Re: [css-values] Concrete proposal for unit algebra in calc()

On Wed, Apr 23, 2014 at 7:56 PM, Tab Atkins Jr. <jackalmage@gmail.com> wrote:
> On the telcon today we accepted this proposal for V&U Level 4

That was fast! Yay.

> , but with some changes:
>
> * We should allow negative infinity; at minimum, it's produced by
> subtraction with a finite LHS and infinite LHS, or the negation of an
> infinite value.

If you allow -Inf then you have to decide what +Inf + -Inf means, and
at that point I suspect it's easiest to just adopt the IEEE rules, NaN
and all.  As there are no comparisons in calc(), the further "NaN !=
NaN" headache does not arise.

But it is then also necessary to decide on the behavior of -Inf and
NaN as the final result.  -Inf logically becomes the most negative
allowed value (which in many cases will be 0), but I'm going to
suggest that NaN be treated as *positive* infinity to raise the odds
that someone will notice the math error.

Gecko can reasonably log a console message at computed-value time
whenever a division by zero occurs, but I'm not sure about used-value
time.

> * We'll discuss whether to accept full IEEE semantics; given that
> impls are probably using them for the values of at least some types
> anyway, it might be worthwhile.  (Lengths are generally fixed-width in
> impls, but I dunno how most impls store other types like resolution or
> time.)

I think it's all C 'float' in Gecko, but don't quote me on that.

> * We'll allow unit powers other than 1/0/-1.

So, calc(1px * 2px / 3px) is valid then?  That's probably most
convenient for authors.

> In addition, you suggest that every computation with an infinity
> results in an infinity. Obviously the intent is to make infinities
> infectious, but as suggested it means that invalid unit math doesn't
> invalidate the calc().  I suspect that the infinity should still
> participate in unit-checking?

Err, yeah, that was an oversight on my part.  Infinities should retain
their dimensions.

Also, given what Matt Rakow said earlier today, I think the inverse
length stuff should be dropped.

zw

Received on Thursday, 24 April 2014 00:09:18 UTC