[css-values-4] Should <percentage> be invalid in calculations matching <number>?

[cdoublev]

I suspect that determining the type of <percentage> should be a failure when the grammar does not accept <percentage>, instead of defaulting to [percent: 1]:

• calc(1% / 1%) should be an invalid z-index because its grammar only accepts <number>
• calc(1% / 1% * 1px) should be a valid width because its grammar accepts both <length> and <percentage>, which resolves to a <length> based on the width or height of the containing block

Note: I assume that the type of calc(1% / 1%) as a value for z-index is currently [percent: 0], which matches <number>.

Or am I missing something?

[tabatkins]

No, the current behavior is somewhat intentional; it's similar to how z-index: calc(2em / 10px) is valid despite z-index not accepting lengths.

Note: I assume that the type of calc(1% / 1%) as a value for z-index is currently [percent: 0], which matches .

Correct.

[cdoublev]

Relative lengths can always be resolved with an absolute base in px: font-size of the element, the viewport width, etc.

So <percentage>s in 1% / 1% do not have any absolute base and are not resolved before applying the operation, which results to 1, ok. Thanks.

[cdoublev]

Please note that step 9.4 of simplify a calculation tree requires to resolve 2em / 10px with a value expressed in the result’s canonical unit, whereas this cannot be achieved (at parse time) with 2em. I reported this problem in #7456 but at the time I created it, I was confused by other questions and preferred to close it.

Related to this issue and this problem: should calc(1% / 1% * 1px) (or calc(1em / 1em * 1px)) be resolved to calc(1px)? I would say no because when 1% === 0px, the result is NaN * 1px. It all depends on whether relative numerics must be assumed to be resolved before resolving operations (except sums, eg. 2em - 1em can be resolved to 1em) or math functions (eg. sign(1%)).

Basically, I think products and math functions can be resolved (simplified) when they only include <percentage>s, if they do not resolve to another type (except <number>). They can not when they include relative dimensions.

[Loirooriol]

Related: #9667