After matching with a String, StringScanner#[:unknown_capture_group] returns nil:

$ irb -rstrscan irb(main):001> s=StringScanner.new("abc") => #<StringScanner 0/3 @ "abc"> irb(main):002> s.scan("a") => "a" irb(main):003> s[:foo] => nil 

After matching with a Regexp, StringScanner#[:unknown_capture_group] raises IndexError:
(using a new process to be sure to not be affected by #135)

irb -rstrscan irb(main):001> s=StringScanner.new("abc") => #<StringScanner 0/3 @ "abc"> irb(main):002> s.scan(/./) => "a" irb(main):003> s[:foo] (irb):3:in `[]': undefined group name reference: foo (IndexError) 

It would be best if the behavior is consistent.

There is a test for the nil case in https://github.com/ruby/strscan/blob/v3.1.2/test/strscan/test_stringscanner.rb#L830-L840
And there are some tests for the IndexError cases in https://github.com/ruby/strscan/blob/v3.1.2/test/strscan/test_stringscanner.rb#L443-L444 and in more places.

The implementation is

This is proving problematic on TruffleRuby.
We would like to upstream our pure-Ruby implementation of StringScanner, but to do so we want to avoid exposing internals as much as possible.
Things like p->regex is something I don't want to expose to code living outside the truffleruby repository.
Having to workaround by keeping extra state for whether the last match was with a String or a Regexp feels very messy and wrong semantically.

Could the behavior be unified for both of these cases?