Re: Adding custom providers for thingatpt.el (was: [PATCH] Add support for 'thing-at-point' to get URL at point)

[Jim Porter]

On 11/6/2023 12:11 PM, Ihor Radchenko wrote:
> [ Branching to emacs-devel for further input from Emacs devs ]
>
> Jim Porter <jporterbugs@gmail.com> writes:
>
> > This is similar to Emacs bug#66752[1]. It would be nice if
> > "(thing-at-point 'url)" would return the URL when point is over an Org
> > link. With this, it's easier to write a function that copies (or browses
> > to) the URL at point without coding so many special cases.
> > ...
> > +(defun org--url-at-point ()
> > +  "`thing-at-point' provider function."
> > +  (org-element-property :raw-link (org-element-context)))
>
> Supporting thingatpt.el is certainly welcome. However, I have some
> doubts about how mature thingatpt.el is.
>
> In particular, I am concerned whether `thing-at-point-provider-alist' is
> reliable enough in non-trivial scenarios like when given URL string is
> not matching some generic URL regexp.

The nice thing about 'thing-at-point-provider-alist' is that your 
provider has absolute control over what to return, so Org's URL provider 
could do whatever it wants. As far as I can tell, this code path 
completely avoids calling 'bounds-of-thing-at-point' ('botap'). However, 
it *would* call 'botap' if point wasn't on an Org link, since it would 
fall back to the last condition in 'thing-at-point'. Still, this is what 
happens today with no provider, so it's not really any worse than before...

Maybe it would make sense for 'thing-at-point' to have a "(catch 
'not-found ...)" form around the loop over 
'thing-at-point-provider-alist'. Then Org could definitively say, 
"There's no URL at point, no matter what anyone else says".

> Looking into the source code of `bounds-of-thing-at-point', I see that
> for standard "things" (like url),
> `thing-at-point-bounds-of-url-at-point' is used unconditionally. In the
> case of Org links, we may have something like [<point>[https://orgmode.org]]
> that will not match default URL regexp as is. AFAIU, there is no
> documented way to customize the behaviour of `bounds-of-thing-at-point'
> and `forward-thing'.

I think it would make sense to add some sort of 
'bounds-of-thing-at-point-provider-alist' (that's a mouthful!) that 
would let modes override the behavior of 'botap', but I don't think 
that's necessary for the narrower purpose of asking, "I want the value 
of THING at point, if any."

> I also have concerns about Org-specific part of the patch, but the above
> is far more important, and we need to discuss it before starting to
> consider anything for Org mode.

For better or worse, I mostly modeled this patch on how EWW integrates 
with thing-at-point, since that's the only place I saw in the Emacs tree 
that did this already.