Did you run the entire testsuite or just the single .mod file?

Was your workaround for byte code something like the following in /mex/sources/bytecode/ErrorHandling.hh

+#ifdef __GNUC__
+#  ifndef isnan
+#     define isnan(x) __builtin_isnan(x)
+#  endif
+#  ifndef isinf
+#    define isinf(x) __builtin_isinf(x)
+#   endif
+#endif

Best,
Houtan

On Jun 8, 2017, at 15:38, Sébastien Villemot <sebastien.villemot@sciencespo.fr> wrote:

I have recompiled Octave 4.2.1 and Dynare 4.5 on Debian Jessie (working around the bytecode issue), in order to mimic the environment of jessie, and I still don't get the crash. I don't understand how I got it in the first place.

It does not look like a problem with consistently uninitialized memory space, because a check with valgrind could not unveil anything.

So I'm giving up on this issue for now.

Le jeudi 08 juin 2017 à 14:34 +0200, Johannes Pfeifer a écrit :
I consistently get it on Sedna. The message is
 
warning: matrix singular to machine precision, rcond = 1.34593e-17
warning: called from
    dyn_first_order_solver at line 313 column 9
    stochastic_solvers at line 264 column 18
    resol at line 144 column 14
    dynare_resolve at line 69 column 44
    dsge_likelihood at line 266 column 57
    hessian at line 92 column 34
    dynare_estimation_1 at line 244 column 24
    dynare_estimation at line 105 column 5
    fs2000_ML at line 218 column 14
    dynare at line 223 column 1
*** Error in `/usr/local/bin/octave-cli-4.2.1': double free or corruption
(!prev): 0x0000000002221150 ***
panic: Aborted -- stopping myself...
attempting to save variables to 'octave-workspace'...
*** Error in `/usr/local/bin/octave-cli-4.2.1': corrupted double-linked
list: 0x00000000021fd560 ***
panic: attempted clean up failed -- aborting...


Thus, the crash happens during or immediately after computing the Hessian.
My hunch is that the Hessian again returns something complex with a NaN in
it and we get the same crash as in the other mod-file when the inverse
Hessian is supposed to be computed.

-----Ursprüngliche Nachricht-----
Von: Dev [mailto:dev-bounces@dynare.org] Im Auftrag von Sébastien Villemot
Gesendet: Donnerstag, 8. Juni 2017 14:18
An: dev@dynare.org
Betreff: Re: [DynareDev] Beta release of 4.5.0

Well, I am no longer able to reproduce the crash in
kalman_filter_smoother/compare_results_simulation/fs2000_ML.mod.

Do you have a way to consistently trigger it?

Le mercredi 07 juin 2017 à 18:16 +0200, Johannes Pfeifer a écrit :
Have you tried on Linux whether the problem also occurs on the Linux GUI? On the Windows CLI, everything runs smoothly, but there are still figures showing up on screen, which is different from the Linux CLI.   Von: Dev [mailto:dev-bounces@dynare.org] Im Auftrag von Sébastien Villemot Gesendet: Mittwoch, 7. Juni 2017 17:39 An: dev@dynare.org Betreff: Re: [DynareDev] Beta release of 4.5.0   Le mercredi 07 juin 2017 à 16:54 +0200, Johannes Pfeifer a écrit :   Thanks a lot. I cannot replicate the third crash on Windows. On Linux, I get   *** Error in `/usr/local/bin/octave-cli-4.2.1': double free or corruption (!prev): 0x000000000                 26faad0 *** panic: Aborted -- stopping myself... attempting to save variables to 'octave-workspace'... save to 'octave-workspace' complete Aborted   The first one seems to be a bug in Octave. I filed a bug report at ht tps://savannah.gnu.org/bugs/index.php?51198   Thanks. I have added some technical information and a tentative patch to the bug tracker.   So I think we're done with the crash in estimation/fs2000_calibrated_covariance.mod.   Let's now investigate the crash in kalman_filter_smoother/compare_results_simulation/fs2000_ML.mod.         Von: Dev [mailto:dev-bounces@dynare.org] Im Auftrag von Sébastien Villemot Gesendet: Mittwoch, 7. Juni 2017 12:10 An: dev@dynare.org Betreff: Re: [DynareDev] Beta release of 4.5.0   Dear all,   Le lundi 29 mai 2017 à 23:46 +0200, Stéphane Adjemian a écrit :   I do not receive integration tests reports for Octave since ages... I don't know why. Before the disappearance of the reports we had some failing tests, it would be nice to fix them. The difficulty with the Octave testsuite is that it often never ends. Some mod files, not always the same, run forever. I am convinced that this is not a bug in these files but rather a consequence of the memory leaks Ferhat has to fix in bytecode.   I have run the testsuite for Octave. The environment is Debian unstable, with Octave 4.2.1 from Debian experimental, and a few Octave forge packages recompiled by hand. The commit tested is d608588ef2 on the 4.5 branch.   Basically there are only 3 test failures : estimation/fs2000_calibrated_covariance.mod estimation/fs2000_model_comparison.mod kalman_filter_smoother/compare_results_simulation/fs2000_ML.mod   The first and the third ones are crashes in Octave. I am going to investigate them. The second failure is simply the consequence of the first failure (the second test needs the output of the first one).   -- Sébastien Villemot Économiste Sciences Po, OFCE _______________________________________________ Dev mailing list Dev@dynare.org https://www.dynare.org/cgi-bin/mailman/listinfo/dev -- Sébastien Villemot Économiste Sciences Po, OFCE _______________________________________________ Dev mailing list Dev@dynare.org https://www.dynare.org/cgi-bin/mailman/listinfo/dev
-- 
Sébastien Villemot
Économiste
Sciences Po, OFCE

_______________________________________________
Dev mailing list
Dev@dynare.org
https://www.dynare.org/cgi-bin/mailman/listinfo/dev