2026-04-29 07:45 UTC
@shajra I'm open to being told I'm wrong or missing something, but I don't get the value add. I did try a number of these libraries at one point and found them just way too slow. The last thing I fuzzed before this I was getting 1M executions/s per core for the fundamental thing; no way I want Python test gen.
For the text editor, maybe I could do it for e2e on the editor, but I'm not sure what the value add is over the simple thing I'm doing that took basically no time.
Replies (1)
-
@robinp@mastodon.social 2026-04-29 10:11
@danluu @shajra fuzzers are more blackbox oriented/autonomous, PBT is greybox and you need to come up with the properties to check (via autogenerated inputs). Coming up with props is somewhat hard, but then serve as a specification of your impl. Main diff sounds like, a fuzzer can discover a crash, not a non-crashing but logically incorrect answer. I can imagine a small intersection in usecase, like your describe the data corruption test. See the Quickchecking Dropbox paper.