Hi all,
(If you don't ever care to run the webkit layout tests, you can skip this note).
As most of you are no doubt aware, we currently can only run the
webkit layout_tests on windows XP. For some of us who primarily
develop on 64-bit Vista, this is inconvenient at best, and this is
only going to get worse over time as more of migrate to 64-bit
machines and (eventually) Windows 7.
So, I'm working on porting the layout tests to Vista. This note is a
writeup of the approach I'm thinking of taking, and I'm looking for
feedback and suggestions, especially since most of you have been on
this code base a lot longer than me.
My basic approach is to try and get something up as quickly as
possible as a proof of concept, and then work to try and reduce the
maintenance over time. So, I've started by cloning the chromium-win
port over to vista, and modifying the test scripts to be aware of the
new set of test expectations. I will then tweak the tests to get
roughly the same list of tests passing on Vista as on Windows. The
main differences will have to do with how the theming renders scroll
bars and a few other native controls. I have most of this now, and
should have the rest of this in a day or two, but this is not a
maintainable solution without a lot of manual overhead.
Next, we'll get a buildbot setup to run on Vista.
While we're doing this, I'll start working on reducing the test set
duplication between XP and Vista. The best way to do this (we
currently think) will be to modify test_shell to *not* draw the native
controls, but rather stub them out in a platform-independent way for
test purposes (e.g., just painting a grey box instead of a real scroll
bar). Then we can write a few platform-specific unit tests to ensure
that the widgets do work correctly, but the bulk of the tests will
become (more) platform-independent. My hope is that we'll have
something that I can demonstrate here in a week or two, and that it
will extend trivially to Win 7.
A stretch hope is that we can even get the rendering to be
platform-independent enough that we may even be able to leverage them
across the linux and mac ports. I don't know if this is realistic or
not, as many of the tests may differ just due to font rendering and
other minor differences.
An alternative strategy is to start looking at more and more of the
tests and making sure they are written to be as platform-independent
as possible. First we'd this by making sure that we don't rely on
pixel-based tests where text-based tests would do. Another option
would be to switch to writing two tests just to ensure that page A
renders the same way as page B (where A and B use two different sets
of layout but should produce the same output). Both of these options
are significantly more work up front, but will payoff in much less
maintenance down the line. Also, all of this work will also overlap
with the webkit test suites, so it'll need to be coordinated with our
upstream buddies.
Comments? Thoughts?
-- Dirk
--~--~---------~--~----~------------~-------~--~----~
Chromium Developers mailing list: chromium-dev@googlegroups.com
View archives, change email options, or unsubscribe:
http://groups.google.com/group/chromium-dev
-~----------~----~----~----~------~----~------~--~---
