Continuous fuzzing

[AdamKorcz]

Hello,

I see that Compress is being fuzzed which is awesome!

I would like to suggest setting up continuous fuzzing by way of Googles oss-fuzz infrastructure. Continuous fuzzing will help with running the fuzzers regularly as well as fuzzing for deeper bugs that take longer time to discover.

Oss-fuzz is a free service that is offered with an implied expectation that found bugs are fixed, so that the resources spent on fuzzing Compress go towards fixing bugs. I will be happy to integrate Compress into oss-fuzz. All I need are the email addresses for the maintainers that should receive potential bug reports. Note that these email addresses will be added to a public list that can be amended at any time.

[klauspost]

Sounds like a good idea. I am missing the continuous fuzzing.

[AdamKorcz]

Sure. To proceed, please leave the email addresses for maintainers to receive bug reports, and I will setup integration.

[klauspost]

Wow this is complicated to set up :/

[AdamKorcz]

yeah, can be tricky. If you need me to set it up, let me know, I have it ready already

[alexey-milovidov]

It is a prerequisite for using S2 in ClickHouse.

[klauspost]

All fuzzing has been migrated to Go 1.18 Fuzzing and there is 500K iteration fuzzing running on every PR now.

I still have no objection to adding oss-fuzz.

[alexey-milovidov]

Wow! It means that the concern is resolved completely. Thank you!

[AdamKorcz]

Okay, I will get a PR up on OSS-Fuzz this weekend for integration.

[klauspost]

@AdamKorcz I don't know if a complication is that is has several Fuzz tests, and as you can see I run them both with and without assembly -tags=noasm - and a few where it is assembly with some features disabled FuzzNoBMI2Dec

I have an extended corpus from running them here. I try to keep the "public", in repo, corpus reasonably small, since it impacts the download size.

It could be set up with a script that has -fuzztime=30m on each, after which it moves on to the next.

If there is any way I can help, let me know. The go.yaml contains all the fuzz tests that are relevant.

[klauspost]

@AdamKorcz

I feel that I am about to get into trouble with adding a seed file. (I have removed that line for now)

Since your PR lists files to copy.

I was thinking of adding a tag=fuzznoext so it would simply skip importing the external seed files.

For dictionaries that is needed for the tests I could embed them in the binary with go:embed.

These are the solutions with the least disruptive updates I can think of, that also would require the least maintenance.