Continuous fuzzing roadmap toward OSS-Fuzz integration

[LucaCappelletti94]

I believe having continuous fuzzer will help us a lot in the process more effectively the increasing number of PR and avoid merging regressions, other than of course discover current errors.

I propose the following sequence of PRs:

1. Switching from honggfuzz to cargo-fuzz/libFuzzer harness. I agree that in the past honggfuzz was the goto solution, and it was also my favourite, but cargo-fuzz has improved a lot in the last few years and is required by OSS-Fuzz and ClusterFuzzLite. Done in Fuzz: Replace honggfuzz harness with cargo-fuzz #2520
2. ClusterFuzzLite PR gate + daily batch, same targets. Ready for review in Fuzz: Wire up ClusterFuzzLite CI fuzzing #2527
3. Adding a harness executing the parse -> Display -> re-parse loop. Ready for review in Fuzz: Display roundtrip #2528
4. Actually committed seed corpus. Ready for review in Fuzz: Add curated seed corpus #2530
5. OSS-Fuzz application (projects/datafusion-sqlparser-rs) which I believe we may be approved for, given the reasonable user pool of this crate. This PR should only be opened after we reach some fuzzer stability.
6. Fuzz-byte-driven weighted recursive query generator (SparkFuzz/SQLsmith style), sidestepping Added derive for arbitrary #2193's Arbitrary blocker which is still blocked to this day because of the ignored upstream PR Added support for Arbitrary akubera/bigdecimal-rs#163 @iffyio opinions?

EDIT: swapped order of steps as mentioned in #2519 (comment)

[alamb]

Adding a harness executing the parse -> Display -> re-parse loop, plus an actually committed seed corpus.
That sounds good to me

In my mind, I would suggest startign small and targeted (rather than trying to automate everything all at once)

So for example, we could start with basic fuzzing in CI and more on demand fuzzing for longer runs.

[LucaCappelletti94]

In my mind, I would suggest startign small and targeted (rather than trying to automate everything all at once)

Yeah, maybe I should move point 3 at the end, since it is going to be the largest one by far.

So for example, we could start with basic fuzzing in CI and more on demand fuzzing for longer runs.

Yes, this is point 5

[LucaCappelletti94]

@alamb does the new roadmap look good? If so, I can start by merge in #2520 and setup ClusterFuzzLite analogously to diesel-rs/diesel#5212

[LucaCappelletti94]

First few bugs identified by the updated fuzzer: #2533 #2532 #2531