[17krishna8]

Adds impl_v3_encode and impl_v3_decode fuzz targets that differentially test lib/v3 against lib, as listed in #106 ("Update OSS-Fuzz to also fuzz lib/v3").

Both targets generate an encoding from the fuzz input with the existing gen module, then check that v3 behaves like v2:

• impl_v3_encode: v3.encode(input) == v2.encode(input)
• impl_v3_decode: v3.decode(input) == v2.decode(input), including error kinds and positions

Since v3 must preserve v2 behavior for simple usages, and encodings like #163 (encode by the greatest multiple of blocks that fits 128 bits) are internal rewrites, continuous differential fuzzing should catch behavioral regressions early.

No divergence was found while preparing this change:

• a few hundred thousand libFuzzer executions of both targets with coverage feedback
• an offline differential of about one million generated specifications and inputs (all bit widths, bit orders, paddings, ignored characters, wrappings, and translations)

Note: this does not change the OSS-Fuzz project configuration in google/oss-fuzz, which builds all targets of lib/fuzz and should pick up the new targets automatically.

[ia0]

Thanks a lot for the PR but I think this is too early to merge. I'm still designing v3 with major changes to come and I don't think it's anywhere ready for fuzzing yet. Once I'm satisfied with v3 design, I'll probably refresh this PR to make sure v3 covers all use-cases of v2.

[17krishna8]

Thanks for taking a look! That makes sense - I'll leave the PR open for whenever you are happy with the v3 design, and I'm happy to rebase or rework it however is useful at that point.