[css-values-4] Agressively simplifying a Sum node can change the sign of 0

[cdoublev]

#13069 added the simplification of Sum nodes containing an unresolved value and negate nodes.

For example, 1px - (1em + 1px) is simplified to -1em.

But this simplification can change the sign of zero, like in order: sign((-1 * 0px) - (1em - 1em * sibling-count())). If I am not mistaken, with the new simplification requirement, it simplifies to sign(-0px + -1em + (1em * sibling-count()), which resolves to sign(-0px - 1px + 1px) assuming 1em resolves to 1px and sibling-count() to 1. -0px - 1px + 1px results to 0. On the other hand, if it resolves to sign(-0px - 0px), -0px - 0px results to -0px.

0⁻ + 0⁻ or 0⁻ - 0⁺ produces 0⁻. All other additions or subtractions that would produce a zero produce 0⁺.

Note that signed zeros cannot be written in CSS. At least, they are not supposed to.

Also note that ordering calculation terms can have a similar effect.

^{edit: (1em - 1em) simplifies to 0em because units are the same, so added sibling-count() into play.}

[LeaVerou]

Is this author-observable? E.g. if authors do sign(var(--foo) * infinity) where --foo is such an express, can the result ever be a non-zero?
Would it make sense to define sign(-0) as returning a regular 0?

[cdoublev]

Sorry for this bad example. I initially thought the sign of -0 would be -1 instead of -0.

But this is definitely observable in this simpler example:

<div>
  <div id=element></div>
</div>
<script>
  const resolved = getComputedStyle(element)
  element.style.order = 'calc( 1 / ( (-1 * 0) - (1 - sibling-count()) ) )'
  resolved.order; // positive in Chrome, 0 in Firefox (initial value)
  element.style.order = 'calc( 1 / ( (-1 * 0) - (1 - 1) ) )'
  resolved.order; // negative in Chrome and Firefox
</script>

Not sure what is going on in Firefox. Firefox does not support sibling-count() at the moment.