Re: Bug#725516: lbzip2: FTBFS: problems running testsuite
On Tue, Oct 15, 2013 at 9:21 PM, Steven Chamberlain <steven@pyro.eu.org> wrote:
> On 13/10/13 21:21, Mikołaj Izdebski wrote:
>> Upstream lbzip2 uses automake 1.11 and it is not compatible
>> with automake >= 1.13. Attached patch to use automake 1.11.
>> The patch will be included in the next lbzip2 upload.
>
> Thanks a lot for this. The test suite now runs, but I see consistent
> fail results on kfreebsd-amd64:
>
[...]
>
> If I invoke ./Tester or ../src/lbzip2 directly from the shell, it works
> fine for all the valid test cases, and produces the same output as
> minbzcat. It only fails when invoked by 'make check' (using only -j1).
This can be (yet another) problem with Automake. Could you try new upstream
version? The package is not yet in unstable, but you can download it with:
dget -x http://mentors.debian.net/debian/pool/main/l/lbzip2/lbzip2_2.3-1.dsc
This package also uses newer Automake (1.14) which may solve the issue.
>
>> + ./timeout ./minbzcat
>> + rc1=0
>> + ./timeout ../src/lbzip2 -dcqn4
>> lbzip2: unable to create a POSIX thread: Resource temporarily unavailable
>> + rc2=1
>
> (To see that. I added +x to the shell invocation of ./Tester and removed
> 2>/dev/null from the decompressor command lines).
>
> From ktrace/dump:
>
> 92870 lbzip2 CALL
> mmap(0,0x20001000,0x3<PROT_READ|PROT_WRITE>,0x1002<MAP_ANON|MAP_TYPE|MAP_PRIVATE>,0xffffffff,0)
> 92870 lbzip2 RET mmap -1 errno 12 Cannot allocate memory
> 92870 lbzip2 CALL write(0x2,0x61e820,0x4a)
> 92870 lbzip2 GIO fd 2 wrote 74 bytes
> "lbzip2: unable to create a POSIX thread: Resource temporarily
> unavailable
> "
This doesn't seem right. It looks like pthread_create() tries to allocate 500 MB
of memory, which is much more than it should ever need. (lbzip2 itself never
allocates more than 5MB in one chunk.)
>
> It is interesting that only some of the .bz2 test cases trigger the
> problem, while minbzcat has no problem with them. I tried changing the
> 1GB "ulimit -v" in ./Tester to 2GB or 256MB which made no difference.
> The system had approx. 5GB free memory anyway.
In the cases that succeed lbzip2 is supposed to fail (and it does, but
with memory
allocation problem). I suspect that in all cases the mmap() problem occurs.
Thank you for for analysis. If new package doesn't solve this then I will
have a closer look myself.
>
> Regards,
> --
> Steven Chamberlain
> steven@pyro.eu.org
Reply to: