Ignore more exceptions in stubtest

[JelleZijlstra]

See python/typeshed#6801.

[sobolevn]

Do we need to log it somehow?

[JelleZijlstra]

Not sure there's a good framework for that. Perhaps we can add a constant similar to MISSING, or fall back to a runtime value of Any if a non-AttriibuteError exception is thrown.

[hauntsaninja]

This looks good, although it might make it possible for verify to be called with verify(MISSING, MISSING, ...), which gives a not ideal error:

In [4]: print(list(verify(MISSING, MISSING, ["asdf", "fdsa"]))[0].get_description())
error: asdf.fdsa is not present in stub
Stub:
MISSING
Runtime:
MISSING

[JelleZijlstra]

I don't think that would be new from this change, since it already did getattr(..., ..., MISSING).

[hauntsaninja]

It is new. The way we source entries, they're guaranteed to exist either in the stub or at runtime. Now if we have something that isn't in the stubs, shows up in vars but raises an error when you getattr it, we'll end up with MISSING, MISSING. It's niche and I thought stubtest would choke, but it looks like it'd just ends up giving a slightly confusing message, which is fine.

Actually, we should not be calling verify if getattr fails with an unexpected Exception. So something like this will avoid confusing errors:

try:
    runtime_attr = getattr(runtime, mangled_entry, MISSING)
except Exception:
    pass
else:
    yield from verify(stub_to_verify, runtime_attr, object_path + [entry])

[JelleZijlstra]

Thanks, I applied that suggestion.

[hauntsaninja]

Looks like githubstatus.com just acknowledged issues with Github Actions (I ran into issues on another repo as well)