rendering regression on filtered transformed objects

[jimmac]

Attached is a side by side rendering of an SVG which is also included. On
the left side there's an SVN build (from Oct 21st), on the right hand side
a superior rendering of 0.45.1.

I think the core of the issue are filtered objects (they are 'linked
offsets' to be able to colorize them differently but follow the shape of
the original). The whole helmet is a group and has been sqished and rotated
during the process of creation. You can see the linked offset objects fly
back to their position when you ungroup.

[jimmac (jimmac)]

Originator: YES

File Added: gimp-news-development.svg

[Buliabyak-users (buliabyak-users)]

Originator: NO

jimmac, screenshots are nice but please always attach source svg!

[Niko Kiirala (kiirala)]

Originator: NO

I feel that calling the 0.45.1 rendering 'superior' is a bit of an
overstatement. But yes, the regression is there nevertheless. The code that
handles rotation and skewing in filters has gone through big changes
between 0.45.1 and SVN head.

It seems that you've bumped into especially bad combination of squishing
and rotation with this, too.

The good solution would be to implement a pixblock transformer with either
bilinear or bicubic filtering in Inkscape. The fast solution would be to
just increase the resolution used for images inside the filters rendering
code, when applicable. The latter is pretty much, how 0.45.1 archieves this
better result.

[Tom Davidson (tjd-mit)]

See the dupes (left column) for some more dramatic examples of how bad this can look...

[Niko Kiirala (kiirala)]

Fixed in SVN revision 16938, using a raster image transformer with bicubic filtering.