GNU bug report logs -
#72388
31.0.50; Use tree-sitter-cuda grammar but with tree-sitter-cpp's font-lock/indentation rules
Previous Next
Full log
View this message in rfc822 format
> On Jan 18, 2025, at 4:42 AM, Ergus <spacibba <at> aol.com> wrote:
>
> On Fri, Jan 17, 2025 at 05:29:00PM -0800, Yuan Fu wrote:
>>
>>
>>> On Jan 17, 2025, at 5:04 PM, Yuan Fu <casouri <at> gmail.com> wrote:
>>>
>>>
>>>
>>>> On Jan 16, 2025, at 11:35 AM, Ergus <spacibba <at> aol.com> wrote:
>>>>
>>>> Hi Yuan:
>>>>
>>>> It has been a long time since this emails.. But today finally I made my
>>>> serious attempt to make a `cuda-ts-mode`.
>>>>
>>>> While the treesit-language-remap-alist helped to reuse an important part
>>>> of the code, there were a few issues I needed to manage manually and
>>>> maybe they could be improved somehow:
>>>>
>>>> 1. The indentation rules still uses the 'cpp prefix, so I had to
>>>> manually tune the code:
>>>>
>>>> ```
>>>> (defun cuda-ts-mode--simple-indent-rules ()
>>>> (let ((cpp-rules (c-ts-mode--simple-indent-rules
>>>> 'cpp c-ts-mode-indent-style)))
>>>> `((cuda . ,(alist-get 'cpp cpp-rules)))))
>>>>
>>>>
>>>> (setq-local treesit-simple-indent-rules
>>>> (cuda-ts-mode--simple-indent-rules))
>>>> ```
>>>>
>>>> In order to manage it properly
>>>>
>>>> 2. I see that font-lock is working, but the fontlock rules check is
>>>> using the cpp parser.
>>>>
>>>> That's because the c-ts-mode--font-lock-settings still receives cpp as input.
>>>> I am trying to find a way to get it from c++ and then change the
>>>> :language cpp
>>>> with
>>>> :language cuda
>>>> because if I call
>>>> (c-ts-mode--font-lock-settings 'cuda) I obviously get an error.
>>>>
>>>> Is there some way to go around these in a cleaner way?
>>>
>>> Thanks Ergus, once you map a language to another, it should be transparent, and you shouldn’t need to change cpp to cuda. It’s most likely a bug somewhere. I’ll fix it.
>>>
>>> Yuan
>>
>> Ok, now using c-ts-mode’s font-lock and indentation rules verbatim should work. BTW, you should have
>>
>> (setq-local treesit-language-remap-alist '((cpp . cuda)))
>>
>> in cuda-ts-mode’s body.
>>
>> Yuan
>>
>
> Hi Yuan:
>
> Very thanks for your answer and quick fix ;)
>
> However when I try this:
>
> ```
> (define-derived-mode cuda-ts-mode c-ts-base-mode "Cuda"
> (when (treesit-ready-p 'cuda)
> (setq-local treesit-primary-parser (treesit-parser-create 'cuda))
> (setq-local treesit-language-remap-alist '((cpp . cuda)))))
> ```
>
> `treesit-simple-indent-rules' and `treesit-font-lock-settings' are nil.
>
> On the other hand:
>
> If I do instead:
>
> ```
> (define-derived-mode cuda-ts-mode c++-ts-mode "Cuda" ...
> ```
>
> Then I have fontlock, but it seems taken from C++ grammar and
> indentation still doesn't work. It actually creates a 'cpp grammar,
> which I think is not desired in this case.
>
> Indentation doesn't work in this case because:
>
> ```
> (treesit-node-language (treesit-node-at (point)))
> ```
>
> returns 'cuda, but `treesit-simple-indent-rules' has 'cpp key; and
> the `treesit-simple-indent' function has:
>
> ```
> (language (treesit-node-language parent))
> (rules (alist-get language treesit-simple-indent-rules))
> ```
>
> The code in `c-ts-mode--simple-indent-rules' ends with:
>
> (pcase mode
> ('c `((c . ,rules)))
> ('cpp `((cpp . ,rules))))
>
> So it can only create 'cpp or 'c keys and `treesit-node-language' uses
> `treesit-parser-language' (which is correct). So I still need to remap
> things manually.
>
> In font-lock I cannot see in such details because the var
> `treesit-font-lock-settings' is more opaque, but I suppose that there
> may be happening something similar.
>
Thanks Ergus. I was trying to avoid using the heavy hammer but it seems we have to make it completely transparent. I pushed a change to master. Now when you create a cuda parser using the remapping from cpp, it should completely look and talk like a cpp parser.
Font-lock should work though. It worked for me when I tested it locally.
Yuan
This bug report was last modified 135 days ago.
Previous Next
GNU bug tracking system
Copyright (C) 1999 Darren O. Benham,
1997,2003 nCipher Corporation Ltd,
1994-97 Ian Jackson.