search for: yugr

...>>>>>> https://bugzilla.samba.org/show_bug.cgi?id=3959 >>>>>>>> ). >>>>>>>> >>>>>>>> I've recently developed a simple proof-of-concept tool SortChecker ( >>>>>>>> https://github.com/yugr/sortcheck) which intercepts calls to >>>>>>>> qsort and >>>>>>>> friends (via LD_PRELOAD) and performs various sanity checks before >>>>>>>> passing >>>>>>>> control to libc e.g. >>>>>>...

...e this will usually result in slightly invalid output (not fully sorted arrays, rare failed bsearches, etc.) but may as well cause aborts on some systems (https://bugzilla.samba.org/show_bug.cgi?id=3959). I've recently developed a simple proof-of-concept tool SortChecker (https://github.com/yugr/sortcheck) which intercepts calls to qsort and friends (via LD_PRELOAD) and performs various sanity checks before passing control to libc e.g. * sign(cmp(a, b)) == - sign(cmp(b, a)) for all array elements * etc. Results were quite inspiring: I've found several errors in popular open-source...

...alid output (not >> fully sorted arrays, rare failed bsearches, etc.) but may as well cause >> aborts on some systems (https://bugzilla.samba.org/show_bug.cgi?id=3959). >> >> I've recently developed a simple proof-of-concept tool SortChecker ( >> https://github.com/yugr/sortcheck) which intercepts calls to qsort and >> friends (via LD_PRELOAD) and performs various sanity checks before passing >> control to libc e.g. >> * sign(cmp(a, b)) == - sign(cmp(b, a)) for all array elements >> * etc. >> >> Results were quite inspiring: I...

...e failed bsearches, etc.) but may as well cause >>>> aborts on some systems (https://bugzilla.samba.org/show_bug.cgi?id=3959 >>>> ). >>>> >>>> I've recently developed a simple proof-of-concept tool SortChecker ( >>>> https://github.com/yugr/sortcheck) which intercepts calls to qsort and >>>> friends (via LD_PRELOAD) and performs various sanity checks before passing >>>> control to libc e.g. >>>> * sign(cmp(a, b)) == - sign(cmp(b, a)) for all array elements >>>> * etc. >>>> &g...

...t;> aborts on some systems ( >>>>>> https://bugzilla.samba.org/show_bug.cgi?id=3959 >>>>>> ). >>>>>> >>>>>> I've recently developed a simple proof-of-concept tool SortChecker ( >>>>>> https://github.com/yugr/sortcheck) which intercepts calls to qsort and >>>>>> friends (via LD_PRELOAD) and performs various sanity checks before >>>>>> passing >>>>>> control to libc e.g. >>>>>> * sign(cmp(a, b)) == - sign(cmp(b, a)) for all array ele...

...need to consider at least 3. > > As I privately replied to Kostya, I was only willing to add constant overhead. Sure, there are various conflicting tradeoffs. SortChecker started as an experiment so I was as agressive as possible. > SortChecker adds N*N overhead: > https://github.com/yugr/sortcheck/blob/master/src/sortchecker.c#L427 > except he caps it at N=32 on line 413 I'm afraid it's even N^3 (with N=32 cap). This indeed sounds scary but I have not seen significant slowdowns when running instrumented distro. > (and thus will not detect all > violations either)...