Re: Bounding box of a group with a transformed child

I'm not saying that there aren't potentially large variations between the
cheat and the correct bbox.  Just that in most cases where you are applying
a pattern or gradient, the shape probably has an aspect ration that is not
far from 1:1.

The fact that it appears not a lot of people have complained about it
probably indicates that either:
(a) people don't care
(b) haven't noticed the difference
(c) use userSpaceOnUse coords most of the time

I'm also definitely not arguing that it shouldn't be fixed. :)


On 28 May 2014 00:37, Dr. Olaf Hoffmann <> wrote:

> Paul LeBeau:
> > It seems that pretty much all of the browsers, except FF, cheat and
> return
> > the extent of the transformed-bbox.  Probably because calculating the
> true
> > bbox of some elements (like complicated paths) is a little tricky.
> >  Especially since the bbox needs to be calculated and available before
> any
> > rendering takes place.  And in most cases it doesn't make a lot of
> > difference if it is a little too big.
> >
> If you have for example:
> <g stroke-width="10" stroke="url(#MyGradientUsingBoundingBoxUnits) red">
> <path d="M0,0 1000,1000" transform="rotate(-45)" />
> </g>
> it would be a big difference.
> If you determine the boundingBox before transformation:
> '0 0 1000 1000' and apply the transformation to this rectangle:
> '0 -1000/sqrt(2) 1000*sqrt(2) 1000*sqrt(2)'
> correct:
> '0 0 1000*sqrt(2) 0', this means, that one cannot apply for example
> gradients and patterns with objectBoundingBox units to this - here
> the stroke will be red instead of the gradient.
> Indeed, even for rect elements the difference can be relevant:
> <rect width="100" height="100" rx="50"  transform="rotate(-45)" />
> This is as well a circle, the size in width and height of the boundingBox
> is
> always 100 and not sqrt(2)*100.
> And what about scaling?
> <path d="M0,0 1000,1000" transform="scale(1,-1)" />
> Will this 'cheating' result in '0 0 1000 -1000'? ;o)
> More problems with skewing and general matrix transforms.
> The difference can be of several orders of magnitude - or as
> shown for the gradient example - arbitrary.
> The cases, where the difference is not relevant, seem to be
> the exceptions ;o)
> If one has a good algorithm to determine the boundingBox of an object
> without transformation, the extension to get it right for transformed
> objects
> should be a solvable problem (because one only has to transform all points
> and control points of the equivalent paths of the objects and analyse this
> instead).
> Obviously this is more difficult for the stroke of an object.
> Olaf

Received on Tuesday, 27 May 2014 14:02:48 UTC