Frame background mode

Frame background mode

[Juri Linkov]

I suggest to improve the formula used to determine whether the
background is light or dark.  The new formula uses the optimal
weighting coefficients for RGB color components to reflect the
perception of color luminance.

Index: lisp/faces.el
===================================================================
RCS file: /cvsroot/emacs/emacs/lisp/faces.el,v
retrieving revision 1.297
diff -u -r1.297 faces.el
--- lisp/faces.el	23 Dec 2004 18:28:13 -0000	1.297
+++ lisp/faces.el	26 Dec 2004 18:44:06 -0000
@@ -1536,11 +1536,15 @@
 		 'dark)
 		((equal bg-color "unspecified-fg") ; inverted colors
 		 'light)
-		((>= (apply '+ (x-color-values bg-color frame))
-		    ;; Just looking at the screen, colors whose
-		    ;; values add up to .6 of the white total
-		    ;; still look dark to me.
-		    (* (apply '+ (x-color-values "white" frame)) .6))
+		((let ((bg-color-values (x-color-values bg-color frame))
+		       (white-values (x-color-values "white" frame)))
+		   (>= (+ (* (nth 0 bg-color-values) 0.30)
+			  (* (nth 1 bg-color-values) 0.59)
+			  (* (nth 2 bg-color-values) 0.11))
+		       (* (+ (* (nth 0 white-values) 0.30)
+			     (* (nth 1 white-values) 0.59)
+			     (* (nth 2 white-values) 0.11))
+			  .5)))
 		 'light)
 		(t 'dark)))
 	 (display-type

The most noticeable consequence of such change is that many green
colors (which really look like light colors) are moved to `light'
category.

Here is the code that displays the color changes between old and new
formulas (it needs the patch to facemenu.el I sent earlier):

(let ((frame (selected-frame)) diff-list new-dark-list new-light-list)
  (dolist (color (list-colors-duplicates))
    (let* ((bg-color (car color))
	   (old (if (>= (apply '+ (x-color-values bg-color frame))
			(* (apply '+ (x-color-values "white" frame)) .6))
		    'light 'dark))
	   (new (if (let ((bg-color-values (x-color-values bg-color frame))
			  (white-values (x-color-values "white" frame)))
		      (>= (+ (* (nth 0 bg-color-values) 0.30)
			     (* (nth 1 bg-color-values) 0.59)
			     (* (nth 2 bg-color-values) 0.11))
			  (* (+ (* (nth 0 white-values) 0.30)
				(* (nth 1 white-values) 0.59)
				(* (nth 2 white-values) 0.11))
			     .5)))
		    'light 'dark)))
      (if (eq new 'light)
	  (setq new-light-list (cons color new-light-list))
	(setq new-dark-list (cons color new-dark-list)))
      (unless (eq old new)
	(insert (format "%-22s %-5s %s\n" bg-color old new))
	(setq diff-list (cons color diff-list)))))
  (list-colors-display (reverse diff-list) "*Colors-Diff*")
  (list-colors-display (reverse new-dark-list) "*Colors-Dark*")
  (list-colors-display (reverse new-light-list) "*Colors-Light*"))

The changes in color mode it computes are the following (the second
column shows the mode computed by the old formula, the third column
by the new formula):

