Re: Sticky tooltips

[Arthur Miller]

Eli Zaretskii <eliz@gnu.org> writes:

>> From: Arthur Miller <arthur.miller@live.com>
>> Cc: emacs-devel@gnu.org
>> Date: Tue, 29 Sep 2020 05:36:29 +0200
>> 
>> > You propose to switch to "tooltips" that pop up in a special buffer?
>> I was mostly proposing for "pure Emacs" frame, special buffer or not, I
>> don't know. Why not?
>> 
>> That makes implementing "sticky" tooltip as someone suggested quite
>> trivial.
>> 
>> There is only one tooltip active at a time (I think), so there is just
>> one buffer/frame needed; and tooltip text could be erased/set when a
>> tooltip is shown. Would that be too inneificient?
>
> You are, in effect, describing the way our native tooltips are
> implemented: they are special small frames.  See tooltip.el.

Haha; typical me :-) 

I took looked in tooltip.el thanks. It works really nice, and it
switches flawlesly between gtk/native tooltips; I am impressed.

Yeah, you are correct, what tooltip already does, as you say is
logically what I described, but the implementation is different:
tooltip.el seems to outsource all it's work to xfns.c, and the code for
tooltip frame creation; I used just "a naked frame" and some Lisp to
create an effect of a tooltip.

By the way; "tooltip-show" (in tooltip.el) seems to take a string
argument; rather than a buffer, and that widget library "widget.el",
needs a buffer (for the minor mode, etc). Maybe it can be done
somehow, I don't know; however I am not so sure if "sticky tooltip" is
that good idea; echoing to message buffer is maybe good enough for most
people? 

Just as a side curious question: did tooltips really needed own c
implementation? Isn't code in x-create-tip-frame and x-show-tip seems
redundant to other frame creation routines/display? Couldn't tooltips be
implemented using normal frames, with "tooltip-special" code written in
lisp? 

I dont' mean to ditch it away, just a reflection on complexity. Maybe I
don't see all the complexity with tooltips; but it is still ~900+ lines
of c code that need to be maintained for X11, and there are well w32 and
ns implementations, so around 3k locs? 

Maybe special-buffers is not a bad idea? Just thinking out of the box :-).