bug#60953: The :match predicate with large regexp in tree-sitter font-lock seems inefficient

[Dmitry Gutov <dgutov@yandex.ru>]

On 31/01/2023 01:57, Yuan Fu wrote:
> 
> 
>> On Jan 30, 2023, at 11:58 AM, Dmitry Gutov <dgutov@yandex.ru> wrote:
>>
>> On 30/01/2023 21:05, Eli Zaretskii wrote:
>>>> Date: Mon, 30 Jan 2023 21:01:02 +0200
>>>> Cc:casouri@gmail.com,60953@debbugs.gnu.org
>>>> From: Dmitry Gutov<dgutov@yandex.ru>
>>>>
>>>> But that doesn't answer the question "Could it?".
>>> I don't understand what you are asking.  "Could" in what sense?
>>
>> Like, would it make sense to try to modify it that way, or extract a function that would do that, without writing it from scratch.
>>
>> Or create a new function which would reuse some common code.
>>
>> We would call the new function something like match_buffer_substring. Optionally, also expose it to Lisp.
> 
> Another option is to change user/programmer’s expectation of the anchor: we could say that the regexp must match the entirety of the node text. IOW, \\` \\' are implied.

Huh, I guess that's an option too.

A couple reasons not to do that would be:

- Potential breakage in all existing TS modes, a week (?) before we're 
going to release Emacs 29 pretest. Maybe that's okay, I can't say. But 
the breakage from that kind of change could be subtle.

- Compatibility reasons? People writing TS modes for Emacs might be 
coming from other editors/TS integrations.

While TreeSitter docs say the predicates are not handled by it, it does 
show this example:

   (#match? @constant "^[A-Z][A-Z_]+")

The use of '^' anchor seems to imply that the regexp doesn't have to 
otherwise match the whole node text (OTOH it's not clear why the example 
doesn't just say "^[A-Z]" or "^[A-Z][A-Z_]").

The doc also references the Rust crate and WebAssembly binding which 
support #match?.

IIUC Rust uses "re.is_match", which is documented to use "implicit .*? 
at the beginning and end". Which matches our current semantics. 
WebAssembly uses "regex.test", same effect.