light slate gray       dark  light
medium slate blue      light dark
deep sky blue          dark  light
dark turquoise         dark  light
cadet blue             dark  light
medium sea green       dark  light
light sea green        dark  light
spring green           dark  light
lawn green             dark  light
green                  dark  light
chartreuse             dark  light
medium spring green    dark  light
lime green             dark  light
yellow green           dark  light
goldenrod              dark  light
dark goldenrod         dark  light
peru                   dark  light
chocolate              dark  light
orange                 dark  light
dark orange            dark  light
tomato                 dark  light
magenta                light dark
snow4                  dark  light
seashell4              dark  light
AntiqueWhite4          dark  light
LemonChiffon4          dark  light
cornsilk4              dark  light
ivory4                 dark  light
honeydew4              dark  light
LavenderBlush4         dark  light
MistyRose4             dark  light
azure4                 dark  light
SlateBlue2             light dark
SteelBlue3             dark  light
DeepSkyBlue1           dark  light
DeepSkyBlue2           dark  light
LightCyan4             dark  light
PaleTurquoise4         dark  light
turquoise3             dark  light
cyan3                  dark  light
SeaGreen3              dark  light
PaleGreen3             dark  light
SpringGreen1           dark  light
SpringGreen2           dark  light
SpringGreen3           dark  light
green1                 dark  light
green2                 dark  light
chartreuse1            dark  light
chartreuse2            dark  light
chartreuse3            dark  light
OliveDrab3             dark  light
DarkOliveGreen3        dark  light
khaki4                 dark  light
LightYellow4           dark  light
yellow3                dark  light
gold2                  dark  light
gold3                  dark  light
goldenrod2             dark  light
goldenrod3             dark  light
DarkGoldenrod1         dark  light
DarkGoldenrod2         dark  light
DarkGoldenrod3         dark  light
IndianRed2             dark  light
sienna1                dark  light
sienna2                dark  light
sienna3                dark  light
tan3                   dark  light
chocolate1             dark  light
chocolate2             dark  light
salmon3                dark  light
LightSalmon3           dark  light
orange1                dark  light
orange2                dark  light
orange3                dark  light
DarkOrange1            dark  light
DarkOrange2            dark  light
coral1                 dark  light
coral2                 dark  light
tomato1                dark  light
tomato2                dark  light
HotPink3               dark  light
PaleVioletRed3         dark  light
maroon1                light dark
magenta1               light dark
magenta2               light dark
MediumOrchid3          light dark
DarkOrchid1            light dark
DarkOrchid2            light dark
thistle4               dark  light
gray51                 dark  light
gray52                 dark  light
gray53                 dark  light
gray54                 dark  light
gray55                 dark  light
gray56                 dark  light
gray57                 dark  light
gray58                 dark  light
gray59                 dark  light

-- 
Juri Linkov
http://www.jurta.org/emacs/

Re: Frame background mode

[Eli Zaretskii]

> From: Juri Linkov <juri@jurta.org>
> Date: Sun, 26 Dec 2004 21:52:54 +0200
> 
> I suggest to improve the formula used to determine whether the
> background is light or dark.

Do we really need this so close to a pretest start?  The new formula
will certainly change some backgrounds from dark to light and vice
versa, which will potentially annoy some users whose color
customizations will need to be reworked.

RE: Frame background mode

[Drew Adams]

    > I suggest to improve the formula used to determine whether the
    > background is light or dark.

    Do we really need this so close to a pretest start?

Dunno. Maybe not.

    The new formula will certainly change some backgrounds from dark
    to light and vice versa, which will potentially annoy some users
    whose color customizations will need to be reworked.

That argument seems independent of _when_ such a change might be made; it
does not argue for not making the change _now_.

I think Juri is on the right track with this change. And experimentation
could help refine the formula.

Re: Frame background mode

[Juri Linkov]

"Eli Zaretskii" <eliz@gnu.org> writes:
> The new formula will certainly change some backgrounds from dark to
> light and vice versa,

The new formula will affect only users who have background color set
to one of 98 colors I listed in the previous message.

> which will potentially annoy some users whose color customizations
> will need to be reworked.

No color customization needs to be reworked.  The only change needed
for users affected by the new formula is to customize frame-background-mode
user option.

