bug#67036: 30.0.50; treesit-forward-sexp not working properly in ruby-ts-mode

[Juri Linkov]

>>>>>>>     +  # from "elsif" and "then" C-M-f should jump to next 
>>>>>>> "elsif"/"else"
>>>>>>>       if a == 2 then
>>>>>>>         puts "hello"
>>>>>>>       elsif a == 3
>>>>>
>>>>> Try out the change referenced above, but it doesn't do exactly
>>>>> this. Because the tree-sitter parse tree doesn't match the intuition you
>>>>> described above.
>>>> Thanks for adding "then" and "else" that works not bad.
>>>> Then "elsif" could be added as well.
>>>
>>> You can try it and report back. My impression is that it made navigation
>>> worse: the "elsif" nodes are not siblings of one another, they are
>>> nested. So it doesn't exactly match your expectations.
>> Now I tried and indeed it does wrong thing: jumps to the end of the
>> last "else" instead of the end of own block because they are nested.
>> But maybe possible to skip "condition" and jump to the end of its
>> "consequence" before "alternative"?
>
> With custom code? Like with other questions, I'm not sure where to plug it
> in.
>
> I guess it's possible to set up a ruby-ts-mode specific wrapper for
> treesit-forward-sexp, but that may not be the most optimal way to do
> that. Anyway, I haven't studied this direction yet.

Shouldn't treesit-forward-sexp provide a way for ts-modes to override
the default handling?  By default for all ts-modes such a hook could
check if point is inside strings/comments then to use syntax navigation.
And ruby-ts-mode could use such a hook to implement exceptions like
handling 'else'.

>>>>> Also, interactive forward-sexp never reports "No next sexp" when inside
>>>>> parens or begin...end. It will do forward-up-list instead.
>>>> On the one hand, it's inconsistent with the default non-treesit behavior of
>>>> forward-sexp.  On the other hand, the default behavior is too annoying
>>>> when it screams all the time with "Containing expression ends prematurely!"
>>>> instead of doing something useful.
>>>
>>> Looks like I have been spoiled by Paredit's paredit-forward which catches
>>> such errors and does the appropriate thing. Might be nice to bring this
>>> behavior to Emacs as forward-sexp-command, for example.
>> Agreed, at least as opt-in.
>
> Sounds worth a separate bug-report/feature-request.

Such option would be nice, but still need to investigate here
why "No next sexp" not reported inside begin/class/module.

>>>>>>> +# when point is after @, C-M-f should jump to the end of symbol
>>>>>>>     zzz @abc,
>>>>>>>         4
>>>>>
>>>>> This is something that would need to be changed somewhere inside
>>>>> treesit-forward-sexp (or treesit--navigate-thing). The default 
>>>>> forward-sexp
>>>>> behaves differently when in the middle of a symbol.
>>>> Agreed, more general changes are needed when point is inside symbols,
>>>> strings, comments, etc.
>>>
>>> This is regarding behavior "inside" a thing as understood by
>>> treesit. Unrelated to Emacs's syntactic entities.
>> "Inside a thing" could be handled the same way as "inside
>> strings/comments".
>
> IIUC, treesit knows about the bounds of said thing, it just chooses not to
> move in such situation. That just seems like a bug, not a fundamental
> limitaiton.

This could be fixed like described above.

>>>>>>> +# C-M-f on '[' doesn't jump to after ']'
>>>>>>> +hash['key']
>>>>>>> +
>>>>>
>>>>> As discussed previously, there is no specific node which spans from [ to
>>>>> ]. Some custom code could probably be written (there *are* leaf nodes for 
>>>>> [
>>>>> and ]), but the current capabilities of treesit-thing-settings don't offer
>>>>> a good way to plug that in.
>>>> Like for point inside strings, this might require more general changes
>>>> that take into account syntax tables.
>>>
>>> Possibly, but I expect a solution that doesn't use the syntax table would
>>> be tried first.
>> Since there is no available information from treesit, handling
>> treesit-forward-sexp inside strings/comments could forward to the syntax 
>> table
>> like `prog-fill-reindent-defun' forwards to `fill-paragraph'.
>
> Maybe.

Idem.