[css-pseudo] Safari’s ::selection “wash” and UA tweaks to highlight colors

[delan]

Safari has a special approach to painting text ::selection, where it only paints a “wash” on top of the text being selected (see #6022 irc log, “blue wash”, “magic transluscent”). The way this is achieved is interesting: iff the author sets the ::selection background to a fully-opaque color, Safari overrides the alpha value to make the color translucent (demo).

But @fantasai says:

UA cannot do these weird things when the author has specified colors. They have to use the author-specified colors with no modification. But when using UA-chosen colors, they can choose anything.

I don’t think we can allow Safari to paint ::selection as a translucent “wash” and forbid Safari from making opaque ::selection backgrounds translucent, because any author unaware of Safari’s different approach might set the ::selection background to an opaque color (such as a named color or other color without alpha) while expecting the selected text to stay legible.

I want to agree with @fantasai here and forbid all UA tweaks to author highlight colors (see also #6150). But I think I’m biased here, because I also feel like we should forbid the “text wash”, because both of these would make it easier to specify and test highlight painting precisely.

(@smfr, @mrego)

[bradkemper]

I agree, but we should also allow mix-blend-mode (especially with multiply against a light underlying element, and screen against dark), and probably backdrop-filter too. These would produce much better typically Mac-like selections, IMO, than just translucency.

[smfr]

mix-blend-mode and backdrop-filter both have stacking context side-effects. They are not simple painting effects, so we can't make them work for selection.

[fantasai]

If Safari makes opaque colors specified by the author translucent, then it's going to have different highlight colors than e.g. Blink and Gecko. This can, in fact, result in color/background pairs that while readable in one browser are unreadable in another--and this has occurred on real websites in the past.

So either everyone has to apply magic alpha the same way, or there is no magic alpha. We can't have every UA choosing independently on whether to apply alpha magic to author-specified ::selection colors, if we want authors to be able to reliably specify sufficiently-contrasting colors.

[css-meeting-bot]

The CSS Working Group just discussed [css-pseudo] Safari’s ::selection “wash” and UA tweaks to highlight colors, and agreed to the following:

• RESOLVED: If there is color in the author origin the UA must respect that color
• RESOLVED: If there isn't a color from the author origin the UA may apply magic

