From: Jeff King <peff@peff.net>
To: Clemens Buchacher <clemens.buchacher@intel.com>
Cc: git@vger.kernel.org, junio@pobox.com
Subject: Re: [PATCH] allow hooks to ignore their standard input stream
Date: Fri, 13 Nov 2015 01:17:29 -0500 [thread overview]
Message-ID: <20151113061729.GC32157@sigill.intra.peff.net> (raw)
In-Reply-To: <20151111144222.GA24717@musxeris015.imu.intel.com>
On Wed, Nov 11, 2015 at 03:42:22PM +0100, Clemens Buchacher wrote:
> On Wed, Nov 11, 2015 at 03:39:20PM +0100, Clemens Buchacher wrote:
> > + if (write_in_full(proc.in, buf.buf, buf.len) < 0) {
> > + /* We do not mind if a hook does not read all refs. */
> > + if (errno != EPIPE)
> > + ret = -1;
>
> I can reproduce the pipe error reliably with the test below. I did not
> include it in the patch since I am in doubt if we should add an optional
> sleep to the code.
Yeah, we definitely do not want to do so. The reliable way to get a
SIGPIPE is:
1. Have one process write more than a pipe buffer's worth of data.
2. Have the other process exit without reading anything.
Then even if the writer process goes first, it will block, and
eventually get EPIPE when the other side closes the descriptor.
Unfortunately, the pipe buffer size varies from system to system. It's
128K by default on modern Linux platforms. Using that is probably
enough. The worst case is that the system has a larger buffer, and the
test will succeed even without the fix. But that is OK as long as
somebody somewhere is running the test with a smaller buffer and would
catch a regression.
Of course, convincing git to generate 128K worth of hook input is often
quite inefficient. The existing test in t5571 is quite painful because
it calls "git branch" in a loop. We can do much better than that.
The test below reliably fails without your patch and passes with it, and
seems to run reasonably quickly for me:
diff --git a/t/t5571-pre-push-hook.sh b/t/t5571-pre-push-hook.sh
index 6f9916a..61df2f9 100755
--- a/t/t5571-pre-push-hook.sh
+++ b/t/t5571-pre-push-hook.sh
@@ -109,23 +109,28 @@ test_expect_success 'push to URL' '
diff expected actual
'
-# Test that filling pipe buffers doesn't cause failure
-# Too slow to leave enabled for general use
-if false
-then
- printf 'parent1\nrepo1\n' >expected
- nr=1000
- while test $nr -lt 2000
- do
- nr=$(( $nr + 1 ))
- git branch b/$nr $COMMIT3
- echo "refs/heads/b/$nr $COMMIT3 refs/heads/b/$nr $_z40" >>expected
- done
-
- test_expect_success 'push many refs' '
- git push parent1 "refs/heads/b/*:refs/heads/b/*" &&
- diff expected actual
- '
-fi
+test_expect_success 'set up many-ref tests' '
+ {
+ echo >&3 parent1 &&
+ echo >&3 repo1 &&
+ nr=1000
+ while test $nr -lt 2000
+ do
+ nr=$(( $nr + 1 ))
+ echo "create refs/heads/b/$nr $COMMIT3"
+ echo >&3 "refs/heads/b/$nr $COMMIT3 refs/heads/b/$nr $_z40"
+ done
+ } 3>expected | git update-ref --stdin
+'
+
+test_expect_success 'filling pipe buffer does not cause failure' '
+ git push parent1 "refs/heads/b/*:refs/heads/b/*" &&
+ test_cmp expected actual
+'
+
+test_expect_success 'sigpipe does not cause pre-push hook failure' '
+ echo "exit 0" | write_script "$HOOK" &&
+ git push parent1 "refs/heads/b/*:refs/heads/c/*"
+'
test_done
next prev parent reply other threads:[~2015-11-13 6:17 UTC|newest]
Thread overview: 8+ messages / expand[flat|nested] mbox.gz Atom feed top
2015-11-11 14:39 [PATCH] allow hooks to ignore their standard input stream Clemens Buchacher
2015-11-11 14:42 ` Clemens Buchacher
2015-11-13 6:17 ` Jeff King [this message]
2015-11-13 9:33 ` Clemens Buchacher
2015-11-13 23:23 ` Jeff King
2015-11-16 8:05 ` Clemens Buchacher
2015-11-16 13:59 ` Jeff King
2015-11-13 6:18 ` Jeff King
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20151113061729.GC32157@sigill.intra.peff.net \
--to=peff@peff.net \
--cc=clemens.buchacher@intel.com \
--cc=git@vger.kernel.org \
--cc=junio@pobox.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).