bug#59498: 29.0.50; c++-ts-mode get wrong-type-argument error when enabled

[Yuan Fu]

> On Nov 26, 2022, at 11:18 PM, Stefan Monnier <monnier@iro.umontreal.ca> wrote:
> 
>>> I added treesit-comment-start/end to help indenting comments. So this is
>>> the correct way to use them. The following comment explains why I created
>>> new variables:
>>> 
>>> ;; `comment-start' and `comment-end' assume there is only one type of
>>> ;; comment, and that the comment spans only one line.  So they are not
>>> ;; sufficient for our purpose.
>> 
>> ??? This is surprisingly unclean, IMO.  For starters, the names of the
>> variables are confusing.  The need to define two sets of comment-start and
>> comment-end regexps is also a nuisance and a source of errors.
>> 
>> How do non-treesit modes handle this issue?  Why do the treesit-based modes
>> need something special here?
>> 
>> Stefan, any ideas?
> 
> `comment-start` and `comment-end` do not describe the set of possible
> comment delimiters.  They describe the comment delimiters that should be
> *inserted* when we do things like `comment-dwim`.
> 
> To find/match comment delimiters we have `comment-start-skip` and
> `comment-end-skip`.  They're not ideal, but they've been good enough so far.
> They don't say which comment starter matches which comment-ender (that
> was done by the syntax-tables), but tree-sitter should be able to tell
> us that when we need it.

Ah, I should’ve done more research, sorry. comment-start/end-skip can 
completely replace treesit-comment-start/end.

> It would be nice if we could avoid the need to set/use
> `comment-start-skip` and `comment-end-skip` when using tree-sitter.
> Maybe we can compute their values from the tree-sitter grammar.
> But getting rid of uses of those vars will take a fair bit more work,
> I think.

Tree-sitter puts the whole comment in a single “comment” node, so there is no 
hope getting comment-start/end from it, sadly.

Yuan