How to deal with custom styles in dark mode?

[xfq]

Description of problem

URL to affected spec or repo:

• Code: Add dark mode support w3c/bp-i18n-specdev#130
• Preview: https://deploy-preview-130--bp-i18n-specdev.netlify.app/

What happened (e.g., it crashed)?:

1. Open https://deploy-preview-130--bp-i18n-specdev.netlify.app/#lang_misc when the system is in light mode.
2. Select 'dark' in the lower left corner of the spec.
3. The colour is incorrect.

If the OS is in dark mode, the colour is correct.

Expected behavior (e.g., it shouldn't crash):

Colours should follow the @media (prefers-color-scheme: dark) media query.

[marcoscaceres]

I think the problem is with .links being applied to the aside (i.e., this is problem with that spec... not ReSpec, as ReSpec doesn't set the style for that)

So that local.css should define

@media (prefers-color-scheme: dark) { .links { /**/} }

[sidvishnoi]

Using the media query won't work with forced dark-mode via the toggle in sidebar (that comes from fixup.js). See w3c@031abec#diff-cefc979f9d55dee149c3b7b5b410cf2d5db0c13d16bf3b379780a1a7ccaab5e2R24-R30 for example.

So, the best strategy would be to use one of the variables from https://www.w3.org/StyleSheets/TR/2021/dark.css (and https://www.w3.org/StyleSheets/TR/2021/base.css) to support consistent dark mode styles (including custom styles). Even with ReSpec's dark mode support, I avoided media query as much as possible and used variables from W3C stylesheet.

[sidvishnoi]

Opinion: I'm not in favor of existence of toggles. The prefers-color-scheme is already the preference set by user that we're after. Adding site-wise color scheme should be something that either the browsers or extensions should do.

[sidvishnoi]

Alternative would be to watch whether dark.css applies using JS (mutation observer on that link element), and then add a class to html and use that instead of (only) media query. But yeah, mostly the pain of having a toggle.

[marcoscaceres]

I like/use the toggle myself... because I don't like the specs on the dark background, even when the rest of the OS is in dark mode... so I get it.

[sidvishnoi]

So maybe fixup.js should add a .dark class when it's using dark mode. Then we can adapt the styles on our end - using media query along with :not(.dark).
Can you file an issue there (tr-design)?

[xfq]

I see. Thanks.

Filed w3c/tr-design#349

[sidvishnoi]

Someday in future we can use style queries to simplify that. Someday...

[xfq]

fixup.js now adds a .darkmode class when it's using dark mode.

[sidvishnoi]

I actually cannot find a way to make it work in case JS is disabled, while avoiding dark style duplication. With duplication, I'm hoping something like following:

.foo {
  /** light mode styles */
}

/** prefers dark mode, and not forced light mode */
@media (prefers-color-scheme: dark) {
  body:not(.lightmode) .foo {
    /** dark mode styles */
  }
}

/** forced JS darkmode */
body.darkmode .foo {
  /** dark mode styles again */
}

@tabatkins any guidance 🙏🏽?

[tabatkins]

The better route, rather than add a class to the body, is to add (or manipulate an existing) <meta name=color-scheme> to have content="light" or content="dark", which'll force the page into light/dark mode and make the MQ match appropriately.

[tabatkins]

(And ideally, generate the spec with content="light dark" so it'll respect the OS setting by default.)

[marcoscaceres]

Hmmm... so, we do generate content="light dark", but we should be checking if the author has included a <meta name=color-scheme> in the document first.

I think we should:

• Check if meta name=color-scheme is present. If no, add it with "light dark".
• Deprecate and remove conf.darkMode

That would solve for this.

[marcoscaceres]

Sent https://github.com/w3c/respec/pull/4700

[sidvishnoi]

@tabatkins Not sure I followed right, but shouldn't following result in forced dark mode?

(Chrome 124 on Mac OS 14)

[tabatkins]

Sigh, I thought it did, but I guess I just tested its effect on the used color scheme, not the MQ. It looks like it does indeed not affect the MQ. I'm gonna raise this in the CSSWG.

[xfq]

Tracked in w3c/csswg-drafts#10249

[xfq]

Any progress on this issue? I assume we are able to solve this in a way that does not depend on changes to CSS specs?