Bug#745795: cache corruption in python-tables on amd64 machines

Antonio Valentino antonio.valentino at tiscali.it
Mon Jun 2 07:21:52 UTC 2014


Hi Wouter,

> From: Wouter Verhelst <wouter at debian.org>
> To: Debian Bug Tracking System <submit at bugs.debian.org>
> Subject: cache corruption in python-tables on amd64 machines
> Date: Fri, 25 Apr 2014 09:19:32 +0200
>
> Package: python-tables
> Version: 2.3.1-3
> Severity: important
>
> Hi,
>
> At a customer, a problem was found with python-tables where, if a file
> was opened twice, the second (and subsequent) time(s) python-tables
> would return junk.
>
> See attached test case. Run like so:
>
> wouter at carillon:~$ python
> Python 2.7.6 (default, Mar 22 2014, 15:40:47)
> [GCC 4.8.2] on linux2
> Type "help", "copyright", "credits" or "license" for more information.
>>>> import bugTesting as b
>>>> b.bugTest('bugged_file.h5')
> [208 135 150   2   0   0   0   0 128 230]
> [0 0 0 0 0 0 0 0 0 0]
> [0 0 0 0 0 0 0 0 0 0]
>>>> b.bugTest('bugged_file.h5')
> [112  22 157   2   0   0   0   0 128 230]
> [0 0 0 0 0 0 0 0 0 0]
> [208 201 160   2   0   0   0   0   0   0]
>>>> b.bugTest('bugged_file.h5')
> [112  22 157   2   0   0   0   0 128 230]
> [0 0 0 0 0 0 0 0 0 0]
> [208 133 159   2   0   0   0   0   0   0]
>>>>
>
> On an i386 machine the output of the above is the same every time, as
> one would expect. On a machine running the amd64 port, however, this is
> not the case.
>
> While I'm filing this bug report on my laptop (which runs unstable), the
> customer in question is running stable and has seen the same problem.
>
> Regards,

The problem seems to be in the HDF5 library rather in PyTables.

I was able to reproduce the issue in a pure C program (attached).

The problem does not occur is one uses the attached testdataset.h5 file
that, in analyzed with "h5dump -A -p",  seems to be substantially
identical to the one provided in the original problem report.

Can you please provide more info about how the test bugged_file.h5 has
been generated?


cheers

--
Antonio Valentino
-------------- next part --------------
A non-text attachment was scrubbed...
Name: debian_745795.txz
Type: application/x-xz-compressed-tar
Size: 1040 bytes
Desc: not available
URL: <http://lists.alioth.debian.org/pipermail/debian-science-maintainers/attachments/20140602/39998c6a/attachment-0001.bin>


More information about the debian-science-maintainers mailing list