Sudden surge of regressions/changes in (filter) rendering
I just ran the rendering tests again, and compared to last week there were suddenly 16(!) new results. I've had a quick look at some of them and they all seem to be regressions related to filter rendering (affecting quite a lot of different filters). Could anyone who has touched the general filtering code in the last week please recheck their work? (I'm currently swamped with work, so I have no time to go after this myself.)
Jasper van de Gronde wrote:
I just ran the rendering tests again, and compared to last week there were suddenly 16(!) new results. I've had a quick look at some of them and they all seem to be regressions related to filter rendering (affecting quite a lot of different filters). Could anyone who has touched the general filtering code in the last week please recheck their work? (I'm currently swamped with work, so I have no time to go after this myself.)
As always the results of the rendering tests can be found at: http://home.hccnet.nl/th.v.d.gronde/inkscape/ResultViewer.html
-----Original Message----- From: Jasper van de Gronde [mailto:th.v.d.gronde@...528...] Sent: Monday, December 15, 2008 10:40 To: Inkscape Devel List Subject: Re: [Inkscape-devel] Sudden surge of regressions/changes in (filter) rendering (added link)
As always the results of the rendering tests can be found at: http://home.hccnet.nl/th.v.d.gronde/inkscape/ResultViewer.html
I like this page a lot :-) Could you add an explanation next to the category names?
Category Pass (the result corresponds to a pass reference file) Fail (the result corresponds to a fail reference file) New (the result does not correspond to either a pass or fail reference, hence it is new and should be checked ?) No references (there are no reference files for this testcase) Other (?)
I have a question (again, sorry, prbably I forgot) about "new": does it mean it does not correspond to either pass or fail? Wouldn't that just make it a "fail" ? (or did you make it like this in case there is no pass reference?)
Most of all: thanks for this awesome webpage !!
Cheers, Johan
Mon, 15 Dec 2008 10:39:31 +0100 Jasper van de Gronde <th.v.d.gronde@...528...> kirjoitti:
I just ran the rendering tests again, and compared to last week there were suddenly 16(!) new results. I've had a quick look at some of them and they all seem to be regressions related to filter rendering (affecting quite a lot of different filters). Could anyone who has touched the general filtering code in the last week please recheck their work? (I'm currently swamped with work, so I have no time to go after this myself.)
I've worked on the filtering code lately. Still, I'm unable to see any new problems in image export. Could you post some example images of these regressions?
And while I'm at it, let's describe what I've done there: In Inkscape preferences, there is now "Filter Quality" setting. It controls the rendering resolution and resampling settings in filters. As was with the old Blur quality setting, this also affects only the display, export will always use the highest quality.
Highest quality uses full resolution and bicubic resampling while rotating images High quality uses full resolution, but nearest neighbour resampling while rotating Normal quality and below uses limited resolution and nearest neighbour resampling while rotating
There are some problems when editing images at lower resolution. Transparent areas and odd lines are common. (but only in editing mode, export still should be as good as it has been)
Niko Kiirala wrote:
Mon, 15 Dec 2008 10:39:31 +0100 Jasper van de Gronde <th.v.d.gronde@...528...> kirjoitti:
I just ran the rendering tests again, and compared to last week there were suddenly 16(!) new results. I've had a quick look at some of them and they all seem to be regressions related to filter rendering (affecting quite a lot of different filters). Could anyone who has touched the general filtering code in the last week please recheck their work? (I'm currently swamped with work, so I have no time to go after this myself.)
I've worked on the filtering code lately. Still, I'm unable to see any new problems in image export. Could you post some example images of these regressions?
Pretty much any of the tests that use filters among the rendering tests in SVN will do, apparently... But I've attached a specific output I got for the test for bug 181995 (the SVG and reference are in SVN). As you can see the result is pretty weird (colours change and everything).
And while I'm at it, let's describe what I've done there: In Inkscape preferences, there is now "Filter Quality" setting. It controls the rendering resolution and resampling settings in filters. As was with the old Blur quality setting, this also affects only the display, export will always use the highest quality.
Hmm... Weird, the tests obviously use png export, so they shouldn't be affected.
For reference, the tests are located here: https://inkscape.svn.sourceforge.net/svnroot/inkscape/gsoc-testsuite/tester/
Jasper van de Gronde wrote:
And while I'm at it, let's describe what I've done there: In Inkscape preferences, there is now "Filter Quality" setting. It controls the rendering resolution and resampling settings in filters. As was with the old Blur quality setting, this also affects only the display, export will always use the highest quality.
Hmm... Weird, the tests obviously use png export, so they shouldn't be affected.
Sorry to have caused a bit of a fuss, I found it a bit too strange (also because apparently no one else was having problems), so I decided to simply do a completely clean build of inkscape. This solved all the regressions... (Weirdest case of a "dirty" build I've seen yet.)
I've modified my test script to always perform a clean build, so this shouldn't happen again.
participants (3)
-
unknown＠example.com -
Jasper van de Gronde -
Niko Kiirala