Re: coreutils-8.3: a few tests fail

[Jim Meyering]

Pádraig Brady wrote:

> On 13/01/10 10:51, Jim Meyering wrote:
>> Christophe LYON wrote:
>>> I have built coreutils-8.3 and ran 'make check' on a few systems and I
>>> have noticed some FAILs as follows. Given their old age, I am not sure
>>> it is worth reporting, though.
>>>
>>> * Linux RedHat 3/x86: tail-2/inotify-hash-abuse, log extract:
>>>    + kill -HUP 15829
>>> [1]+  Hangup                  tail $debug -qF $(seq $n)>out 2>&1
>>> + wait 15829
>>> ./tail-2/inotify-hash-abuse: line 59: wait: pid 15829 is not a child
>>> of this shell
>>> + st=127
>>> + echo tail died via unexpected signal: 127
>>> tail died via unexpected signal: 127
>>> + fail=1
>>>
>>> * Linux RedHat 3/x86: tail-2/inotify-hash-abuse2, log extract:
>>> + kill -HUP 16044
>>> ./tail-2/inotify-hash-abuse2: line 40: kill: (16044) - No such process
>>> + wait 16044
>>> + st=124
>>> + echo tail died via unexpected signal: 124
>>> tail died via unexpected signal: 124
>>> + fail=1
>>>
>>> * Linux RedHat 3/x86_64: test-utimens (test-utimens.h:101: assertion failed)
>>>
>>> * Linux RedHat 4/x86: test-nanosleep (test-nanosleep.c:78: assertion failed)
>>> * Linux RedHat 4/x86: test-sleep (test-sleep.c:53: assertion failed)
>>
>> Thanks for the report.
>> Can you repeat the tests with the very latest snapshot?
>>
>>    http://meyering.net/cu/coreutils-ss.tar.gz      9.4 MB
>>    http://meyering.net/cu/coreutils-ss.tar.xz      4.0 MB
>>
>> If any of those failures persist, we *might* have time to address
>> one or two before coreutils-8.4, which is slated to be released today.
>
> I think we may be encountering shell issues here?
>
> We get 127 rather than 129 under dash
> once the "hangup" message is displayed before
> `wait` is called
>
> $ /bin/sleep 20 & bg=$!
> $ kill -HUP $bg
> [1] + Hangup                     /bin/sleep 20
> $ wait $bg
> $ echo $?
> 127

I think so to, and had just been working on that.
What do you think of these changes?

>From 2eb54470541557fd89b934a7b2245fe4b1515b82 Mon Sep 17 00:00:00 2001
From: Jim Meyering <address@hidden>
Date: Wed, 13 Jan 2010 12:46:25 +0100
Subject: [PATCH] tests: decrease risk of race in tail-2/inotify-hash-abuse2

* tests/tail-2/inotify-hash-abuse2: Try to avoid killing the
tail process before we start waiting for it to die.
Also, don't assume that kill -HUP is the same as kill -1.
Use the latter.
Also, always use tail's -s .001 option, to make the non-inotify
case act more like the inotify-enabled one.
---
 tests/tail-2/inotify-hash-abuse2 |   12 +++++++++---
 1 files changed, 9 insertions(+), 3 deletions(-)

diff --git a/tests/tail-2/inotify-hash-abuse2 b/tests/tail-2/inotify-hash-abuse2
index 4fe080b..5966b03 100755
--- a/tests/tail-2/inotify-hash-abuse2
+++ b/tests/tail-2/inotify-hash-abuse2
@@ -26,9 +26,9 @@ fi

 touch f || framework_failure

-debug='---disable-inotify -s .001'
+debug='---disable-inotify'
 debug=
-timeout 10 tail $debug -F f & pid=$!
+timeout 10 tail $debug -s .001 -F f & pid=$!

 for i in $(seq 200); do
   kill -0 $pid || break;
@@ -37,8 +37,14 @@ for i in $(seq 200); do
 done

 # Fixed tail will not have aborted.  Kill it.
-kill -HUP $pid
+( touch witness; sleep 1; kill -1 $pid ) &

+# Wait for the killer subshell to start, so that we decrease the
+# risk of 'wait'ing *after* the tail process has been killed.
+until test -f witness; do :; done
+
+# Hope that we begin to wait for $pid before it is killed.
+# FIXME: this is racy...
 wait $pid
 st=$?

--
1.6.6.511.gf46c4