Re: [Aquamacs-bugs] Re: incorrect fontification of non-ascii chars on Mac OS X 10.4

[David Reitter]

On 30 Jun 2005, at 11:40, YAMAMOTO Mitsuharu wrote:
> I'm not aware of that (I only define one fontset on startup).
> Removing the Foptimize_char_table call in Fset_fontset_font
> (fontset.c) seems to improve the speed at least for this case.

I removed this (and the checks etc.) and it does improve the speed  
significantly.
I'm down to maybe 2 seconds for those 25 fontsets. However, this is  
still too long - especially considered we get more fontsets in the  
future.

>  What should trigger fontset creations?  If it is restricted to user
> commands such as menu item selection, then such lazy creation could be
> implemented at the user level.

I suppose one would need to create a fontset whenever it is needed.  
That means that if some mode fontifies something, or if the frame  
font changes (programmatically or interactively), it should be  
created. I wouldn't know how to do this at this point.

An alternative would probably be if we gave users the option to  
'activate a font', but offer a library of fontsets.
I assume that wouldn't be too hard: we scan the font list, extract  
the font names and properties, and user can activate a font.
Upon starting, we can create fontsets for only the activated (or  
'loaded') fonts.

Obviously, this is not a quick solution - so right now I'm struggling  
with a situation where a simple bugfix (to support bold umlauts,  
etc.) means an extra 2-second delay on startup.

Can the resulting table, that is, the whole fontset, be serialized  
and byte-compiled?

> I think that's because the current Emacs internal code, aka
> emacs-mule, is created so that it matches the X11 font system.  You
> don't want to change the Emacs internal code for the Carbon port, do
> you?

It's probably smart to wait for unicode support. But at that point it  
would make sense to be able to sidestep Mule and use the system's  
input and text display (=font selection) methods. And that applies to  
all ports where the underlying systems offer good inputs methods.

Maybe it would help if someone (more knowledgeable than me) took a  
good look at how it's done in the Emacs-on-Aqua port.