bug#66183: elixir-ts-mode test failure

[Wilhelm Kirschbaum]

Yuan Fu <casouri@gmail.com> writes:

> > On Sep 24, 2023, at 9:53 AM, john muhl via Bug reports for GNU
> > Emacs, the Swiss army knife of text editors 
> > <bug-gnu-emacs@gnu.org>
> > wrote:
> >
> > The change in 5cba5ee8905 caused an elixir test to start 
> > failing. The
> > white space at the start of the test is now in a heex range so 
> > pressing
> > tab there matches a heex rule instead of the elixir one. Moving 
> > the
> > dummy range to the end fixes the test.
>
> Thank you. I’ll make sure to remember also running elixir tests 
> when I make a change.
>
> I looked around and found elixir-mode’s language-at-point 
> function
> works in a way that isn’t what tree-sitter functions
> expect. Tree-sitter functions expect the major mode to derive 
> language
> at point by querying the host language and do some pattern 
> matching,
> rather than using language parser’s range. I took a quick look 
> at
> elixir’s grammar and came up with this version for a POC. You 
> probably

Sorry about that.  I had absolutely no idea how it was expected to 
work.

> need to adjust it but the idea is there.
>
> With the new language-at-point definition, treesit-language-at 
> can
> return the correct language, and the test passes again.
>
> Yuan
>
> [2. text/x-patch; example.diff]...

This seems to work if you set named on: treesit-node-prev-sibling,
otherwise it just tries to match the \"\"\" against H or F.

  (let* ((node (treesit-node-at point 'elixir)))
    (if (and (equal (treesit-node-type node) "quoted_content")
             (string-match-p
              (rx bos (or "H" "F") eos)
-              (treesit-node-text (treesit-node-prev-sibling 
              node))))
+              (treesit-node-text (treesit-node-prev-sibling node 
t))))
        'heex
      'elixir)))

I will spend some time in the next couple of hours to catch up to 
all
the changes.

Wilhelm