[petsc-dev] Valgrind report a bug in PetscOptionsFindPair_Private

[Smith, Barry F.]

   Fixed in barry/fix-petscstrlcat/maint

   I also doubled the length of the work strings used in processing options so your options won't get truncated and hence not get processed properly.

    Barry


> On Jun 20, 2018, at 8:48 AM, Eric Chamberland <Eric.Chamberland at giref.ulaval.ca> wrote:
> 
> Hi,
> 
> since we recently upgraded our nighlty valgrind validations to PETSc 3.9.2, valgrind found a bug in PetscOptionsFindPair_Private, called from:
> 
> PetscOptionsHasName("M_AssembleurGD_H_/pmi/cmpbib/compilation_BIB_gcc_valgrind/COMPILE_AUTO/BIBTestValidation/Ressources/opt/Test.assemblageDynamique/Ressources.Lettre_C_2016_gsf_P2B5_Z450_mp/Lettre_C_2016_gsf_P2B5_Z450_mp_A_SolveurLin_pour_solveur_","-::GIREF_NOM_PREFIXE_OPTIONS_PETSC");
> 
> valgrind complains with:
> 
> ==253393== Source and destination overlap in strncat(0x1ffeff62e0, 0x1ffeff6299, 108)
> ==253393==    at 0x4C2C976: strncat (vg_replace_strmem.c:348)
> ==253393==    by 0x1522912F: PetscStrlcat (str.c:460)
> ==253393==    by 0x151DE08A: PetscOptionsFindPair_Private (options.c:1361)
> ==253393==    by 0x151DF153: PetscOptionsHasName (options.c:1502)
> ...
> 
> We noticed that the PETSc code has changed close from there, so maybe it is related?
> 
> We do not have the bug with Petsc versions: {3.5.3,3.8.3} that we used to tests with valgrind.
> 
> Thanks,
> 
> Eric