Re: [platform-testers] new snapshot available: coreutils-8.24.150-4ab87

[Pádraig Brady]

On 13/01/16 18:30, Pádraig Brady wrote:
> On 13/01/16 17:29, Assaf Gordon wrote:
> 
>> On OpenBSD 5.8:
> 
>> FAIL: tests/misc/sort-debug-warn
>> ================================
>> -sort: failed to set locale; using simple byte comparison
>> +sort: using simple byte comparison
> 
> Interesting. setlocale() succeeds there with LC_ALL=missing
> I'll look into whether we can test this case.

Patch attached.

> 
>> freebsd101-test-suite.log
>>
>> FAIL: tests/misc/head-c
>> =======================
>>
>> + page_size=4096
>> + page_size=4
> 
>> + ulimit -v 10000
>> + head -c1 /dev/null
>> Segmentation fault (core dumped)
> 
> Hmm, we should also be disabling core dumps
> for this ulimit -v determination, with something like?
> 
> trap '' SEGV; ulimit -c 0;

Patch attached

> 
>> + ulimit -v 14004
>> + head --bytes=-9223372036854775807
>> head: memory exhausted
> 
> Does the test pass when changing s/1000/2000/ on the ulimit line?
> 
>> FAIL: tests/misc/kill
> 
>> + SIGINVAL=32
>> + returns_ 1 env kill -l 32 0
>> LWP
>> EXIT
>> + fail=1
> 
> This is a new test, and the scheme used to pick an
> invalid signal number is not general.
> I'll see if I can get sig2str() to return -1 generally.

Looking further it seems that NSIG is only for older signals on FreeBSD
and there is a newer _SIG_MAXSIG that we can add to gnulib like:

diff --git a/lib/sig2str.h b/lib/sig2str.h
index f347170..2730774 100644
--- a/lib/sig2str.h
+++ b/lib/sig2str.h
@@ -44,6 +44,8 @@ int str2sig (char const *, int *);

 #if defined _sys_nsig
 # define SIGNUM_BOUND (_sys_nsig - 1)
+#elif defined _SIG_MAXSIG
+# define SIGNUM_BOUND (_SIG_MAXSIG - 2) /* FreeBSD >= 7.  */
 #elif defined NSIG
 # define SIGNUM_BOUND (NSIG - 1)
 #else


As for the test to get an invalid signal number, this would be more general:

  env kill -t | sort -k 1b,1 > sig.list
  seq 256 | sort -k 1b,1 > all.list
  invalid=$(join -v2 sig.list all.list | head -n1)
  if test "$invalid"; then
    ...
  fi

Related to this I also notice that we wrap signal specs
which doesn't seem appropriate here:

$ bash-kill -l 257
bash: kill: 257: invalid signal specification
$ util-linux-kill -l 257
kill: unknown signal: 257
$ coreutils-kill -l 257
HUP
$ coreutils-kill -l 1
HUP


>> opensuse421-test-suite.log
>>
>> FAIL: tests/df/df-symlink
>> =========================
> 
>> + diff -u exp out
>> --- exp      2016-01-13 12:07:31.789282664 -0500
>> +++ out      2016-01-13 12:07:31.793282621 -0500
>> @@ -1,2 +1,2 @@
>>  Filesystem     Mounted on
>> -/dev/vda2      /
>> +/dev/vda2      /tmp
>> + fail=1
> 
> $ df --out=source '.'
> /dev/vda2
> $ df --out=source,target /dev/vda2
> /dev/vda2      /
> $ df --out=source,target '.'
> /dev/vda2      /tmp
> 
> Upon failure we really should have df tests
> output /proc/mounts and /proc/self/mountinfo if they exist
> to help debug.
> 
>> FAIL: tests/mv/dup-source
>> =========================
>>
>> + diff -u exp out
>> --- exp      2016-01-13 12:07:44.597145567 -0500
>> +++ out      2016-01-13 12:07:44.593145610 -0500
>> @@ -2,4 +2,4 @@
>>  mv: cannot stat 'a': No such file or directory
>>  mv: cannot stat 'b': No such file or directory
>>  mv: cannot move './b' to a subdirectory of itself, 'b/b'
>> -mv: warning: source directory 'b' specified more than once
>> +mv: cannot move 'b' to a subdirectory of itself, 'b/b'
>> + fail=1
> 
> Interesting. The mv case is different to cp on this system.
> I can't think why off the top of my head.  The "earlier_file"
> must not be found in the hash for some reason.
> 
>> centos65-test-suite.log
>>
>> ERROR: tests/cp/fiemap-extents
>> ==============================
> 
>> + fallocate -l 4MiB -n unwritten.withdata
>> fallocate: invalid length value specified
>> + framework_failure_
> 
> It looks like IEC number suffix parsing was only added
> to fallocate recently.  We can generalize that or skip on older systems.

Patch attached

cheers,
Pádraig.

8.25-snap1-fixes.patch
Description: Text Data