I thought the plan for windows was to use interceptors like the sandbox for the system level calls (DrawThemeBackground, etc) and just mock everything. This would fix running the tests on Vista and not require a certain theme on XP. AFAIK, no one has started on this work yet.
On Mon, 5 Jan 2009, Evan Martin wrote: > > On Sun, Jan 4, 2009 at 10:14 PM, Brett Wilson <[email protected]> wrote: > > I just upgraded my dev box to one of the new quad-cores and put Vista > > 64 on it so I can use more memory for the disk cache. (Note: XP64 > > isn't one of the standard corp OS images.) I suspect more people will > > want to start doing this (I'm still doing tests, will send out the > > results). > > > > I just realized that I have a problem now: I have no way to run webkit > > pixel tests since the form control reference images all require XP's > > Luna theme. I have three boxes: Vista, Mac, and Linux, so it's not > > reasonable to get another one. It also seems impossible to require > > everybody use XP forever. > > > > WebKit solves this by putting the layout test results of each build in > > a public place on the buildbot (I don't think ours are, are they?) and > > when you change something that has a reference that isn't for your > > system, you just grab it off there, double-check that it's correct, > > and check it in. However, this requires more tolerance of test redness > > than we have had so far. > > > > The other option is to use platform-independent form control > > rendering. This actually isn't too hard. WebKit already has form > > control rendering for when you have funky CSS styles applied (think > > when you set "border-top:2px solid red;" on a text fiend) and we can > > enable this easily enough in our theme control by returning the > > correct values (maybe there are a few edge cases that need cleaning > > up). > > > > This has the advantage of letting people running XP, Vista or either > > of those with the classic theme run the layout tests in a meaningful > > way. It has the disadvantage that none of our native theme rendering > > will be covered by pixel tests, which could lead to regressions. > > It would also let the Linux output be (nearly) shared with Windows. > Currently we match Windows up to the details of antialiasing so the > fuzzy differ finds them identical enough unless form controls are > involved. > > But at least in the Linux world, we've had a ton of bugs related to > form control rendering, and they appear to be subtle (e.g. taking > improperly taking some transformation into account when clipping), > triggering only a few layout test failures. It is important to > continue testing those code paths. > > Does Vista not support classic-mode rendering? Clearly the solution > is to require everyone to use classic. :P > > Slightly more seriously, you're welcome to check in Vista or OSless > baselines as long as you're willing to maintain them when they break. > Otherwise it'll be yet one more "platform" people will need to > rebaseline when they fix a bug somewhere. :\ > > > --~--~---------~--~----~------------~-------~--~----~ Chromium Developers mailing list: [email protected] View archives, change email options, or unsubscribe: http://groups.google.com/group/chromium-dev -~----------~----~----~----~------~----~------~--~---
