threads / patch / 28428

patcht4014: remove Message-Id/timestamp before comparing patches

Subject: [PATCH] t4014: remove Message-Id/timestamp before comparing patches

## tl;dr

5 messages between Sep 19, 2011 and Sep 20, 2011. Diffs are folded; open one to read it.

replies: 4people: 3as markdown or json

Thomas Rast· Sep 19, 2011, 06:45 UTC · lore

The test introduced in 787570c (format-patch: ignore ui.color, 2011-09-13) has a race condition: if the two format-patch invocations do not ask for the current time in the same second, their Message-Id headers will disagree.

Normally this would be a pretty unlikely occurrence. However, under valgrind format-patch runs so slowly that the race triggers every time, with a time difference of 2-3s on my hardware.

To avoid this problem, replace the contents of the Message-Id header with a dummy before comparing.

Signed-off-by: Thomas Rast <trast@student.ethz.ch>
---
 t/t4014-format-patch.sh |    7 +++++++
 1 files changed, 7 insertions(+), 0 deletions(-)
Show changes to t/t4014-format-patch.sh +7 −0
diff --git a/t/t4014-format-patch.sh b/t/t4014-format-patch.sh
index 7e405d7..8700089 100755
--- a/t/t4014-format-patch.sh
+++ b/t/t4014-format-patch.sh
@@ -886,11 +886,18 @@ test_expect_success 'empty subject prefix does not have extra space' '
 	test_cmp expect actual
 '
 
+strip_msgid () {
+	sed 's#\(Message-Id: *\)<[^>]*>#\1<MESSAGE@ID>#' "$1" >"$1+" &&
+	mv "$1+" "$1"
+}
+
 test_expect_success 'format patch ignores color.ui' '
 	test_unconfig color.ui &&
 	git format-patch --stdout -1 >expect &&
+	strip_msgid expect &&
 	test_config color.ui always &&
 	git format-patch --stdout -1 >actual &&
+	strip_msgid actual &&
 	test_cmp expect actual
 '
 

-- 
1.7.7.rc1.366.ge210a6
Junio C Hamano· Sep 19, 2011, 17:08 UTC · re: Thomas Rast · lore

Re: [PATCH] t4014: remove Message-Id/timestamp before comparing patches

Thomas Rast <trast@student.ethz.ch> writes:
Show 5 quoted lines
> +strip_msgid () {
> +	sed 's#\(Message-Id: *\)<[^>]*>#\1<MESSAGE@ID>#' "$1" >"$1+" &&
> +	mv "$1+" "$1"
> +}
> +

As this is munging the contents of the line, if future versions of format-patch start coloring the message ID values for some mistake, the test using strip_msgid won't catch them anyway, so I'll amend the stripping to actually strip the message id, like this:

	sed '/^Message-Id: /d' <"$1" >"$1+"
before applying.
Thanks.
Jeff King· Sep 19, 2011, 19:15 UTC · re: Thomas Rast · lore

Re: [PATCH] t4014: remove Message-Id/timestamp before comparing patches

On Mon, Sep 19, 2011 at 08:45:43AM +0200, Thomas Rast wrote:
Show 11 quoted lines
> The test introduced in 787570c (format-patch: ignore ui.color,
> 2011-09-13) has a race condition: if the two format-patch invocations
> do not ask for the current time in the same second, their Message-Id
> headers will disagree.
> 
> Normally this would be a pretty unlikely occurrence.  However, under
> valgrind format-patch runs so slowly that the race triggers every
> time, with a time difference of 2-3s on my hardware.
> 
> To avoid this problem, replace the contents of the Message-Id header
> with a dummy before comparing.

Hmph. I was confused at first reading this, because git format-patch doesn't produce message-ids. Usually.

But it does if you turn on threading, which obviously requires it. And earlier tests set format.thread without bothering to clean up after themselves.

So perhaps this is a better solution:
-- >8 --
Subject: [PATCH] t4014: clean up format.thread config after each test

The threading tests turn on format.thread, but never clean up after themselves, meaning that later tests will also have format.thread set.

This is more annoying than most leftover config, too, because not only does it impact the results of other tests, but it does so non-deterministically. Threading requires the generation of message-ids, which incorporate the current time, meaning a slow-running test script may generate different results from run to run.

Signed-off-by: Jeff King <peff@peff.net>
---
This uses the test_color helper I introduced recently in d960c47. This
should be OK, because ph/format-patch-no-color (which introduces the
problem) is built on top of jk/color-and-pager (which introduces
test_config).

There are several other places in the script that could use the same cleanup, but I didn't bother as they are more complex (e.g., using "--add" to build up multi-value header config) and don't actually cause problems (because they're deterministic, and don't clutter the output in a meaningful way)

 t/t4014-format-patch.sh |    8 ++++----
 1 files changed, 4 insertions(+), 4 deletions(-)
Show changes to t/t4014-format-patch.sh +4 −4
diff --git a/t/t4014-format-patch.sh b/t/t4014-format-patch.sh
index 7e405d7..6797512 100755
--- a/t/t4014-format-patch.sh
+++ b/t/t4014-format-patch.sh
@@ -457,22 +457,22 @@ test_expect_success 'thread deep cover-letter in-reply-to' '
 '
 
 test_expect_success 'thread via config' '
-	git config format.thread true &&
+	test_config format.thread true &&
 	check_threading expect.thread master
 '
 
 test_expect_success 'thread deep via config' '
-	git config format.thread deep &&
+	test_config format.thread deep &&
 	check_threading expect.deep master
 '
 
 test_expect_success 'thread config + override' '
-	git config format.thread deep &&
+	test_config format.thread deep &&
 	check_threading expect.thread --thread master
 '
 
 test_expect_success 'thread config + --no-thread' '
-	git config format.thread deep &&
+	test_config format.thread deep &&
 	check_threading expect.no-threading --no-thread master
 '
 

-- 
1.7.7.rc1.3.gb95be

← back to recent threads