bug#65470: 29.1.50; js-ts-mode: regex pattern can cause incorrect parenthesis matching

["Augustin Chéneau" <btuin@mailo.com>]

[-- Attachment #1: Type: text/plain, Size: 1992 bytes --]

Le 25/08/2023 à 02:18, Dmitry Gutov a écrit :
> On 24/08/2023 22:47, Theodor Thornhill wrote:
>> Eli Zaretskii<eliz@gnu.org>  writes:
>>
>>>> Date: Wed, 23 Aug 2023 16:23:13 +0300
>>>> From: Dmitry Gutov<dmitry@gutov.dev>
>>>>
>>>> On 23/08/2023 12:05, Augustin Chéneau (BTuin) wrote:
>>>>> With the mode js-ts-mode, matching tokens (such as '()', '[]') can be
>>>>> incorrectly paired by `show-paren-mode`. This is trivially 
>>>>> reproducible
>>>>> with this simple example:
>>>>>
>>>>> (/foobar)/)
>>>>>
>>>>> The first parenthesis is matched with the second one, which is 
>>>>> inside a
>>>>> regular expression pattern (between slashes), and the last one is not
>>>>> paired.
>>>>>
>>>>> The behavior should be the same as for string, the content of the 
>>>>> regex
>>>>> pattern should have no influence on the structure of the code. The 
>>>>> first
>>>>> parenthesis should match with the third one. Here, the first 
>>>>> parenthesis
>>>>> is matched with the last one:
>>>>>
>>>>> ("foobar)")
>>>>>
>>>>> js-mode behaves correctly in both cases.
>>>> Sounds like js-ts-mode also needs a syntax-propertize-function, similar
>>>> to c-ts-mode, ruby-ts-mode and rust-ts-mode.
>>>>
>>>> Others (typescript-ts-mode?) probably need it as well, at least modes
>>>> for those languages that have dedicated regexp or heredoc syntax.
>>> Would someone please add syntax-propertize-function in modes that need
>>> it?  I think this should be done on the emacs-29 branch.
>>>
>>> TIA
>> I'll add it to my list, but if someone will grab it that's fine, as I'm
>> a little short on time the next couple of weeks 🙁
> 
> This one seems to work for js-ts-mode.
> 
> typescript is a bit more fiddly (two separate modes, one with jsx and 
> one without), but should be able to follow the similar pattern.

Thanks for the patch, but I still have some issues. I believe that the
START and END arguments of `put-text-property` are off by one. The
attached modified patch seems to work better.

[-- Attachment #2: js-ts--syntax-propertize-v2.diff --]
[-- Type: text/x-patch, Size: 1745 bytes --]

diff --git a/lisp/progmodes/js.el b/lisp/progmodes/js.el
index 9d2990e7bc9..a74f67d4466 100644
--- a/lisp/progmodes/js.el
+++ b/lisp/progmodes/js.el
@@ -3829,6 +3829,7 @@ js-ts-mode
                 (append "{}():;,<>/" electric-indent-chars)) ;FIXME: js2-mode adds "[]*".
     (setq-local electric-layout-rules
 	        '((?\; . after) (?\{ . after) (?\} . before)))
+    (setq-local syntax-propertize-function #'js-ts--syntax-propertize)
 
     ;; Tree-sitter setup.
     (treesit-parser-create 'javascript)
@@ -3864,6 +3865,26 @@ js-ts-mode
     (add-to-list 'auto-mode-alist
                  '("\\(\\.js[mx]\\|\\.har\\)\\'" . js-ts-mode))))
 
+(defvar js-ts--s-p-query
+  (when (treesit-available-p)
+    (treesit-query-compile 'javascript
+                           '(((regex pattern: (regex_pattern) @regexp))
+                             ((variable_declarator value: (jsx_element) @jsx))
+                             ((assignment_expression right: (jsx_element) @jsx))))))
+
+(defun js-ts--syntax-propertize (beg end)
+  (let ((captures (treesit-query-capture 'javascript js-ts--s-p-query beg end)))
+    (pcase-dolist (`(,name . ,node) captures)
+      (let ((syntax (pcase-exhaustive name
+                      ('regexp
+                       (string-to-syntax "\"/"))
+                      ('jsx
+                       (string-to-syntax "|")))))
+        (put-text-property (1- (treesit-node-start node)) (treesit-node-start node)
+                           'syntax-table syntax)
+        (put-text-property (treesit-node-end node) (1+ (treesit-node-end node))
+                           'syntax-table syntax)))))
+
 ;;;###autoload
 (define-derived-mode js-json-mode js-mode "JSON"
   (setq-local js-enabled-frameworks nil)