fuzzbench
fuzzbench copied to clipboard
Unique code coverage plots unreliable
Unique code coverage plots are currently unreliably populated in the final FuzzBench report. Investigate why and fix this.
@jonathanmetzman could you add a bit more details here on what needs to be fixed?
This seems like it's related to https://github.com/google/fuzzbench/issues/1491 due to this condition:
https://github.com/google/fuzzbench/blob/669805b3b107c1b13c27c636a3e964ad929df0ba/experiment/dispatcher.py#L183
Yeah basically the problem is we only generate the report once, after completion. This is fragile and in #1491 we see experiment measuring is prone to not completing.