Re: How to add pseudo vector types

[Yuan Fu]

> On Jul 29, 2021, at 12:21 PM, Eli Zaretskii <eliz@gnu.org> wrote:
> 
>> From: Yuan Fu <casouri@gmail.com>
>> Date: Thu, 29 Jul 2021 11:57:56 -0400
>> Cc: Stephen Leake <stephen_leake@stephe-leake.org>,
>> Clément Pit-Claudel <cpitclaudel@gmail.com>,
>> monnier@iro.umontreal.ca,
>> emacs-devel@gnu.org
>> 
>>>> Anyway, I found a way that avoids this issue: the bounds of tree-sitter’s 
>>>> visible region never changes, and the next time when lisp narrows to a 
>>>> different region, we update tree-sitter’s bound to match that of the 
>>>> narrowing. Here is the latest patch. If the code is not entirely 
>>>> straightforward, I’m happy to add more comment to explain it.
>>> 
>>> I'm not sure we should do this, because it means we second-guess what
>>> the Lisp program calling TS intends to do.  Why should we do that,
>>> instead of leaving it to the Lisp program to DTRT?  And what happens
>>> if our guess is wrong?
>> 
>> I don’t think the current implementation guesses anything. Let me turn 
>> around and ask you what is TRT: if the buffer is xxxAAAxxx, and lisp narrows 
>> to AAA and creates a parser, parser sees AAA; now widen, user inserts BBB in 
>> front of AAA, what do we tell tree-sitter? Nothing changed, or BBB inserted 
>> at the beginning?
> 
> Neither.  We should tell TS that instead of AAA there's now
> xxxBBBAAAxxx, because the narrowing was removed.

This is the common usage that I imagined:

Narrow
Calls tree-sitter (for fontification etc)
Widen

Users edit the buffer

narrow
Calls tree-sitter (for fontification etc)
Widen

Ideally, tree-sitter only sees the narrowed region because everytime it is 
called, the buffer is narrowed. However, tree-sitter doesn’t work that way, it 
needs to be updated when user edits the buffer, when the buffer is widened. If 
your goal is give lisp control of what tree-sitter sees, we can’t just give 
tree-sitter the whole buffer whenever the user makes some change.

> 
>> To where should lisp narrow? BBBAAA, or AAA, or BBB?
> 
> It's the question for the Lisp program, not for the low-level code
> which we are discussing.
> 
> Anyway, you are once again bothered by a scenario that should not
> happen at all: a Lisp program should not call TS first with, then
> without narrowing (or the other way around).  I don't see why such
> situation should happen, and if they do, the Lisp programs which need
> them will have to figure out what to do and how.

Even if lisp always call tree-sitter with narrowing, we still need to update 
tree-sitter when the buffer is widened. Does that make sense?

Yuan