[Mesa-dev] [PATCH v4 00/18] configure.ac: Hopefully the last!

[Tobias Droste]

Am Mittwoch, 19. Oktober 2016, 14:33:19 CEST schrieb Emil Velikov:
> Hi Tobias,
> 
> Thanks for sticking around and re-spinning the lot. I hope you'll
> agree that things look a lot better now.

True :-)

> 
> On 16 October 2016 at 01:20, Tobias Droste <tdroste at gmx.de> wrote:
> > This series leaves out everything from the previous series that was
> > to controversial, requires more discussions or could be done as a
> > follow up.
> > 
> > It just cleans up how LLVM version/target/component checks are used.
> > No version or other requirements were changed.
> > 
> > I reordered the patches a little bit and reduced the code movement
> > a lot.
> > There's only 1 line that is moved twice (call of the function
> > "llvm_set_environment_variables()"), but there's no way around this
> > (patch 12 and patch 14). The reason why is explained in patch 12.
> > 
> > After each patch I made sure that these drivers:
> > r300,r600,radeonsi,radv,llvmpipe
> > either build or fail to configure under these conditions:
> > a) LLVM installed and version >= version required
> > 
> >     --> build
> > 
> > b) LLVM not installed
> > 
> >     --> fail to configure
> > 
> > c) LLVM installed and version < version required
> > 
> >     --> fail to configure
> > 
> > d) LLVM installed, version >= version required but missing
> > target/component
> > 
> >     --> fail to configure
> > 
> > e) "enable-gallium-llvm" needed but "disable-gallium-llvm" used
> > 
> >     --> fail to configure
> > 
> > I also made sure that i965 ("classic" dri driver) and softpipe build
> > regardless of the LLVM version or if LLVM is installed.
> > 
> > @Emil:
> > I left "llvm_add_component()" to only take one component at the time.
> > This way the error message can state which component is actually
> > missing. If you still want it changed to take multiple components
> > at once, this could be easily done.
> 
> Indeed. Let's not bother with that for now.
> 
> > Patches 5,6 and 7 could be one patch as they just move code around,
> > but the diff was really unreadable so I split it up.
> > I also kept the reviewed by tag for patch 1 and (now) patch 8,
> > because they didn't change.
> 
> All the patches which don't have any comments/suggestions are
> Reviewed-by: Emil Velikov <emil.velikov at collabora.com>
> 
> For the rest, feel free to send the odd v5 rather than re-spinning the
> whole lot.
> 

I will send out a v5 for patch 12 and 18 (split up).
Patch 11 and 17 may get a v5 depending on your feedback.

> Thanks again for untangling LLVM bits !
> Emil