Bug#1010202: MatIO test failures against HDF5 1.12.0
Control: forcemerge 1011805 1010202 Le jeudi 19 mai 2022 à 23:39 +0200, Gilles Filippini a écrit : > Thomas Beutlich a écrit le 19/05/2022 à 22:36 : > > Am 19.05.2022 um 10:06 schrieb Sébastien Villemot: > > > Le mardi 03 mai 2022 à 20:59 +0200, Thomas Beutlich a écrit : > > > > Hi Sébastien, > > > > it is the same test case that fails for three different HDF5-based > > > > MAT files. My guess is some concurrency issue of the hdf5lib. Can you > > > > please try: > > > > * to run the testsuite sequentially > > > > * to not build libhdf with --enable-parallel, i.e. to have the > > > > serial version of hdf5lib > > > > * set env var HDF5_USE_FILE_LOCKING=FALSE > > Setting the above environment variable fixes the problem when running > the testsuite in parallel mode. Thanks to both of you for your help on this issue. Actually, it turns out that the problem is not related to HDF5 1.12.0, but to a new debhelper version. More specifically, debhelper 13.7 enabled parallelization of autotest testsuites by default, which exposed the underlying issue related to concurrency in HDF5. In particular, the problem is also present in unstable (see #1011805). Incidentally, this debhelper release also enabled verbose output of the testsuite, which unfortunately makes the output look rather bad. I am going to set HDF5_USE_FILE_LOCKING=FALSE when running the testsuite. Cheers, -- ⢀⣴⠾⠻⢶⣦⠀ Sébastien Villemot ⣾⠁⢠⠒⠀⣿⡁ Debian Developer ⢿⡄⠘⠷⠚⠋⠀ https://sebastien.villemot.name ⠈⠳⣄ https://www.debian.org signature.asc Description: This is a digitally signed message part -- debian-science-maintainers mailing list debian-science-maintainers@alioth-lists.debian.net https://alioth-lists.debian.net/cgi-bin/mailman/listinfo/debian-science-maintainers
Bug#1010202: MatIO test failures against HDF5 1.12.0
Hi, Thomas Beutlich a écrit le 19/05/2022 à 22:36 : Hi Sébastien, seems I was right (and somehow remembered correctly) with my guess of some concurrency issue when running the testsuite. For my own test setup I chose to go with serial execution of the testsuite (if matio is built with libhdf5 support). Best regards, Thomas Am 19.05.2022 um 10:06 schrieb Sébastien Villemot: Hi Thomas, It turns out that running the testsuite sequentially fixes the problem, as explained by Gilles Filippini (who maintains HDF5 in Debian), see: https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1010202#25 Where do we go from here? We now have a workaround, but that’s hardly satisfactory. If possible, please reply by putting p...@debian.org and 1010...@bugs.debian.org in CC, so that this it is properly tracked in our system. Best regards, Le mardi 03 mai 2022 à 20:59 +0200, Thomas Beutlich a écrit : Hi Sébastien, it is the same test case that fails for three different HDF5-based MAT files. My guess is some concurrency issue of the hdf5lib. Can you please try: * to run the testsuite sequentially * to not build libhdf with --enable-parallel, i.e. to have the serial version of hdf5lib * set env var HDF5_USE_FILE_LOCKING=FALSE Setting the above environment variable fixes the problem when running the testsuite in parallel mode. Best, _g. -- debian-science-maintainers mailing list debian-science-maintainers@alioth-lists.debian.net https://alioth-lists.debian.net/cgi-bin/mailman/listinfo/debian-science-maintainers
