Description
Consider:
<!DOCTYPE html>
<html><head>
<meta charset="utf-8" />
<meta http-equiv="X-UA-Compatible" content="IE=Edge" />
<meta name="viewport" content="width=device-width, initial-scale=1" />
<style>
html, body { margin: 0; padding: 0; font-size: 36px; }
html { forced-color-adjust: none; }
body { forced-color-adjust: auto; background: red; }
</style>
</head><body>
</body>
</html>Suppose red is not a system color, then:
- Per https://www.w3.org/TR/css-color-adjust-1/#forced-colors-properties, forced-color adjustment changes used values, not computed values (unless inherited with
forced-color-adjust: preserve-parent-color). Therefore, the computed values follow normal rules. The computedbackground-colorofhtml,bodyaretransparent,red. Moreover, the computedbackground-imageofhtml,bodyare bothnone. According to the adjustment rules, the used background ofbodyisCanvasin this case. - Per https://drafts.csswg.org/css-backgrounds/#body-background, when the computed
background-color,background-imageofhtmlaretransparent,none, the computed background properties ofbodyare propagated tohtml(as if they were "specified"), and the used background ofbodyis initial value, which istransparent. Note that the propagated specified values tohtmlbecome computed values.
We reach a contradiction on the used background of body. The order of propagation and adjustment should be specified.
In addition, regardless of the used background on body:
- The computed background of
htmlbecomesred, and itsforced-color-adjustisnone, so it should be red.
However, MSEdge 136.0.3240.50 (Official build) (64-bit) will show Canvas background. Would you suggest that I file an issue with Chromium? (I don't suggest myself reporting that as a bug.)
I feel the correct definition should be this:
- The background on
bodyis computed, then potentially adjusted, arriving at a used value. - If the computed background on
htmlistransparent, then it acquires the used background frombody(not the computed ofbody) as its own computed value and forcibly resetsbodytotransparentbackground. Potentially, the background ofhtmlundergoes adjustment from its computed value (either propagated or cascaded).
I'll call the propagation condition "html has trivial background". Enumerate the scenarios for my proposed definition:
htmlhas trivialbackground, either or both have adjustment, then the overall result isCanvasbackground.htmlhas trivialbackground, neither has adjustment, then the overall result is developer-specified background frombody.htmlhas non-trivialbackground, then the canvas background directly determined byhtml(consideringhtmlwillingness to adjust).
In all cases, the result is intuitive (and consistent with what MSEdge currently does).
Note that in the proposed definition, in the second step, whether the forcible reset of body to transparent is followed by another potential adjustment on body, is academic and not interesting:
- Per CSS color-adjust-1, upon adjustment, the background color retains its alpha channel. Therefore, the body background simply becomes another transparent color. There is no difference in rendered result.
- Keeping
bodysimplytransparentreenforces the (correct) idea that setting onebackgroundon eitherhtmlorbodysimply sets the canvas background. The range of this background has nothing to do with the bounds ofhtmlorbody.