On 16.01.2014 17:20, Jonathan Schleifer wrote:
Am 16.01.2014 um 17:13 schrieb Stephan Aßmus <superstippi@xxxxxx>:On 16.01.2014 16:53, Jonathan Schleifer wrote:* If you allow to use any installed Clang, it means you will also test Haiku with Clang from trunk and thus *if* there is a problem realize it early enough to fix it → we can make sure a Clang release never breaks for Haiku, meaning a new Clang release will *never* break, completely eliminating the need to require a specific version. Therefore version x.y or up will be fine and there's no need to always build a bundled Clang.Actually, the user who is trying to compile Haiku on his system will notice it. He or she will probably be frustrated.That is what I said about enabling -Wall -Werror. But there it was ok because it reveals bugs. And now it's suddenly not ok for another even though it reveals? Sounds like double standards to me…
Actually, what you say doesn't make sense. -Werror would be tested and enabled for exactly the version of Clang that would be in buildtools. New warnings would simply not appear out of nowhere -- they would if we followed your suggestion of using any random Clang version that might be installed on a user's system (which may have introduced new warnings).
The idea is to reduce variables. Just imagine someone reports a bug with his self-build Haiku version, and the bug happens to actually be a compiler bug with the specific version of Clang he has installed, perhaps he build it himself. I don't want project resources wasted with someone trying to investigate that bug, while he couldn't possibly reproduce it with his setup (different Clang version).
Best regards, -Stephan