I wrote a script which tests several rasterization programs, by using the official W3C SVG test suite and comparing the rasterized png with the expected pngs pixel by pixel.
The problem is, with v. 1.1 first edition (2011) and v. tiny 1.2 (2008) test suites, in a lot of images the vectors doesn't match with the expected png, because the revision number is not the same, making a lot of false-positives (more than 90%), like this one.
However it's ok with the v. 1.1 first edition test suite.
I could trunk the png to remove the area with the revision number, but it's a really derpy solution.
So by which png should I compare the rasterized vectors ?
Thanks.
There is no non-derpy solution to this problem, for a few reasons. The test images from this time were never meant to be ref-tests (that is, they are not pixel-by-pixel matches). Also, some of the tests that appear in the later test suites were not accepted as legitimate tests, so the revision numbers were not updated.
The later SVG 1.1 2nd edition test suite should be considered canonical, but even that contains some revision-mark errors, like coords-trans-06-t.
This is actually an issue for the SVG WG to resolve, and I'll raise it with them. The revision numbers in all approved tests should match the PNG references, and we can revise the tests so that the revision numbers match.