Re: master 15afa72460b: Fix 'script-representative-chars' for the 'han' script

[Eli Zaretskii <eliz@gnu.org>]

> From: Po Lu <luangruo@yahoo.com>
> Cc: emacs-devel@gnu.org
> Date: Tue, 06 Aug 2024 07:58:54 +0800
> 
> Eli Zaretskii <eliz@gnu.org> writes:
> 
> > If you remove U+2E90, U+2F00, U+3300 and U+3400 from the list and
> > rebuild Emacs, what happens if you insert U+2F75?  Does Emacs succeed
> > to find another font which support that codepoint or does it appear as
> > tofu?  If the latter, what happens if you in install some additional
> > font which does support U+2F75?
> 
> I'll ask, but my intuition is that no font will be discovered, since a
> font must support all of any characters defined as lists in
> script-representative-chars to be eligible.

Note that I said "if you remove those characters".

If you did note that, then does it mean when U+2F75 needs to be
installed and the current font for han doesn't support it, Emacs will
never try to look for _another_ font which supports han characters?
Or will it try, but always fail?

> > IOW, I'm interested to know what happens on GNU/Linux if more than one
> > font is available that together cover both the "usual" han characters
> > and those additional ones which you think we should remove from
> > script-representative-chars, but neither of these fonts supports all
> > of those characters.  Can Emacs solve this by itself on GNU/Linux, or
> > does it need "help" from the user's customization of the fontset?
> 
> Probably the latter, unless `han' is divided into scripts for
> characters, obsolete characters, radicals, and the like.

That is again quite disappointing, since I always thought font
backends based on Fontconfig can do a better job, because (AFAIR)
Fontconfig caches the font information and makes it available for
programs that search fonts covering specific characters.

What you describe happens on MS-Windows, but there we don't have a way
to test whether a font supports a character without actually loading
the font (the 'has_char' backends method always fails).