bug#61913: java-ts-mode query error for string highlight, due to recent commit in tree-sitter-java.

[Yuan Fu <casouri@gmail.com>]

> On Mar 3, 2023, at 11:10 PM, Eli Zaretskii <eliz@gnu.org> wrote:
> 
>> From: Yuan Fu <casouri@gmail.com>
>> Date: Fri, 3 Mar 2023 14:32:02 -0800
>> Cc: Eli Zaretskii <eliz@gnu.org>,
>> 61913@debbugs.gnu.org
>> 
>>>>> Error during redeployment: (jit-lock-function 8295) reported
>>>>> (treesit-query-error "Node type error at" 42 "(string_literal)
>>>>> @font-lock-string-face (text_block) @font-lock-string-face"
>>>>> "Debugging the query with `treesit-query-validate'")
>>>>> 
>>>>> 
>>>>> The attached patch fixes the problem. 
>>>> 
>>>> Wouldn't removing text_block from our code cause problems if someone
>>>> uses tree-sitter-java from before the removal?
>>>> 
>>>> Is there a way to write code which handles text_block if it exists?
>>>> 
>> 
>> Tree-sitter language grammars really need some versioning system. I
>> wonder how do we propose such things to them tho...
> 
> That'd be hard, given that many of them don't even make releases.
> 
> It should be enough for our purposes to have a possibility of querying
> the grammar about support for specific features.  Then the version
> will not matter, only the supported features will.  Maybe we could do
> that ourselves, on-the-fly, like we do with programs when we want to
> know whether they support some command-line switch?

We can test whether a node type exists in the grammar, by trying to compile a query using that node type. If it returns successfully, then the node type exists. But there are other assumptions we make about a grammar, like “the 2nd child of a if node must be the condition”. Hopefully these more subtle things don’t change easily, because we can’t easily test them.

Yuan