bug#65347: 29.1; Underscore in query replace prevents case-matching

[Spencer Baugh]

Eli Zaretskii <eliz@gnu.org> writes:

>> From: Spencer Baugh <sbaugh@janestreet.com>
>> Date: Wed, 16 Aug 2023 17:27:34 -0400
>> 
>> 
>> 1. emacs -Q
>> 2. (insert "Foo_bar")
>> 3. M-% foo RET quux RET
>> 4. Observe the buffer contains Quux_bar
>> 5. Undo, so the buffer contains Foo_bar again
>> 6. M-% foo_bar RET quux RET
>> 7. Observe the buffer contains quux
>> 
>> Expected behavior:
>> 7. Observe the buffer contains Quux
>> 
>> It seems to me that this would match the docstring better:
>> 
>>   Replacement transfers the case pattern of the old text to the
>>   new text, if both ‘case-fold-search’ and ‘case-replace’ are
>>   non-nil and FROM-STRING has no uppercase letters.
>
> This is a documentation issue: the complex heuristics used by
> replace-match to decide whether and how to "transfer the case pattern"
> depend on whether _all_ the words in the replaced text use the same
> case pattern.  In your example, Foo_bar is 2 words (what is a "word"
> is determined by the buffer's syntax table), and only one of them is
> capitalized.  So Emacs does not capitalize the replacement.  Try doing
> the same with Foo_Bar, and you will see Quux in the replacement.
>
> These subtleties come up from time to time, so I think it's high time
> we documented that (in the hope this will stop this from popping up),
> so I've now done so on the emacs-29 branch.
>
> I think this is all we should do here, as the behavior is correct and
> justified, has been like that since time immemoriam, and we should not
> change it, therefore.

Ah, that makes sense!  I have no further complaint then and we should be
able to close the bug.

Although, a further relevant question: is there some way to make _ not
be a word separator, in this specific context?  Without changing the
syntax table of the buffer overall?  So that foo_bar is interpreted as
one word?

Maybe some kind of escaping?