Re: [RFC PATCH] Per-window face support

[Daniel Colascione]

>> Cc: address@hidden
>> From: Daniel Colascione <address@hidden>
>> Date: Sat, 16 Jun 2018 07:24:13 -0700
>>
>> >>> +    if (w) {
>> >>> +      Lisp_Object found = assq_no_quit (parameter,
>> w->window_parameters);
>> >>> +      if (!NILP (found) && EQ (XCDR (found), value))
>> >>> +        match = true;
>> >>> +    }
>> >>
>> >> [...]
>> >>
>> >> Also, I wonder whether EQ should actually be equal_no_quit, because
>> >> the latter would allow parameters with values that are not just
>> >> integers or symbols.
>> >
>> > I still wonder why we only allow EQ there, it sounds unnecessarily
>> > restrictive.
>>
>> Because there's no need for greater power. All you need is simple
>> classification.
>
> That's not the Emacs way, IMO.  Being able to use only integers or
> symbols as filtering parameters sounds too restrictive to me, and I
> see no reason for such restrictions.

There's no loss of generality in allowing only eq at face filtering time:
you can set an integer or symbol to reflect the result of any other
computation, as I do at
https://github.com/dcolascione/emacs-window-highlight/blob/master/window-highlight.el.

If you want to evaluate a condition more complex than eq can handle, you
arrange to run some code when your condition changes, then store the
result of your comparison in a property that the face filter can inspect
with eq. Anything you can do by allowing equal in filters, you can do
using this technique.

 Also, the code and the
> documentation were clearly written with an eye towards future
> extensions, which seems to contradict "no need for greater power".
>
> But if I'm the only one who's bothered by these restrictions, then so
> be it.

Contractually allowing something more powerful than eq here can really
screw us over later: what happens when we equal-compare large data
structures? What happens when we eventually get custom type-by-type
equality predicates? What happens if we want to cache the result of
applying face filters? The way it is now, with an eq-test, the result of a
face filter operation can change _only_ after a call to
set-window-parameter or an update of a face spec somewhere. I don't want
to give up this invariant for no reason.

Limiting the test to eq greatly simplifies the contract while not reducing
power. _That's_ the Emacs way.