Re: prettify-symbols-mode, derived modes, and compose-region

[Eli Zaretskii]

> From: D <d.williams@posteo.net>
> Cc: emacs-devel@gnu.org
> Date: Fri, 5 Mar 2021 22:24:44 +0100
> 
> > And that's even before you consider
> > complex text shaping which could combine several codepoints into a
> > single wide grapheme cluster.
> 
> I don't think that using a character capable of that would be of any
> use as a decorative element.

I was talking about the need to align the "decorative elements" with
text that includes such grapheme clusters.

> > Anyway, I think it should be very easy to add a new display property
> > type that would override the pixel width of a font glyph with a fixed
> > value calculated in some way by a Lisp program.
> 
> That would of course be the perfect solution.  But, doesn't this have
> to exist in one way or another already?  For example, the checkbox
> widget in custom interfaces to my knowledge visually replaces a simple
> "[X]" with a bitmap.  When I examine a checkbox with C-u C-x = it
> seems to be the case that display does the legwork and somehow obtains
> the image size from the xpm file (I suppose).

In your original use case you didn't have images, you had characters.
Or did I misunderstand?

> Same with superstar, essentially.  For example, every decorative
> bullet can have a "fallback character" for terminal displays, and this
> behaves correctly from the perspective of the frame you opened the
> buffer with.  I include a simple redisplay command for emacsclient
> users for this reason.  So I suppose I operate under the same
> limitations as shr?

Yes, I think so.  And on text terminals this issue is moot anyway,
because every character takes a single column (if you forget about
those rare ones that take 2).

> In that case, I probably should look into that character size
> function.  Could you point me in the right general direction?  I'm not
> quite sure how to find it.

shr-fill-line, I guess.

You can also try using font-get-glyphs, it might be more suitable for
your needs.  I think shr.el doesn't use it because that turned out to
be slower than the current approach, but your use case is different.