GitLab

It turns out subivision algorithm and KoPathShape
don't go well together. KoPathShape is a pretty expensive
class memory-wise and is not supposed to be created-freed
rapidly.
By removing it, the performance gains is between 2-20x!

While this should work in creating new meshgradient
on new objects. The reverse hasn't been added yet i.e
Manager -> UI

if the meshpatch, in mesharray were backwards (which
I wasn't sure was allowed till I tested pepper.svg from
inkscape), then patch would be rendered outside the
QImage

Maniphest: T13101

Primary reason for this is, that rendering meshpatches
every single time a translation is done, is very
expensive. So, we render it the first time and reuse it.

The way we do it is:

1) First we render the patch in scaled-up version of
   QImage, this helps us with loss of quality which is
   a consequence of `QPainter&` passed in
   `KoMeshGradientBackground::paint` as a scaled version.

2) We render the whole patch on QImage, not just the patch
   inside clipped region.

3) Then we render the QImage on QPainter in user coordinates
   i.e independent from the shape.

NOTE: One thing which has to be taken care of is, when we resize
the shape, the patch is resized as well. Which is one of the
expected behavior. But there is one more behavior, that is
the patches should not be scaled, which can help us with
having only selected portion of the patches in view.

Maniphest: T13101

Maniphest: T13101

Maniphest: T13101

and remove `SvgMeshStop.h/cpp` -- just a blank file
which wasn't included anywhere

We don't check for color variance yet. So, this is based on
hardcoded recursion.

Maniphest: T13101

Maniphest: T13101

Whenever SvgParser is destroyed (which is often). The
pointer is deleted and the data is copied over to a new
one. Hence the need.

Maniphest: T13101

Maniphest: T13101

Maniphest: T13101

Maniphest: T13101