The full IRC log of that discussion

fantasai: The problem here is that certain UAs make tweaks to highlight color spec by author such as making semi-transparent. That means rendering differs across UAs. I believe FF is only one using colors given as is. This can create contrast problems b/c author can use a color that is fine when semi-transparent but unreadable when opaque
fantasai: There was a bug report on this with white text and white brackground
fantasai: Either we decide we're adding alpha channel magic and use the same magic or we should agree to use colors as specified
q?
fantasai: I don't want every UA to do whatever and they end up unreadable
florian: Desire to match platform is understandable but leads to this where authors use something not interpretable
Rossen_: Yeah, this is a problem when unreadable
fantasai: Default it's fine to do whatever but we need to agree on how the colors are used
s/not interpretable/not interpretable, and thereby unreadable in some casees/
s/white background/white background. Looked great on Safari, was completely unreadable in Firefox./
Rossen_: Anyone have an opinion on if we want to add alpha channel magic or agree on the specific colors? Which path forward?
q+
florian: I suspect what fantasai hinted at which is once color is spec use that it's the easiest path forward. Havign to spec magic I'm not sure if it's doable. Matching native isn't something we can set in stone. Do native magic by default but when a color is set do that
Rossen_: I agree that it would be an uphill battle
GameMaker: Can someone explain the demo page? I see a slider that changes color but I'm not sure what this is trying to show
iank_: If you slide all the way to the right on Safari you'll see the magic alpha applied on Safari
GameMaker: That makes sense
GameMaker: In general against trying to codify this. We work with hi and it might change in 10 years. Letting UAs do what they want when there isn't a color that's great. But when author says I want this color I think we can get behind that
ack GameMaker
Rossen_: Magic described will match the current magic giving the alpha in the last stop on Safari?
q+
Rossen_: The demo page which in Safari adds an alpha channel in the last stop when it's opaque. My question is if we allow for other browsers to match that magic is that the effect we want?
q+
iank_: Proposal is the opposite
florian: Opposite b/c in the demo author has spec the color. If the author hadn't all browsers could do whatever magic they want
ack heycam
heycam: I think there's a question of exact mech to determine author has given a color. What's the value of the selection color going to be by default. A system color keyword?
fantasai: Either system color or not specified other than an internal keyword. I think system colors is most straightforward
heycam: Author can't read because getComputedStyle?
fantasai: There's a compat vs security discussion on there
florian: Worried about round tripping?
how do you access the ::selection getComputedStyle ?
heycam: More about the exact way ofdetermining the author wrote a color
s/getComputedStyle?/getComputedStyle::selection/
florian: As long as it's written you don't have a preference?
heycam: I prefer value of property rather than result of cascade. Seeing if there's a rule with a color property
ack GameMaker
GameMaker: Sounded like we might lean toward defining this and I would say we don't want to define special magic
florian: Not looking for special magic, but looking for when it does and does not apply
GameMaker: Sounds like on same page
Rossen_: Let's look for resolution
fantasai: Here's what's tricky. Spec says each system color computes to corres. color in color space. So at computed time we don't know it's a system color
florian: We would need to do what heycam doesn't prefer
heycam: Do we have other examples of looking at cascade to determine if something was spec?
iank_: Maybe highlight?
florian: I think there's logic around bg and fg. Can we use same logic around having set things?
https://drafts.csswg.org/css-pseudo-4/#paired-defaults
florian: I think 2 part resolution. Part 1: When the author has not specified a color the UA may do what they want. When author has specified a color UA must use as is
florian: Second is investigating what do we consider author has set a color
florian: Maybe we push second part to later?
iank_: Inside of Blink we have a mech to detect if something is spec by an author. We use it rarely, appearance and highlight stuff. I don't know if WK has that
florian: Found what I was talking about. Condition is if bg color yields a cascaded value from author origin. We could use that
florian: Not as simple as heycam wanted
fantasai: But it's something you're already detecting
heycam: If other things use this mech I guess it's okay
Rossen_: Prop: For colors that originate in the...
florian: If there is color in the author origin the UA must respect that color
Rossen_: Objections?
RESOLVED: If there is color in the author origin the UA must respect that color
s/that color/that color exactly, including its alpha/
Rossen_: Do we need to resolve on the magic?
florian: If there isn't a color from the author origin the UA may apply magic
Rossen_: Happy to resolve on that
Rossen_: Objections?
RESOLVED: If there isn't a color from the author origin the UA may apply magic
q+
ack PaulG
PaulG: Was the alpha transparency magic from Safari b/c they were rendering on top of the text? Do we need to add to spec selection is rendered under text?
fantasai: That is in the spec
chrishtr: Does resolution say UA must not apply magic otherwise?
florian: That's the first resolution
fantasai: Spec allows magic that's not controllable by CSS like dimming content. Need to clarify they can't change colors

[fantasai]

Edited. @delan Let me know if the new prose is clear or if it could be written better.

[delan]

Thanks @fantasai!

However if there are no colors in the [=author origin=], the UA may apply additional color processing.

Does this mean setting any -taking property suppresses UA magic in all such properties? For example, if the author sets ‘text-shadow’ on ::selection, does that suppress UA magic in ‘color’ or ‘background-color’?

[fantasai]

@delan That was my interpretation of the resolution; though it does seem a bit odd so maybe we need to tweak it. CC @megangardner

[schenney-chromium]

My recollection of the discussion, after reading the transcript, is that if there are any colors then magic doesn't apply. I kinda thought it was odd at the time but I can see why: As soon as there is one color that is author changed the UA can't know what might happen with the other colors and should take what the author said.

On the other hand, what if you give a foreground color but not a background color, and the browser only has a unified magic way of setting all default highlight colors? What should the browser do for the background? I suspect that is not a real issue in which case this spec text is fine by me.

Otherwise it's probably back to the WG for a resolution saying something about "you must give all colors if you give one", and that seems a stretch.