-
Show log
Commit
-
Hash : 9eab3ac7
Author :
Date : 2022-07-21T12:35:19
[CoverageFormat2] Remove hand-written loop While on a fuzzer-found test case (added) that loop was faster, on real fonts, including NotoNastaliq in our benchmark, it was actually slower, which intuitively I would have expected. Still no idea why on that fuzzer case it's faster though. :(
-
README
-
In order to build the fuzzer one needs to build HarfBuzz and harfbuzz/test/fuzzing/hb-fuzzer.cc with: - Using the most recent Clang - With -fsanitize=address (or =undefined, or a combination) - With -fsanitize-coverage=edge[,8bit-counters,trace-cmp] - With various defines that limit worst case exponential behavior. See FUZZING_CPPFLAGS in harfbuzz/src/Makefile.am for the list. - link against libFuzzer To run the fuzzer one needs to first obtain a test corpus as a directory containing interesting fonts. A good starting point is inside harfbuzz/test/shaping/fonts/fonts/. Then, run the fuzzer like this: ./hb-fuzzer -max_len=2048 CORPUS_DIR Where max_len specifies the maximal length of font files to handle. The smaller the faster. For more details consult the following locations: - http://llvm.org/docs/LibFuzzer.html or - https://github.com/google/libfuzzer-bot/tree/master/harfbuzz - https://github.com/harfbuzz/harfbuzz/issues/139