Revert "ci: Run fuzzer task for the master branch only" #24292
Conversation
This reverts commit 5a9e255.
This gives them a newer clang version, which may have more sanitizers available.
|
I remember part of the original reasoning to disable it for branches was that the corpus is generated for master, and release branches might differ enough for it to run into problems. Is this resolved? |
|
Fuzz inputs are literally untrusted, arbitrary data. The goal of fuzzing is to harden the code base. I think this is especially important for release branches, since releases are built from them. If the fuzz tests on a release branch fail, I think it should be investigated, not suppressed upfront so that we don't even learn about the issue. |
|
And about the fuzz inputs only being generated for one branch: In fact they are only generated for one commit. Any further commit or code change may render them (or parts of them) "invalid". I don't think we can maintain all fuzz inputs for every commit, at least I won't do that and I don't think it is necessary. My point is that we should investigate any issue that comes up, not completely disable the fuzz test (not even trying to compile them). I think, on a case-by-case basis, where it makes sense, we could checkout a specific commit of the qa-assets repo. Or alternatively discard it completely, so that at least compilation is tested. However, finding the appropriate solution can be done when there is a problem. |
|
Assigned 23.0 |
|
It's ok with me as long as it doesn't cause too many false positives, that everyone is going to ignore anyway at some point. |
|
If there is any fuzz issue, feel free to create an issue and ping me if needed. |
…h only" fa27745 ci: Bump fuzz tasks to jammy (MarcoFalke) fab8cd5 Revert "ci: Run fuzzer task for the master branch only" (MarcoFalke) Pull request description: This reverts commit 5a9e255. I think we should attempt to maintain the fuzz tasks for release branches as well. If it is too difficult for one branch, it could make sense to disable it for that branch, but not for all branches unconditionally. Also, bump to jammy. ACKs for top commit: fanquake: ACK fa27745 - we'll see how we go with the 23.x release branch. Tree-SHA512: d6d08e7dce0884b556c51ff1896aebbbb5a805c22decd58af81a04192d19876978696017b489ec55886ddfd5c022963baaab5f11022369ae5291016826ff8017
…er branch only"" This reverts commit 59a50a1.
This reverts commit 5a9e255.
I think we should attempt to maintain the fuzz tasks for release branches as well.
If it is too difficult for one branch, it could make sense to disable it for that branch, but not for all branches unconditionally.
Also, bump to jammy.