Re: Is t7006-pager.sh racy?

From: Jeff King <peff@peff.net>
Date: Nov 21, 2021, 18:40 UTC
Message-ID: <YZqSgu4XjPWnURju@coredump.intra.peff.net>
In-Reply-To: <20211024170349.GA2101@szeder.dev>

On Sun, Oct 24, 2021 at 07:03:49PM +0200, SZEDER Gábor wrote:

Show 5 quoted lines

> > What makes us expect that the "git log" invocation should trigger a
> > SIGPIPE in the first place?
> 
> A misunderstanding, perhaps, because those 'git log' commands with
> their early-exiting pagers rarely trigger SIGPIPE.

I happened to be looking in this area today[1], and I think it turns out not to be "rarely", but rather "never" for some of the tests.

The test in question sets the pager to "does-not-exist". But in that case we will realize immediately via run-command.c that we could not run the pager, and will not even redirect our stdout to it.

For example, doing this:

  GIT_PAGER=does-not-exist git -c alias.foo='!yes' -p foo

will never get SIGPIPE; it will just write infinitely to the original stdout, and return success.

Whereas this:

  GIT_PAGER=false git -c alias.foo='!yes' -p foo

will reliably get SIGPIPE. But even if we used it (with a while loop to instead of "yes" address the portability concern), the tests in t7006 would still be wrong, because they are sending test-terminal's output to a closed pipe (so we'd still see SIGPIPE regardless of Git's behavior). They should be sending test_terminal's output to a file or /dev/null.

It seems like this thread stalled. Ævar, were you planning to fix these tests?

It's not too hard to swap out a "yes" as I showed above, but it further confuses the trace2 output, because now we have the child yes (or its shell equivalent) exiting as well.

The non-child "log --stdin" example I gave earlier in the thread avoids that, but the fifo hackery is so horrible that I'd just as soon avoid it. I guess yet another option is a builtin which produces infinite output. Perhaps:

  oid=$(git rev-parse HEAD)
  while true; do echo $oid; done |
  test_terminal git -p cat-file --batch-check

That's guaranteed to get SIGPIPE eventually if the pager stops reading. I seem to recall that test_terminal's handling of stdin is somewhat broken, though, and would probably get in our way[2]. Possibly we could just rip it out, as nobody is relying on it (they can't be, because it's broken).

-Peff

[1] https://lore.kernel.org/git/YZqSBlvzz2KgOMnJ@coredump.intra.peff.net/

[2] https://lore.kernel.org/git/20190520125016.GA13474@sigill.intra.peff.net/