Verification failures are printed in label order, making investigating failures difficult about ebpf-verifier HOT 7 CLOSED

The flag --assume-assert is designed to help in this exact case. Doesn't it?

from ebpf-verifier.

@Alan-Jowett did you experiment with that flag?

from ebpf-verifier.

Thanks for the feedback. What does this flag do? Should it be on by default when performing the analysis? I will try it out.

from ebpf-verifier.

The flag forces the invariants to match the assertions, even if they did not. So errors are not repeated and not compounded.

It used to be the default. I don't remember the reasoning behind disabling it, but in part it's because expected failures in tests stress the code less (since many invariants drop to bottom immediately).

Please let me know if the flag helps.

from ebpf-verifier.

That flag is awesome!

From 507 errors -> 1 error, and it's the actual relevant error that is causing the problem. This is even better than my approach. Is there any harm in making this the default in eBPF-for-Window's calls to the verifier?

I will go ahead and close this PR unless you see value in it.

from ebpf-verifier.

I think it should be on by default during debugging, and off by default in tests.

from ebpf-verifier.

Thanks!

from ebpf-verifier.