fitzgen requested alexcrichton for a review on PR #3836.
fitzgen opened PR #3836 from wasm-mutate-fuzz-targets
to main
:
There's definitely more to be done here but I think this is a good starting point.
The big open question for the future in my mind is how this will integrate with the way we heavily rely on swarm testing for Wasmtime configuration. What's great about
wasm-mutate
is that the seed file is the Wasm, so you can just throw a bunch of interesting Wasm files in your corpus and start exploring some really interesting Wasm programs. But our swarm configuration relies on interpreting the start of the seed files as "DNA" for Wasmtime's configuration. If we do that, we lose the ability to easily add interesting Wasm programs to our corpus. Maybe we don't use swarm testing withwasm-mutate
? And choose only N interesting points in the configuration space at which to test?Also we could basically have a
wasm-mutate
version of every fuzz target we have if we made our infra a little more generic and always used fuel to avoid infinite loops. That seems do-able but will require another overhaul of a lot of our infra, since it kind of assumeswasm-smith
right now.Happy to brainstorm on these things, but I think solving them is something for the future.
fitzgen updated PR #3836 from wasm-mutate-fuzz-targets
to main
.
alexcrichton submitted PR review.
alexcrichton submitted PR review.
alexcrichton created PR review comment:
Could wasmparser be updated elsewhere in wasmtime to ensure we've only got one at a time?
fitzgen updated PR #3836 from wasm-mutate-fuzz-targets
to main
.
fitzgen merged PR #3836.
Last updated: Nov 22 2024 at 17:03 UTC