If even such small change is undesirable then maybe it's better to
leave the old formula alone, and to add a new condition keyword
(e.g. `min-background') that will specify the minimum color intensity
(a fraction between 0 and 1) and will use the new formula, for instance:

    ((class color) (min-colors 88) (min-background 0.5))

-- 
Juri Linkov
http://www.jurta.org/emacs/

Re: Frame background mode

[Eli Zaretskii]

> From: "Drew Adams" <drew.adams@oracle.com>
> Date: Mon, 27 Dec 2004 09:36:20 -0800
> Cc: emacs-devel@gnu.org
> 
>     The new formula will certainly change some backgrounds from dark
>     to light and vice versa, which will potentially annoy some users
>     whose color customizations will need to be reworked.
> 
> That argument seems independent of _when_ such a change might be made; it
> does not argue for not making the change _now_.

Yes, it does: changing color defaults tends to prompt a long series of
patches, critique to those patches, discussions, counter-patches, etc.
That is a wrong way to get closer to the pretest.

> I think Juri is on the right track with this change.

I didn't say he wasn't, I just said that it should be postponed.  It's
not a critical issue, so it could wait, IMHO.

Re: Frame background mode

[Eli Zaretskii]

> From: Juri Linkov <juri@jurta.org>
> Date: Mon, 27 Dec 2004 22:15:08 +0200
> Cc: emacs-devel@gnu.org
> 
> The new formula will affect only users who have background color set
> to one of 98 colors I listed in the previous message.

98 colors is not a small number.

> No color customization needs to be reworked.  The only change needed
> for users affected by the new formula is to customize frame-background-mode
> user option.

If they can figure out that this is what they need to do, yes.  I
suspect most users will not know that, and instead will go and change
colors right away.

> If even such small change is undesirable

I don't know if it is undesirable, but Emacs did use the current
recipe since about day one.

> then maybe it's better to
> leave the old formula alone, and to add a new condition keyword
> (e.g. `min-background') that will specify the minimum color intensity
> (a fraction between 0 and 1) and will use the new formula, for instance:
> 
>     ((class color) (min-colors 88) (min-background 0.5))

I'm probably missing something, because I don't see how is this
different from your original suggestion.

Re: Frame background mode

[Juri Linkov]

"Eli Zaretskii" <eliz@gnu.org> writes:
>> then maybe it's better to
>> leave the old formula alone, and to add a new condition keyword
>> (e.g. `min-background') that will specify the minimum color intensity
>> (a fraction between 0 and 1) and will use the new formula, for instance:
>> 
>>     ((class color) (min-colors 88) (min-background 0.5))
>
> I'm probably missing something, because I don't see how is this
> different from your original suggestion.

The original suggestion was to change the formula for computing
conditions (background light) and (background dark) which will
affect most current faces.  

The suggestion above affects none of the current faces and will be
applied for faces that will take advantage of the new `min-background'
keyword.  This idea is not for the incoming release because many faces
need to be changed to use this keyword.

-- 
Juri Linkov
http://www.jurta.org/emacs/

Re: Frame background mode

[Eli Zaretskii]

> Cc: emacs-devel@gnu.org
> From: Juri Linkov <juri@jurta.org>
> Date: Tue, 28 Dec 2004 04:48:06 +0200
> 
> The original suggestion was to change the formula for computing
> conditions (background light) and (background dark) which will
> affect most current faces.  
> 
> The suggestion above affects none of the current faces and will be
> applied for faces that will take advantage of the new `min-background'
> keyword.

If the idea is to add the `min-background' keyword to most of the
faces, the effect will be the same, right?

Re: Frame background mode

[Richard Stallman]

I don't agree with the new classification that your code does.
Several of the colors that it calls "light", I believe are dark.
For instance, light slate gray, cadet blue, dark sea green,
light sea green (!), and gray59 all seem like light colors to me.

I made the test by modifying list-colors-display to display the chosen
color in the background, with foregrounds of white and of black.  That
way I can compare the visibility of both white and black against this
background.  If black is more visible against it, it is a light color.
If white is more visible against it, it is a dark color.

