Re: master d995429e7bc: Use SBYTES instead of strlen in treesit.c

[Stefan Kangas]

Mattias Engdegård <mattias.engdegard@gmail.com> writes:

> 23 juli 2024 kl. 22.42 skrev Stefan Kangas <stefankangas@gmail.com>:
>
>>> We avoid that when they are converted from a sexp (see
>>> treesit_query_string_string).
>>>
>>> Perhaps we should do the same when we get a string?
>
> Tree-sitter is something of which I know very little, but if you mean
> taking care of NULs when we create a Lisp string from what we get from
> tree-sitter, then it doesn't seem to be too broken at a quick glance:

It's not about what we get from tree-sitter, but what we pass to it.

I recently changed calls to ts_node_child_by_field_name and
treesit_query from using strlen for the length argument to using SBYTES.
But that meant that we now pass a length of 7 instead of 3 for Lisp
strings like "abc\^@def".  The question is if tree-sitter will accept
that.

If it doesn't, I proposed that we might want to escape the NULs using
treesit_query_string_string.  I'm struggling to find anything clear in
the tree-sitter documentation about this, but I see your recent change.

The other options are either warning about such Lisp strings when we get
them, or (my least favorite option) just revert back to using strlen.

> (We have too many string constructors in general and several of them
> don't do precisely what the caller thought they would but that's a
> different problem, to be dealt with another day.)

Cleaning that up would be welcome in my book, FWIW.