Delete nonreduced fuzz inputs#263
Conversation
Storage device usage (
|
|
Looks like a few lines of coverage went away. Possibly due to bitcoin/bitcoin#29018 :( Still, I guess this should be good to go. |
|
To see the number of files deleted per test: |
|
Would it be useful to keep relatively new files around, e.g. not delete any files that were added to qa-assets less than 2 years ago, because those might be relevant to old maintained branches even if they are not relevant to the latest release anymore? |
Yeah, I guess that is possible. Though, my preference would probably be to just create a branch of qa-assets (and use that in the previous release) Otherwise, it will be difficult or inefficient to delete any nonreduced fuzz inputs at all, because:
|
|
A more precise cross-diff shows that the larger diffs happen in the targets that are more non-deterministic: |
Yeah, that's a possibility too. Alternatively, the delete_nonreduced script could run merging with multiple releases/branches. It was just an idea I wanted to throw out there, but to be clear it's not an objection to this PR or even a suggestion to address right now. |
That'd probably be ideal. Just creating a branch, or just merging against master will not supply new fuzz inputs for older releases, whereas merging against the previous branches will do that, while also deleting nonreduced inputs that are "unused" on all branches. |
|
rfm? |
|
Are we concerned about the valgrind failure here? https://github.com/bitcoin-core/qa-assets/actions/runs/22990374949/job/66749878719?pr=263 |
|
I think this is just a known upstream valgrind bug. A possible temporary workaround is in bitcoin/bitcoin#34589 |
As per the usual process to avoid wasted CI resources and timeouts when CI runs on large and presumed irrelevant inputs.
Normally, deletion of non-reduced fuzz inputs should happen after feature-freeze on the
masterBitcoin Core branch, but before branch-off, so that the latest release branch retains mostly valid fuzz inputs.Previous: #239
To "reproduce"
Install a fresh VM, as explained in the bash script's doc, and run it:
To "test"
CI
CI should pass, except for a
lintfailure, which should light up on any changes like this pull request, which delete fuzz inputs.