This is with my screen gamma set to 1.5, since it is an LCD screen.
It could be that this LCD screen needs a different gamma value.
I don't know how to tell what is "correct".  (It is a Thinkpad T21.)

How about if you try this exercise on your screen, and tweak the
threshold and see if you can get good results?

Re: Frame background mode

[Luc Teirlinck]

Juri Linkov wrote:

   I suggest to improve the formula used to determine whether the
   background is light or dark.  The new formula uses the optimal
   weighting coefficients for RGB color components to reflect the
   perception of color luminance.

and:

   +		   (>= (+ (* (nth 0 bg-color-values) 0.30)
   +			  (* (nth 1 bg-color-values) 0.59)
   +			  (* (nth 2 bg-color-values) 0.11))

Where did you get the .30 .59 .11 from?

The way colors are perceived, including their brightness, depends on
many things.  Your monitor, your eyes, the background colors you are
currently looking at (background color of your screen, color of your
walls), the kind of light that is present other than the light
produced by your computer (daylight, artificial light, the exact
wavelength distribution of that artificial light, no additional
light...), the intensity and angle of that light and so on.

If you have an abnormal color vision, like me (I am a very abnormal
trichromat) you have to customize your monitor.  Customizing every
single color for every single program you are using is hopeless. 

If I use the default settings of my monitor, the most basic color,
white, 255 255 255, looks to me like light green with some shade of
light blue, in other words, a greenish cyan.  It probably looks
perfectly white to most other people.  (Actually, if I allow direct
sunlight to hit my monitor, it turns white, even for me.)  Green, 
0 255 0 looks tremendously brighter than blue, 0 0 255, which in turn
looks tremendously brighter than red, 255 0 0.  I can _see_ red text
on a black background, but not read it.  Same with cyan on white,
which is actually cyan on lighter and greener cyan.  Same with yellow
on white, which is actually light green on a more bluish light green.

So I have to customize my monitor.  The question is: what strategy do
I need to use to customize it?  What I did is I tried to make red,
green and blue equally bright.  Now, under my usual (day)light
conditions, white looks white and I have less trouble with
unreadability due to bad (for me) color combinations.  If I allow
direct sunlight to hit my monitor, then my de-cyanized white turns
bright rose, but I better not allow that to happen anyway.  Ideally, I
would need different customizations of my monitor for day and night
use, but the difference, while noticeable, is not really big enough to
be worth that trouble.

If I understand correctly the old code assumed that red, green and
blue have exactly the same brightness.  For my monitor, eye and usual
conditions, that is actually correct, because I made it correct.
Again, if I understood correctly, your code assumes that green is, to
some specified degree, brighter than red, which is brighter than blue.
Again, where did you get these numbers from?  What brightness relation
do people who construct monitors _try_ to achieve under which
conditions?

I agree with Eli that it is better not to mess with this until after
the release.

Sincerely,

Luc.

Re: Frame background mode

[Juri Linkov]

Luc Teirlinck <teirllm@dms.auburn.edu> writes:
> Juri Linkov wrote:
>    +		   (>= (+ (* (nth 0 bg-color-values) 0.30)
>    +			  (* (nth 1 bg-color-values) 0.59)
>    +			  (* (nth 2 bg-color-values) 0.11))
>
> Where did you get the .30 .59 .11 from?

When I looked at different colors in the GIMP color selector,
I noticed that the graphical marker changes its color to black
on light colors, and to white on dark colors.  Then I looked
at the GIMP source code for its idea about color intensity and
found that it uses these numbers.

But these coefficients were not invented by GIMP authors, of course.
They are scientifically measured weightings that reflect the relative
sensitivity of human vision to red, green, and blue.  These numbers
work for the average person who didn't change the default relative
brightness of monitor color components.  That is why they are good
for the default formula.

The strongest argument so far to not change the formula now is that it
will change background modes for some users.

-- 
Juri Linkov
http://www.jurta.org/emacs/