[css-typed-om-1] Should css typed om support the same simplifications as css-values-4?

[danielsakhapov]

So, the question is should https://drafts.css-houdini.org/css-typed-om-1/ support the same simplifications (like distributing multiplication over sum and sorting the dimensions alphabetically) as css-values-4? Like https://drafts.csswg.org/css-values-4/#calc-simplification and https://drafts.csswg.org/css-values-4/#sort-a-calculations-children?

[andruud]

Reading the specs, this is what it looks like to me:

• cssom/css-values simplifies during parsing. (E.g. distribution, but not sorting).
• cssom/css-values sorts by unit during serialization.

However:

• css-typed-om does not simplify during "create an internal representation".
• css-typed-om does not sort by unit during reification.

So this means:

• Any internal representation you create from a CSSMathValue is the same thing you get when it's reified back. It's not simplified nor sorted. (Specified values).
• An internal representation produced from parsing, will be simplified (but not sorted), and so it will reify as the simplified structure (again, unsorted).
• An internal representation produced from a CSSMathValue will be sorted when serialized (cssom/css-values), but not simplified.

This is a mess, but I think we can make it more sane if we do one of the following:

• Move all sorting and simplification to parsing, or
• Move all sorting and simplification to serialization. (Maybe better? [css-values-4] Simplification changes semantics #9437)

In any case, preserving CSSMathValue-produced values as-is seems like nice behavior worth maintaining.

cc @tabatkins

[tabatkins]

The behavior is well-defined across the specs. Simplification occurs at parse-time, and when transitioning between some value stages (specified->computed, computed->used).

Typed OM represents a value at one of these stages; whatever the form is at that moment is what it should be reflected as in the Typed OM. If you build one yourself, no simplification happens on the object graph (but once you assign it to a property, the internal value will be simplified as normal).

(The "create an internal representation" algo in Typed OM is a bit too handwavey, and could use some editorial tweaking, tho. I'll untag this from Values and leave it as a Typed OM issue.)

[andruud]

but once you assign it to a property, the internal value will be simplified as normal

OK, it sounds like you're really expecting the creation of an internal value (whether that comes from parsing or typed-om or something else) to invoke simplify a calculation tree. Currently, only parsing invokes that, so we're bypassing it with typed-om (for specified values).

We can currently also bypass sorting if we reify via typed-om and serialize the CSSStyleValue. Is that intentional?

[Mstanic05]

Can you assign this to me!

[emilio]

I agree that the current state of things is a mess. We found this while implementing TypedOM in gecko and would rather sort and simplify always at the same time. @andruud any objections from chromium? There are some tests that currently depend on this behavior.

[andruud]

gecko [...] would rather sort and simplify always at the same time. @andruud any objections from chromium?

No objections to that.