Thomas Dullien

Results 6 issues of Thomas Dullien

Investigate terrible x64 function-entry-point recognition performance from dyninst

Python tests are similarly ultra-brittle to changes in the hashing algo

Many of the tests are brittle when adding new features to the SimHash

1 comment

This was a desired feature initially to make sure the underlying disassembly is good, but makes tweaking / improving / adding new features difficult without breaking the tests. Particular culprits...

Upgrade to latest DynInst to see if the ASAN-identified memleaks are gone

ASAN complains about memleaks from DynInst; check if these are still present in the latest version.

De-duplicate the feature generator between dyninstfeaturegenerator.cpp and flowgraphwithinstructions.cpp

There's a comment about de-duplicating this code in dyininstfeaturegenerator; the current duplication of the code makes it very brittle. The performance cost of forcing the dyninstfeaturegenerator to construct a Flowgraphwithinstructions...

Modified the loop that counts non-ASCII characters in _string_from_ge…

When scanning lots of data, the existing code can spend significant amounts of time in the loop that checks for non-ASCII characters. Unfortunately, all compilers I tested failed to properly...