Add Murch’s Inputs July 2025 (2nd attempt)

murchandamus commented at 5:51 pm on July 15, 2025: contributor

Following up on #229 with a second attempt to upstream my inputs.

dergoegge commented at 4:23 pm on July 16, 2025: member

MSan CI fails due to OOM. Since it’s only the MSan job, I’d guess that the MSan instrumentation’s overhead is the cause. Maybe just delete the offending inputs?

murchandamus commented at 4:33 pm on July 16, 2025: contributor

I removed the five slow and the one oom txorphan inputs:

deleted:    115c827e96a74fd0a0de3c4d9454f95d2dd33a7e
deleted:    4e87c2fe521894325f07804c1879fb274d7878aa
deleted:    557d1498f51bbdb9adf44c0cfd749dd85a5420da
deleted:    9d8f0ff4776aab8f425fb5f42d666d68b9ce3151
deleted:    bb0d053ebfd17f042e6aff39b31f56c9f5dc278a
deleted:    f49790635cf83a0c87b445d93e6fc05d034f9d7d

murchandamus force-pushed on Jul 16, 2025

murchandamus commented at 8:57 pm on July 16, 2025: contributor

Squashed it, because CI was unhappy about the deletions, as I should have expected.

maflcko commented at 5:33 am on July 17, 2025: contributor

There should be no need to delete pre-existing inputs from main. Only deleting your own added input(s) should be enough.

ci is trying to tell you that 4e87c2fe521894325f07804c1879fb274d7878aa already exists and was deleted by you.

dergoegge commented at 8:54 am on July 17, 2025: member

Oh, I assumed the troublesome inputs would be new ones. Did we change anything about the MSan CI or the failing harnesses that could cause existing inputs to start failing?

maflcko commented at 9:27 am on July 17, 2025: contributor

the oom inputs was a different one : 115c827e96a74fd0a0de3c4d9454f95d2dd33a7e

murchandamus commented at 4:54 pm on July 21, 2025: contributor

Thanks, I’ll amend the PR tomorrow, when I’m in the office.

murchandamus force-pushed on Jul 22, 2025

maflcko commented at 7:15 pm on July 22, 2025: contributor

Used 9935aeb3bdd556efa5d483777412da03884aa7d6, to include the other pull in the meantime:

https://drahtbot.space/host_reports/DrahtBot/reports/coverage_fuzz/monotree/7129c9ea8e950f50/9935aeb3bdd556ef/fuzz.coverage/index.html

vs main:

https://drahtbot.space/host_reports/DrahtBot/reports/coverage_fuzz/monotree/7129c9ea8e950f50/e6e82b895a44365a/fuzz.coverage/index.html

maflcko commented at 1:20 pm on July 23, 2025: contributor

My report still does not show any function or line coverage difference. Also, according to the CI logs, the coverage is mostly identical: https://cirrus-ci.com/task/5416573638279168?logs=ci#L4498 vs https://cirrus-ci.com/task/6084064303644672?logs=ci#L8614

So I guess there is still something wrong on your side?

murchandamus commented at 2:45 pm on July 23, 2025: contributor

Uh, only 7,000 files after weeks of fuzzing does look wrong. o.0 I’m going to try rebuilding the submission and see if I get a different result, but so far don’t understand what it might be that I have done differently than before my move.

maflcko commented at 3:26 pm on July 23, 2025: contributor

Ok, I am thinking about going with #232 first, and then wait for you to revisit your fuzzing workflow, if this sounds good?

murchandamus force-pushed on Jul 23, 2025

murchandamus commented at 3:39 pm on July 23, 2025: contributor

Please feel free to go ahead with #232. My submission crafting is pretty quick on the new computer, and should my workflow now actually be right, it is easy for me to rebase on the pruned commit history.

murchandamus commented at 3:42 pm on July 23, 2025: contributor

I redid my submission (with one more week of fuzzing under the belt) and pushed. Going to take a look at the CI logs and see if I can find the fuzz_coverage report, when it finishes running. If it still doesn’t improve, I’m gonna comb through my documentation of the process again to make sure I’m not setting myself up to fail in some manner, but if that still yields no explanation, I could probably use some help.

murchandamus marked this as a draft on Jul 23, 2025

maflcko force-pushed the base branch on Jul 24, 2025

maflcko commented at 8:17 pm on July 24, 2025: contributor

I guess you’ll have to cherry-pick your commit after the force push?

murchandamus commented at 9:54 pm on July 24, 2025: contributor

At least compared to the CI logs we used for comparison above, I could find a few targets for which the coverage improved. Will rebase. E.g., autofile went from 468 to 474.

Add Murch’s inputs July 2025 10347b1814

murchandamus force-pushed on Jul 24, 2025

murchandamus marked this as ready for review on Jul 24, 2025

maflcko commented at 10:49 am on July 25, 2025: contributor

Yeah, seems fine to add inputs that trigger coverage internal to the sanitizer instrumentation. However, it would be better if there also was at least one real line of code covered additionally :)

I’ll go ahead and merge this nonetheless.

this pull: https://drahtbot.space/host_reports/DrahtBot/reports/coverage_fuzz/monotree/5ad79b203505fe7b/10347b1814bf734d/fuzz.coverage/index.html

main: https://drahtbot.space/host_reports/DrahtBot/reports/coverage_fuzz/monotree/5ad79b203505fe7b/fd7e08cd37a175b3/fuzz.coverage/index.html

maflcko merged this on Jul 25, 2025

maflcko closed this on Jul 25, 2025

maflcko commented at 2:24 pm on July 30, 2025: contributor

Just checked for comparison on a 8-core vm, running for two weeks, the coverage increase was 35 lines and two new functions. (https://github.com/bitcoin-core/qa-assets/pull/233#issuecomment-3136576838)

Happy to check your logs, if you want, to see if there is anything standing out. But maybe it was just a randomly odd run with little new coverage for you? :man_shrugging:

murchandamus commented at 3:28 pm on July 30, 2025: contributor

That does make me think that something is wrong about my process. What data could I best provide to help?

murchandamus commented at 7:22 pm on August 1, 2025: contributor

I think I’ll just try to craft another submission next week and see if it provides any additional code coverage. If it does, I’ll chalk it up to happenstance, if not, we could investigate any logs that seem useful.

Add Murch’s Inputs July 2025 (2nd attempt) #230