The recommended way is to add the failing inputs to the corpus once the bug is fixed.
This way oss-fuzz will regularly run these inputs.
We also recommend to have your own CI system run the inputs (not necessary with fuzzing),
especially if you have pre-submit testing, to find the regressions earlier.
oss-fuzz does not have any SLA regarding the turnaround time, and we can typically detect a bug within 24 hours after submission. It's very unlikely that we will report a regression earlier than 3-4 hours after the commit.
Having a separate unit test that detects the same bug is redundant to some extent,
but still might be a good idea sometimes (especially if you do run unit tests and don't execute fuzz targets on their corpora before submit)
Do you still have the #21018 (moved) test case? I think you found it with libfuzzer?
(Or is it already in the corpus?)
For #21450 (moved), I think a unit test might be the easiest option, as we want something that differs on i386 and x86_64.
(Are we fuzzing on i386, or only x86_64?)
We should also add a case for #21471 (moved) (and a v3 HS descriptor GET and POST) to http.
Oh, and #21471 (moved) will require a GET request for a v3 HS descriptor, and disabling the encrypted connection check in the function (or changing the mocked connection so it looks like it's encrypted).
We should also add a case for #21471 (moved) (and a v3 HS descriptor GET and POST) to http.
Oh, and #21471 (moved) will require a GET request for a v3 HS descriptor, and disabling the encrypted connection check in the function (or changing the mocked connection so it looks like it's encrypted).
dgoulet wrote a unit test for this, so it's not as essential.
Once these are merged, we can close this ticket as done.
I want to defer fuzzing for issues with v3 hidden services (#21471 (moved)) - I've opened a ticket to implement v3 descriptor fuzzing: #21509 (moved). We can do this as part of testing hidden services in 0.3.0 or 0.3.1.
I've merged test21470-029 to master, because #21507 (moved) isn't backported yet. Once it's backported, we can consider backporting this to 0.2.9 or 0.3.0.