On Thu, Sep 24, 2015 at 9:41 AM, drago01 <drago01@gmail.com> wrote:
On Thu, Sep 24, 2015 at 6:04 PM, Dave Johansen <davejohansen@gmail.com> wrote:
> On Mon, Sep 21, 2015 at 6:27 AM, Adam Jackson <ajax@redhat.com> wrote:
>>
>> On Sun, 2015-09-20 at 21:00 +0800, Christopher Meng wrote:
>> > On 9/20/15, David Airlie <airlied@redhat.com> wrote:
>> > > It's llvm, there is never a good time to upgrade it and no llvm
>> > > release is
>> > > backwards compatible.
>> >
>> > Yes, so unless someone is pretty familiar with LLVM upstream, keep
>> > reading every changeset, no body can guarantee what will happen.
>>
>> That's not really a fair assessment.  By far the most important thing
>> in Fedora that needs llvm is Mesa.  The llvm build system has a fairly
>> comprehensive test suite (which, btw, fails on at least arm with 3.6.1,
>> but passes on all arches with 3.7), and Mesa has another test suite in
>> the form of piglit that we can use to ensure that changing llvm doesn't
>> regress llvmpipe or r600 or radeonsi.
>>
>> So we can in fact be pretty confident that llvm upgrades won't break
>> the things we actually use llvm for in Fedora.
>
>
> It sounds like there's a lot of momentum behind the update to 3.7 for F23,
> but iwyu ( https://github.com/include-what-you-use/include-what-you-use ) is
> built on top of clang and there's still not an upstream release that's
> compatible with 3.7 so updating in F23 will break iwyu until a release is
> made. So, I personally would prefer that the update to 3.7 only happen in
> Rawhide where breakage like this is expected, but I realize that I'm only
> one voice of many and that iwyu is used a LOT less than other packages like
> Mesa.

There is a llvm34 package for such software.

The currently release version builds against 3.6, so this would mean intentionally running an older version just to avoid the breakage. I'll probably just deal with the breakage until the new version is released.