unexpected behaviour of FER of line, with blur, and after rotation

attached is a screenshot of the two bboxes

Attaching another sample file which better illustrates where Inkscape goes wrong (compare rendering in a current browser): the filter effects region should actually be rotated along with the rotated object (note that filtered objects always get a preserved transform attribute, and filters ought to be applied before the transformation).

Sample SVG redone with latest trunk.

if I understand these images correctly, then, this looks as if the bbox is actually correct, but it is the filter clipping region that is wrong, because the clipping region has not roteted?

On 2013-09-29 19:38 +0200, Alvin Penner wrote:
> (…) but it is the filter clipping region that is wrong, because the
> clipping region has not roteted?

That's my interpretation, yes - but I could be wrong (I'm not an export wrt to the SVG 1.1. spec, and SVG filter effects). So far though I haven't found an SVG renderer which doesn't rotate the filter effects region for the paths on the right (except librsvg, but that usually is not a reference for Inkscape).

here is what I get as a screenshot using Adobe SVG Viewer 3.0 in IE8. Sounds like this is consistent with what you are getting. I think this would be a decent final result if we could achieve it.

(This is the only independent svg viewer I have, IE9 has its own native svg viewer, but it does not render filters at all, as far as I can tell.)

Attached are the screenshots of
- Squiggle (Batik 1.7),
- Opera 12 (old 'Presto' rendering engine),
- Opera Developer 18.0 (Blink-based webkit engine),
- Chromium (lastet snapshot build 32.0.1656.0 (225895)),
- Firefox 0.22 (two releases behind current 0.24).

Correction (Firefox version):
- - Firefox 0.22 (two releases behind current 0.24).
+ - Firefox 22.0 (two releases behind current 24.0).

… and the screenshot of rsvg.view-3 librsvg (git master @1e22da8).

Maybe related (reproduced with r12650, not with 12583):
1. Draw a straight horizontal line.
2. Apply a filter (blur).
-> the bbox expands horizontally, but is set to zero vertically, and thus the line disappears.

-> the bbox expands horizontally, but is set to zero vertically, and thus the line disappears

this behaviour is related to rev 12621, Bug 1229971. Rev 12621 essentially reversed the changes made in Bug 1188336.

if there is no objection, I would like to rename this to focus on the FER not the bbox. As the above screenshots show, the bbox, which is the horizontally aligned square (dotted line) which encompasses the FER, is actually behaving reasonably. The unexpected behaviour is in the FER after rotation. It should be rotated and clipped, and the clipping is not occurring.