Re: [css3-flexbox] resolving flexible lengths

On Fri, Sep 30, 2011 at 9:21 AM, Tab Atkins Jr. <jackalmage@gmail.com>wrote:

> On Thu, Sep 29, 2011 at 8:39 PM, Ojan Vafai <ojan@chromium.org> wrote:
> > There are a couple issues with
> > the http://dev.w3.org/csswg/css3-flexbox/#resolving-flexible-lengths
>  step.
> > 1. It's not clear that the preferred size ignores min/max constraints. I
> > think the spec should explicitly state that it ignores min/max
> constraints
> > unless the preferred size is set to auto, in which case it takes them
> into
> > account.
> > The use-case this allows is the following:
> > <div style="display: flexbox; width: 200px">
> >     <div style="width: flex(1); min-width: 100px;"></div>
> >     <div style="width: flex(3);"></div>
> > </div>
> > Clearly you would expect the first flex item to be 100px wide and the
> second
> > to be 300px. If the preferred width took min-width into account you would
> > instead get 125px and 75px.
>
> I assume you mean that we would expect both of them to be 100px wide,
> right?
>
> 1. They both start at 0px (their preferred size).
> 2. They are set to 50px and 150px, respectively.
> 3. The first is reset to 100px and is now non-flexible.  Restart the
> algorithm.
> 4. The second is the only flexible box, so it gets all of the free
> space and is set to 100px.
>
> I do need to be clear about when min/max is taken into account, and
> when it's *not*.


The spec needs to spell out how to compute the preferred/hypothetical size.
WebKit's current implementation clearly does the wrong thing here and was
modelled after the spec text. I didn't write this part of the code, but when
I was comparing the spec text to WebKit's implementation, it took me a while
to realize what didn't match the spec.

"First, set all the flexible lengths to their preferred size."

"Pretend that the flexbox is display:block, and still establishes a BFC.
Pretend that the flexbox item is the only child of the flexbox (and also
establishes a BFC). Resolve flexible widths/heights into their preferred
sizes. Resolve ‘auto’ widths/heights by shrinkwrapping them. Using all this
pretend knowledge, resolve the width and height."

If the preferred size is a percentage or auto, then we need to compute the
preferred size. When doing this computation, we should take min/max
constraints into account, no?

Ojan