bug#61235: 30.0.50; tree-sit: `treesit-node-check' lacks a way to tell if a node belongs to a deleted parser

[Mickey Petersen <mickey@masteringemacs.org>]

Eli Zaretskii <eliz@gnu.org> writes:

>> From: Mickey Petersen <mickey@masteringemacs.org>
>> Cc: Yuan Fu <casouri@gmail.com>, 61235@debbugs.gnu.org
>> Date: Mon, 06 Feb 2023 12:35:20 +0000
>>
>> > I'm not sure I understand the need.  AFAIU, a parser is deleted only
>> > if we call treesit-parser-delete; are we saying that a Lisp program
>> > doesn't know that it deleted a parser?  What exactly is the practical
>> > situation where this problem happens, and why?
>> >
>> > Frankly, I don't think we should at this stage add APIs without a very
>> > good reason.  We should instead collect experience, both from users
>> > and from Lisp programs, and analyze them before deciding whether more
>> > APIs are necessary.
>> >
>>
>> Because node references are retained even after a parser is deleted.
>>
>> Retrieving a node; somehow deleting the parser (maybe you closed the
>> buffer, or you were doing some off-hand parsing); and then doing
>> _anything_ with the aforementioned node yields an error for which
>> there is no way to test for.
>>
>> This is particularly the case when you mix and match parsers in the
>> same buffer.
>
> I'm asking why the Lisp program cannot track the parsers its uses and
> deletes, and instead expects the core to do the janitor's job for it.

Because I have a proxy-like object of a real node because they're
invalidated if a buffer is edited, even if the parcel of code I hold a
node reference to is untouched. That's just how tree-sitter works, so
I deal with it like this. That part works fine for I can of course use
`treesit-node-check' to determine if it's outdated and thus needs
refreshing (or not.)

The problems begin when the parser is also, for one reason or another,
destroyed.

Now I can no longer determine if it is safe to use the existing (real)
node reference. Any attempt to use a "dangling" node reference
that points to a dead parser throws a nasty error.

Thus, the extension to `treesit-node-check' that can tell if a node
belongs to a dead parser --- something that tree-sitter and the C core
clearly knows about internally.