GNU bug report logs - #59662
29.0.50; [PATCH] Add treesit--indent-defun

Previous Next

Full log

Message #163 received at 59662 <at> debbugs.gnu.org (full text, mbox):

From: Theodor Thornhill <theo <at> thornhill.no>
To: Eli Zaretskii <eliz <at> gnu.org>
Cc: larsi <at> gnus.org, 59662 <at> debbugs.gnu.org, Yuan Fu <casouri <at> gmail.com>,
 monnier <at> iro.umontreal.ca, dgutov <at> yandex.ru
Subject: Re: bug#59662: 29.0.50; [PATCH] Add treesit--indent-defun
Date: Thu, 08 Dec 2022 10:08:17 +0100

Resending because I removed the wrong email to Yuan...

Theodor Thornhill <theo <at> thornhill.no> writes:

> Eli Zaretskii <eliz <at> gnu.org> writes:
>
>>> From: Theodor Thornhill <theo <at> thornhill.no>
>>> Cc: casouri <at> gmail.com, 59662 <at> debbugs.gnu.org, casouri <at> gmail.org
>>> Date: Thu, 08 Dec 2022 09:11:28 +0100
>>> 
>>> > Bonus points for adding something that could use treesit.el
>>> > capabilities, when those are available, instead of syntax-ppss, to
>>> > determine when we are in a comment or in a string.  Or maybe Theo or
>>> > Yuan could suggest such an addition.
>>> 
>>> This works perfectly as is in *-ts-modes.  At least for modes where
>>> syntax-table is set correctly.  I don't think there's a reason to enable
>>> treesit.el capabilities, unless there's a case where this doesn't work.
>>
>> My point is that tree-sitter in many cases parses the program better
>> than syntax-ppss.  So if we can use its information about comments and
>> strings, why not do that?  It's exactly the same logic that Dmitry
>> used:
>>
>
> Something like this?  If we add a defvar such as the below, similar to
> treesit-defun-type-regexp we can use that in the function below:
>
> ```
> (defvar-local treesit-comment-type-regexp nil
>   "A regexp that matches the node type of comment nodes.
>
> For example, \"(line|block)_comment\". ")
>
>
> (defun prog-reindent-defun (&optional argument)
>   "Refill paragraph or reindent the definition that the point is on.
>
> If the point is in a string, or in a comment, or there is a
> comment on the current line, fill the paragraph that the point is
> in or is on the same line.
>
> Otherwise, reindent the definition around or below point."
>   (interactive "P")
>   (save-excursion
>     (if (or (and (treesit-available-p)
>                  (treesit-ready-p (treesit-language-at (point)))
>                  (string-match-p
>                   treesit-comment-type-regexp
>                   (treesit-node-type (treesit-node-at (point)))))
>             (nth 8 (syntax-ppss))
>             (re-search-forward comment-start-skip (line-end-position) t))
>         (if (memq fill-paragraph-function '(t nil))
>             (lisp-fill-paragraph argument)
>           (funcall fill-paragraph-function argument))
>       (end-of-defun)
>       (let ((end (point)))
>         (beginning-of-defun)
>         (indent-region (point) end)))))
> ```
>
> The error handling here is naive, but is this something in the line of
> what you're suggesting?  If so, I can tweak the error handling and
> create a patch for it?  Unless you want to do that, Dmitry
>
>>> +    (if (or (nth 8 (syntax-ppss))
>>> +            (re-search-forward comment-start-skip (line-end-position) t))
>>
>> which falls back on regexp search if syntax-ppss cannot help.  I'm
>> asking why not try tree-sitter before we fall back on syntax-ppss or
>> regexp search?
>>
>> Does that make sense?  If not, why not?
>>
>
> It does.
>
>>> If this ends up in master, should I backport some similar feature for
>>> the *ts-modes?
>>
>> Why do we need this on the release branch?  A general
>> indent-or-fill-comment feature for all the progmodes didn't exist
>> until now, which is why we intend to install this on master.
>
> Yeah, sure :-)
>
> Theo

Previous Next

GNU bug tracking system
Copyright (C) 1999 Darren O. Benham, 1997,2003 nCipher Corporation Ltd, 1994-97 Ian Jackson.