String#casecmp? fails between ascii and utf-8

[notEthan]

Comparing an ASCII-8BIT string with UTF-8, for both of which #ascii_only? is true, using String#casecmp? returns nil rather than a boolean e.g.

''.b.casecmp?('')

It returns boolean on MRI, and the implementation to return boolean is present in truffleruby but not reached:

https://github.com/oracle/truffleruby/blob/vm-22.3.0/src/main/ruby/truffleruby/core/string.rb#L1204-L1213
The check if ascii_only? && other.ascii_only? is not reached after Primitive.encoding_compatible?(encoding, other.encoding) returns nil.

[eregon]

Thank you for the report.
We should use Primitive.encoding_ensure_compatible_str self, other there I suspect.

[eregon]

@notEthan Would you be interested to submit a PR for this, or you prefer the team fixes this? (either is fine)
I'm asking because this seems like it could be an easy contribution.

[notEthan]

I had a look but getting a truffleruby development/build/test environment rolling looks to be more involved than I have time to put into this at the moment.

[eregon]

No worry, the team will fix it then.
Building from source is explained in https://github.com/oracle/truffleruby/blob/master/doc/contributor/workflow.md, it's mostly automated except setting up jt.

[notEthan]

Nice, thanks for fixing.
I do want to note that this applies to nonempty strings too, I just used empty strings in the example for conciseness. The code changes look like they apply to all strings, though the tests only use empty ones. (And, looking at the code changes, kinda glad I didn't try to take this on as my first easy contribution)

[eregon]

Ah yes of course. The change is to use strings for encoding negotiations (which considers code range and empty) vs only encodings of those strings.
The fix is actually just: 7a96f2c, 7dd4301 is the merge commit and include an unrelated cleanup.