Clarification needed in step 5 of the text preparation algorithm

[jfkthame]

The text preparation algorithm described in §4.12.5.1.4 of https://html.spec.whatwg.org/multipage/canvas.html#text-styles says:

5. Form a hypothetical infinitely-wide CSS line box containing a single inline box containing the text text, with its CSS properties set as follows:​
   Property Source
   'direction' direction
   'font' font
   'font-kerning' target's fontKerning
   'font-stretch' target's fontStretch
   'font-variant-caps' target's fontVariantCaps
   'letter-spacing' target's letter spacing
   SVG text-rendering target's textRendering
   'white-space' 'pre'
   'word-spacing' target's word spacing

and with all other properties set to their initial values.

However, this fails to make it clear how the target's fontStretch and fontVariantCaps attributes relate to its font, which (being parsed as a CSS font value) includes font-stretch-css3 and font-variant-css2 components.

Given ctx.font = "condensed 50px Bahnschrift" and ctx.fontStretch = "normal", what is the expected result?

Given ctx.font = "small-caps 50px serif" and ctx.fontVariantCaps = "all-small-caps", what is the expected result?

Given ctx.font = "small-caps 50px serif", does this automatically set ctx.fontVariantCaps to "small-caps"?

Given ctx.font = "50px serif"; ctx.fontVariantCaps = "small-caps", does the value of ctx.font automatically become "small-caps 50px serif"?

Some possible approaches:

(1) font takes precedence: if a stretch and/or caps value is included in the font attribute, it is a fixed part of the font style, and the corresponding separate fontStretch or fontVariantCaps attribute is ignored. So those attributes are only used if the font attribute did not explicitly set a non-normal value.

(2) fontStretch and fontVariantCaps take precedence: these attributes always override whatever may have been set in the font attribute. To avoid breaking existing content that uses e.g. ctx.font = "small-caps 20px serif" and does not mention fontVariantCaps at all, we'll need to define an initial value other than "normal" for these attributes -- something like "unset" or even just an empty string -- so that the font attribute can specify a value and not simply have it overridden by the other attribute's initial value.

(3) Behavior is order-dependent: when the stretch and/or caps value differs between the font attribute and the individual attributes, whichever was set most recently wins. This has two sub-options:

(3a) The context must store the order of setting these attributes, in addition to the current set of values, and save/restore the order along with the values when saving state; or

(3b) Whenever one of these attributes is set, it also automatically updates the other(s) to reflect the new setting, so that the values are consistent with each other. However, the font attribute does not support all the values of fontVariantCaps, so it is unclear to me how this would be managed; there would be some weird asymmetry in the relationship.

Personally, I think either (1) or (2) (or another option not yet proposed...) should be specified The order of setting attributes on the context (3) should not affect the resulting behavior, and neither (3a) nor (3b) are attractive implementation options.

As far as I am aware, currently only the Blink engine implements the fontStretch and fontVariantCaps attributes, and its implementation seems incomplete and confusing (see https://crbug.com/1343549). I think, therefore, that we should not regard this as a model to be adopted, but try to come to a clear resolution that all engines can implement consistently.

[Kaiido]

As a web-dev I think I'd kind of prefer that the behavior of the 2D context matches the CSS one as much as possible.

Would it be an option to add the missing longhands (ctx.fontFamily, ctx.fontSize, ctx.fontStyle, and ctx.fontWeight) and let ctx.font act the same as the CSS shorthand?
So when we update ctx.font all the affected longhands get updated, and when the longhands are updated, ctx.font getter tries to render it or returns "" when it can't, like I believe getComputedStyle(elem).font is supposed to do (though the lack of interop' here is quite problematic, I admit).
If all the longhands are available, I believe ctx.font's getter is less critical and it's ok for it to fail.

---

Ps: cc @whatwg/canvas and particularly @yiyix

[jfkthame]

though the lack of interop' here is quite problematic, I admit

Just FYI, the Firefox bug pointed out in your jsfiddle example (exposing an illegal "shorthand" that includes all-small-caps) has recently been fixed (for Firefox 103) in https://bugzilla.mozilla.org/show_bug.cgi?id=1771215.

[mysteryDate]

As a web-dev I think I'd kind of prefer that the behavior of the 2D context matches the CSS one as much as possible.

+1 (as a former web dev)