Hi John, very strange
1) yes, creduce / run_tests not use TMP/TMPDIR directly 2) TMP=... TMPDIR=.. creduce ... run correctly creduce creates temporaries not in /tmp Note: only one var take effect. Sorry, i don't remember now which one: TMP or TMPDIR. I'll re-check. 3) TMP=... TMPDIR=.. run_tests FAIL after start 4) run_tests starts and create temporaries in /tmp I'll re-check statements 1--4 Thanks for link. Dmitry 2013/5/27 John Regehr <[email protected]>: > Hi Dmitry, > > C-reduce isn't directly using TMP or TMPDIR. Rather, it looks like Perl's > File::Temp module is using these to decide where to create temporary > directories. > > Anyway, I can't think of a great way to avoid using temporary space. Can you > try disabling the tmp-on-tmpfs feature or else point TMPDIR to a different > directory that has more free space? > > http://fedoraproject.org/wiki/Features/tmp-on-tmpfs > > Thanks, > > John > > > > On 05/25/2013 04:37 AM, Дмитрий Дьяченко wrote: >> >> John, i found why run_tests FAIL for me. >> (under bash) if environment variable TMP or TMPDIR is set, then >> run_tests script FAIL. >> >> I not check other shells. >> >> I not check that run_tests PASS. >> Only i check successful run_tests start (see below). >> >> I tried adopt code from creduce-script for run_tests script but fail. >> It's not surprising because of i know nothing about perl :) >> >> Alas, unavailable TMP/TMPDIR may be serious limitation to run tests >> (for new Fedora at least). >> New Fedoras ships with /tmp mounted to tmpfs, i.e. into memory. >> For me free space available in /tmp is under 1Gb. >> >> Thanks, >> Dmitry >> >> 2013/5/20 John Regehr <[email protected]>: >>> >>> Dmitry, I'll be interested to know the cause when you discover it. >>> >>> It might be worth creating a bit of a FAQ about this problem. >>> >>> One cause I've seen is ASLR, which needs to be turned off. >>> >>> Of course any threaded code will be non-deterministic, or thread that >>> executes non-trivial system calls. >>> >>> John >>> >>> >>> >>> >>> On 5/19/13 5:18 PM, Дмитрий Дьяченко wrote: >>>> >>>> Thank John, >>>> hence my fault. i'll try investigate it. >>>> >>>> Dmitry >>>> >>>> 2013/5/20 John Regehr <[email protected]>: >>>>> >>>>> Hi Dmitry, >>>>> >>>>> As far as I can tell, none of the recent commits should make a >>>>> difference >>>>> for the sanity checks! >>>>> >>>>> John >>>>> >>>>> >>>>> >>>>> On 5/19/13 6:03 AM, Дмитрий Дьяченко wrote: >>>>>> >>>>>> >>>>>> John, was your last commit expected to solve sanity checks failures in >>>>>> tests? >>>>>> >>>>>> They appears for me sometimes between early April -- early May 2013. >>>>>> >>>>>> Sounds like perl-code is responsible for fails... >>>>>> >>>>>> Dmitry >>>>>> >>>>>> creduce-trunk-today + clang-trunk-today, Fedora 19 x86_x64, perl >>>>>> 5.16.3 >>>>>> >>>>>> ------------------ tests' run log ------------ >>>>>> ~/build/creduce/tests]$ ./run_tests >>>>>> >>>>>> >>>>>> ===== running test 0 in >>>>>> /home/dimhen/build/creduce/tests/tmp_test0_VevUr >>>>>> ===== >>>>>> >>>>>> >>>>>> ===< 25866 >=== >>>>>> running 4 interestingness test(s) in parallel >>>>>> test (and sanity check) fails at ../../creduce/creduce line 150. >>>>>> cannot remove path when cwd is /tmp/creduce-46BTN3 for >>>>>> /tmp/creduce-46BTN3: at /usr/share/perl5/File/Temp.pm line 902. >>>>>> >>>>>> [...] >>>>>> >>>>>> 2013/5/17 John Regehr <[email protected]>: >>>>>>> >>>>>>> >>>>>>> Ok, I just committed a change that deletes temp dirs more >>>>>>> aggressively >>>>>>> (after each pass) and I don't think it adds any races. >>>>>>> >>>>>>> John >>>>>>> >>>>>>> >>>>>>> >>>>>>> On 05/16/2013 03:05 AM, Konstantin Tokarev wrote: >>>>>>>> >>>>>>>> >>>>>>>> >>>>>>>> >>>>>>>> >>>>>>>> 16.05.2013, 12:51, "Paulo Matos" <[email protected]>: >>>>>>>>> >>>>>>>>> >>>>>>>>> >>>>>>>>> That seems to be an ideal solution. >>>>>>>>> Do we have any rough idea of how much space a long reduction can >>>>>>>>> take? >>>>>>>>> >>>>>>>>> Are we talking of Mbs or Gbs? >>>>>>>> >>>>>>>> >>>>>>>> >>>>>>>> >>>>>>>> I've used test script which produced executable files (and relied on >>>>>>>> creduce >>>>>>>> to clean up all temporary directories). After 10-15 reduction cycles >>>>>>>> it >>>>>>>> has eaten about ~60 Gbs. >>>>>>>> >
