[Webkit-unassigned] [Bug 113282] [GTK] Bump required versions of some dependencies

bugzilla-daemon at webkit.org bugzilla-daemon at webkit.org
Tue Apr 2 08:47:06 PDT 2013


--- Comment #26 from Carlos Garcia Campos <cgarcia at igalia.com>  2013-04-02 08:45:18 PST ---
(In reply to comment #23)
> (In reply to comment #19)
> > Surprisingly jhbuild-optional.modules, already has atk 2.5.3, so we don't need to do anything else. I still find very confusing having modules that are *required* to build webkitgtk+ in a moduleset called optional.
> We have never meant to use the WebKit jhbuild to provide dependencies. It's only used for a particular type of package that requires an exact version to pass layout tests. In the past, it was not uncommon to have hundreds of layout tests fail because the version of Cairo you had was too new. Even though your version of Cairo met the compile-time WebKit dependency requirements, you needed a specific version to pass tests. The jhbuild makes it easy and to be honest, even possible, to pass tests, since it records exactly what version the baselines rely on and provides the appropriate fonts. Before the jhbuild we didn't even know what packages affected layout test output. We're still figure one out now and then.

I understand what jhbuild is for and it was a great idea, but the fact is that bots use it to build, so if you introduce a new dependency that happens to depend on another lib that is not "interesting" it doesn't build.

> The reason the jhbuild doesn't contain every dependency is that it's inefficient to compile everything twice if you don't absolutely need to. The other problem is that it is harder to distinguish normal dependencies from the "interesting" ones. By "interesting" I mean again the ones that require an exact version.

The problem is when a "normal" dependency requires another not "interesting". 

> The -optional moduleset appeared because time and again people wanted to add dependencies to the jhbuild that aren't "interesting," but merely because those dependencies were not not in their distributions package manager. I originally resisted adding these to jhbuild, since instructions for installing dependencies were listed on the trac wiki. At some point I realized I was wrong though, since it should be as easy as possible to compile WebKit. By separating these hard-to-get dependencies into -optional, we still avoid making everyone compile everything twice and know which dependencies are "interesting."

This is because the instructions to build webkit now relies on the internal jhbuild, even for people who will never run the tests. 

> It's possible that at the current moment the main jhbuild moduleset contains dependencies that it shouldn't. Those should be removed where appropriate. Hopefully this comment is helpful. Gustavo and I did not mean the jhbuild to be confusing and infuriating, we meant it to be helpful and to allow something that was just about impossible in the past.

I insist the jhbuild is not the problem, it was a great idea, but if it's only meant for bots and people wanting to run the tests, then the main build should not depend on it. I understand that jhbuild is meant for the bots, but you need to build in the end, and now it's not possible to build with the dependencies we want.

Configure bugmail: https://bugs.webkit.org/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the assignee for the bug.

More information about the webkit-unassigned mailing list