Cairo needs help. It is the main 2D rendering library we use in GNOME, and in particular, it's what librsvg uses to render all SVGs.
My immediate problem with Cairo is that it explodes when called with floating-point coordinates that fall outside the range that its internal fixed-point numbers can represent. There is no validation of incoming data, so the polygon intersector ends up with data that makes no sense, and it crashes.
I've been studying how Cairo converts from floating-point to its
fixed-point representation, and it's a nifty little algorithm. So I
thought, no problem, I'll add validation, see how to represent the
error state internally in Cairo, and see if clients are happy with
getting back a
cairo_t in an error state.
Cairo has a very thorough test suite... that doesn't pass. It is documented to be very hard to pass fully for all rendering backends. This is understandable, as there may be bugs in X servers or OpenGL implementations and such. But for the basic, software-only, in-memory image backend, Cairo should 100% pass its test suite all the time. This is not the case right now; in my tree, for all the tests of the image backend I get
497 Passed, 54 Failed [0 crashed, 14 expected], 27 Skipped
I have been looking at test failures to see what needs fixing. Some reference images just need to be regenerated: there have been minor changes in font rendering that broke the reference tests. Some others have small differences in rendering gradients - not noticeable by eye, just by diff tools.
But some tests, I have no idea what changed that made them break.
Cairo's git repository is accessible through [cgit.freedesktop.org]. As far as I know there is no continuous integration infrastructure to ensure that tests keep passing.
Adding minimal continuous testing
I've set up a Cairo repository at gitlab.com. That branch
already has a fix for an uninitialized-memory bug which leads to an
free(), and some regenerated test files.
The repository is configured to run a continuous integration pipeline on every commit. The test artifacts can then be downloaded when the test suite fails. Right now it is only testing the image backend, for in-memory software rendering.
I've started reporting a few bugs against that repository for tests that fail. These should really be in Cairo's Bugzilla, but for now Gitlab makes it much easier to include test images directly in the bug descriptions, so that they are easier to browse. Read on.
Would you like to help?
A lot of projects use Cairo. We owe it to ourselves to have a library with a test suite that doesn't break. Getting to that point requires several things:
- Fixing current failures in the image backend.
- Setting up the CI infrastructure to be able to test other backends.
- Fixing failures in the other backends.
I think we can make use of modern infrastructure like gitlab and continuous integration to improve Cairo quickly. Currently it suffers from lack of attention and hostile tools. Help us out if you can!