GNU bug report logs - #65885
29.1.50; wrong behaviour of treesit-beginning-of-defun in c++-ts-mode

Previous Next

Full log

View this message in rfc822 format

From: Yang Yingchao <yang.yingchao <at> qq.com>
To: Eli Zaretskii <eliz <at> gnu.org>
Cc: Yuan Fu <casouri <at> gmail.com>, 65885 <at> debbugs.gnu.org
Subject: bug#65885: Acknowledgement (29.1.50; wrong behaviour of treesit-beginning-of-defun in c++-ts-mode)
Date: Thu, 14 Sep 2023 08:17:56 +0800

[Message part 1 (text/plain, inline)]

>> From: Yang Yingchao <yang.yingchao <at> qq.com>
>> Date: Wed, 13 Sep 2023 08:41:27 +0800
>>
>> >>> Date: Tue, 12 Sep 2023 12:41:32 +0800
>> >>> From:  Yang Yingchao via "Bug reports for GNU Emacs,
>> >>>  the Swiss army knife of text editors" <bug-gnu-emacs <at> gnu.org>
>> >>>
>> >>> It seams that there is an issue with the behavior of `treesit-beginning-of-defun` function in
>> >>> `c++-ts-mode` when there are static variables present in the source code. For instance, in the
>> >>> provided code snippet, if the cursor is initially placed on the line containing `int b`, and the
>> >>> "C-M-a" is pressed, the cursor is positioned on line 4 instead of line 1 where the beginning of
>> >>> the function is located.
>> >>>
>> >>> ,----
>> >>> | void func		// c-ts-mode: treesit-beginning-of-defun stops in this line,   RIGHT
>> >>> | {
>> >>> |   char * msg;
>> >>> |   static int a; // c++-ts-mode: treesit-beginning-of-defun stops in this line, WRONG
>> >>> |   int b;        // PUT CUSOR HERE
>> >>> | }
>> >>> `----
>> >>>
>> >>> it works in c-ts-mode works normally.
>> >>
>> >> If you want c++-ts-mode to work like c-ts-mode, set
>> >> treesit-defun-tactic to 'top-level'.  By default, we try to support
>> >> nested defuns in C++, but not in C.
>> >
>> >
>> > I apologize, but I realized that I have already asked this question before.
>> >
>> > However, setting 'treesit-defun-tactic' to 'top-level' could cause other
>> > issues, as seen in the following snippets:
>> >
>> >
>> > ,----
>> > | class Name {
>> > | public:
>> > |   Name()
>> > |   {
>> > |     // cursor here
>> > |   }
>> > |
>> > |   void func()
>> > |   {
>> > |     static int a;
>> > |   }
>> > |
>> > | };
>> > `----
>> >
>> > Suppose the cursor is in line 5, and "C-M-a" is pressed:
>> >
>> > - In 'c++-mode', the cursor will be moved to the start of the function 'Name()', which is the expected behavior.
>> > - In 'c++-ts-mode':
>> >   - With 'treesit-defun-tactic' set to 'nested', it behaves the same as in 'c++-mode'.
>> >   - With 'treesit-defun-tactic' set to 'top-level', the cursor is moved to the beginning of
>> > the class, i.e., line 1.
>> >
>> > Is it possible to make 'c++-ts-mode' behave the same as in 'c++-mode'?
>> >
>> > Regards.
>>
>> Maybe I'm wrong, but I would like to suggest reconsidering the solution to the
>> 'begging-of-defun' issue related to the static *variable*. It feels strange to me to fix this
>> issue by changing the 'treesit-defun-tactic', which is meant to apply to a *function*
>> literally...
>
> Yuan, any ideas or suggestions?

Just FYI, I made some modifications to 'treesit--navigate-thing' in my own branch to fix this specific issue:

,----
| 1 file changed, 8 insertions(+)
| lisp/treesit.el | 8 ++++++++
| 
| modified   lisp/treesit.el
| @@ -2092,6 +2092,14 @@ treesit--navigate-thing
|          (pcase-let
|              ((`(,prev ,next ,parent)
|                (treesit--things-around pos regexp pred)))
| +
| +          ;; YYC: Special handling for static variable inside defun...
| +          (when (and (equal (treesit-node-type prev)
| +                            "storage_class_specifier")
| +                     (equal (treesit-node-type (treesit-node-parent prev))
| +                            "declaration"))
| +            (setq prev nil))
| +
|            ;; When PARENT is nil, nested and top-level are the same, if
|            ;; there is a PARENT, make PARENT to be the top-level parent
|            ;; and pretend there is no nested PREV and NEXT.
| 
`----

Previous Next

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