I have managed to write a DumpRenderTree from the GdkLauncher. On few examples, it looks like results are very similar : differences are in the text size, due to the font differences, but trees are nearly matching those expected ! With a few adaptation of the run-webkit-tests scripts, i am able to execute series of tests. So it seems to be a good start. The technical points i have to resolved now are : * the notification of the "loaded" state. In the .m files, it's done with some delegates. Is there a way of getting this event in webcore, on the frame or frameview interface, for instance ? I have looked for registration/delegation/event patterns but could not find a direct solution. * the LayoutTestController : i suppose i have to rewrite it in C++, in a binding js object, if i want to pass test that uses the "waituntildone/notifyDone" behaviour. Am i right ? Thanks, Ronan On 7/4/06, Mike Emmel < mike.emmel@gmail.com> wrote:
Sorry one more post my suggestion for running the layout tests sooner then later is to use the gdk port to OSX its in the gnome CVS.
You should be able to use the freetype backend. At that point differences are very minimal and at the cairo level if at all and these is more a issue of testing cross platform issues in cairo itself this should be small.
This gives you the needed NSView from the GdkWindow. One of the big reasons I chose gdk actually since I can run in with minor work on OSX/Win32/Linux.
Mike
On 7/4/06, Mike Emmel <mike.emmel@gmail.com> wrote:
On 7/4/06, Eric Seidel <macdome@opendarwin.org > wrote:
On Jul 4, 2006, at 1:35 PM, Mike Emmel wrote:
So why run the layout tests now ? The current ports problems are evident via simple single page
testing.
Layout tests are always useful. Even if the port is otherwise non- functional, the layout tests will help you better understand your changes, when you make them. Sometimes you'll make a one line change and change (fix or break) tens or hundreds of tests at once.
I don't disagree that there not useful if it was a simple matter to get them to run. But it looks like a pretty big project which may not complete for some time. Which brings up a number of questions. Is this the best test code base for cross platform testing ?
There are other tests suites acid2 I believe that may be more amendable. In anycase there are a lot of testing frameworks out ther.
Again I'm just questioning the timing and scope of the project.
Not that it would not be useful if it was available.
Mike
On 7/4/06, Ronan Meneu < rmeneu@origyn.fr> wrote:
* I have compiled the DumpRenderTree tool in the .vcproj, modified it to build a FrameGdk instead of a FrameWin. Purpose of this tool i think is to just dump render tree. It does not have some functionalities of the objective-c code like "image diff". It works on linux, but layer size is always 0x0. Is this the DumpRenderTree tool used on win32 platform ?
This is the correct direction to move in. DumpRenderTree.cpp just hasn't received enough lovin yet.
An ambitious coder might chose to try and port more of DumpRenderTree.xcodeproj (all the associated .m files) to c++ to be shared. It's probably just as easy to write new .cpp files however. Especially since it will be a long time before any sort of delegation model (for editing, loading delegates, etc.) exists for WebKit for any platform other than OS X.
ImageDiff would be simple enough to write using ImageMagick, or any other image library which knew how to read pngs into an RGBA array.
One of the first problems someone needs to solve, is setting up run- webkit-tests to dump separate -expected.txt files for windows or find some way of comparing windows/linux expected.txt output with mac expected.txt output. This is likely impossible due to different fonts on the three systems.
Happy hacking.
-eric
-- Ronan Meneu Origyn Web Browser for Embedded Systems Team Senior Software Engineer