Re: Tracking buffer positions across time, without markers (was Re: PL support)

[Clément Pit-Claudel]

On 10/05/2020 00.32, Eli Zaretskii wrote:
> On May 10, 2020 6:45:13 AM GMT+03:00, "Clément Pit-Claudel" <address@hidden> 
> wrote:
>> On 09/05/2020 19.26, João Távora wrote:
>>> This is done with an idle timer, much the way that Flycheck
>>> works, I believe. 
>>
>> Yup, it's a longstanding issue in Flycheck ^^
>>
>>>> A bit.  But this works only if the interaction is fast enough,
>> right? (Same for syntax errors: positions may well be outdated by the
>> time the errors come back)
>>
>>> By that time another request is probably already underway.
>>
>> As long as the requests are fast.  With slow checkers (typically,
>> proof assistants and automated reasoning tools, a single query can
>> easily take a minute — by that point lots can have changed in a
>> buffer.
>>
>> The APIs that make this work in Visual Studio (spans and snapshots)
>> are described here:
>> https://docs.microsoft.com/en-us/visualstudio/extensibility/inside-the-editor?view=vs-2019
>>
>> I'd love to hear opinions on what the proper implementation of this
>> would for Emacs be.
> 
> Can you explain why using markers is not a good solution for this?

Of course: in brief, because we don't know where to put the markers.

The general problem is this: I send the contents of a buffer to a subprocess.  
Some time later, the subprocess returns a list of positions (offsets from the 
beginning of the buffer, or line/column pairs, etc).  At that point, these 
positions may be stale, since the contents of the buffer may have changed.  I 
need a way to translate these positions (relative to the old buffer contents) 
into positions relative to the new buffer contents.

Since we don't know beforehand what positions the subprocess will return, I 
don't think markers can help; right?