It should match the spec. However, imo, it is nonsense that order: 1.0 and order: 1e1 are invalid. Tab Atkins would probably like to fix one or both notations, see w3c/csswg-drafts#6471 (comment). But it is very low priority so I do not want to push for this change.

What is your opinion? (for the CSS language)

The existing doc is dreadful but has something about it actually. A consequence for <*-dimension> is noted, which refers to <n-dimension>, <ndash-dimension>, etc... which are used in subproductions of <an-b> in :nth-child() and co. For example, :nth-child(1e1) is valid whereas strict conformance only allows :nth-child(1). :nth-child(1e1n) is a bit weird so I would not mind restricting it to an "absolute" integer.

Another example is <zero>. For example, transform: rotate(0.0) is valid whereas strict conformance only allows transform: rotate(0). edit: actually the spec wants <number-token> whose value is 0 so 0e1, 0.0, etc are valid so this example is irrelevant.

it is nonsense that order: 1.0 and order: 1e1 are invalid

I agree. This also leads to weird discontinuities when using math expressions.

10e3 vs. calc(10e3)

https://codepen.io/romainmenke/pen/GRLwLQy

Do you know if there already is a dedicated issue for specifically this aspect?
(Setting type integer when the actual value is an integer.)

There are not any, I think, except the one I linked to in the previous comment and is only about the scientific notation.

I think <integer> only exists for some corner cases of the CSS grammar. There should be a default rounding behavior in properties like order, z-index, etc.

Another argument is that Number.isInteger(1.0) is truthy, probably like in most if not all progaming languages.

I've filed a new issue here : w3c/csswg-drafts#10238