTART). I'm not sure if this violates POSIX or not (I couldn't find a definitive answer to the set of interruptible functions in the standard). But either way, the workaround is probably something like:

  #ifdef FSYNC_NEEDS_RESTART
  #undef fsync /* we'd define to git_fsync() in a header file */
  static int git_fsync(int fd)
  {
	int ret;
	while ((ret = fsync(fd)) < 0 && errno == EINTR)
		; /* try again */
	return ret;
  }
  #endif

-Peff