One of the problems we currently have is that we are using ancient versions of libraries in our JHBuild, and old versions of GCC on the bots.
One of the reasons we do this is to unintentionally avoid breaking the build on systems that require using this old versions of the libraries or the compiler.
However, a majority of our users are using distributions that ship the last version of this libraries or the compiler.
So we end testing WebKit with a different version of the dependencies than what the users end running. Not ideal.
The proposed solution is:
1) Add two new buildbots that buid-test WebKitGTK+ on the last Debian stable and Ubuntu LTS without using the internal JHBuild. This bots will build against the system libraries. This will ensure we don't unintentionally break our policy about dependency versions https://trac.webkit.org/wiki/WebKitGTK/DependenciesPolicy
2) Then we will have complete freedom to raise the dependencies on the JHBuild without fearing unintentionally build breakage on older systems. Ideally we would like to keep the versions of the libraries on our jhbuild synced with the last GNOME modulesets. That way we can test WebKit with the version of libraries that most users will end using.
Created attachment 301703 [details]
Comment on attachment 301703 [details]
Clearing flags on attachment: 301703
Committed r212429: <http://trac.webkit.org/changeset/212429>
All reviewed patches have been landed. Closing bug.
IMHO name of "--default-cmake-features" argument does not imply it is disabling JHBuild.
(In reply to comment #4)
> IMHO name of "--default-cmake-features" argument does not imply it is
> disabling JHBuild.
You can build webkit with the jhbuild and --default-cmake-features, or you can build webkit without jhbuild and --default-cmake-features.
All possible combinations are supported.
But we want this bots to run without jhbuild, so we can be sure webkit remains build-able on the base configuration using only the distribution dependencies.
Checking for this argument on the buildbot config *and* for the gtk port on the buildbot config in order to decide if disable the jhbuild step is not ideal.
But i don't think there is ideal solution here.
If we add a explicit optional argument to disable jhbuild, then we have to take care of filtering that argument on the build step, as otherwise it would be passed to the build... causing the step to fail.
Committed r212437: <http://trac.webkit.org/changeset/212437>