On 05/10/2012 04:56 AM, David Airlie wrote:
----- Original Message -----
> From: "Jon Masters" <jcm(a)redhat.com>
> To: "Development discussions related to Fedora"
> Cc: "Michel Alexandre Salim" <salimma(a)fedoraproject.org>
> Sent: Wednesday, 9 May, 2012 10:57:30 PM
> Subject: Re: Like C++? Not afraid of quirky build systems? Seeking
> On 05/06/2012 02:29 AM, Michel Alexandre Salim wrote:
>> LLVM is becoming an increasingly integral part of our distribution
>> (with mesa now using it to build the LLVMpipe renderer, for
>> that I don't really feel comfortable maintaining it mostly by
> Thanks for the private email about ARM stuff. I've just kicked off
> another scratch build for ARM LLVM that might fix our outstanding
> problems. I'm ok - vaguely - in being a co-maintainer on ARM if there
> nobody else on our end who can represent ARM (as it seems). I started
> going through some of its design over the weekend, in my copious
> non-existent spare time to try to understand the ARM bits.
> More broadly though, I feel that GCC is well represented in terms of
> engineering knowledge but I'm *concerned* that we run the risk of
> growing a dependence on LLVM that is more critical than the LLVMpipe
> stuff. Before we can blink, we might need LLVM for building lots of
> other fundamental stuff. I am wondering if as a distribution we ought
> have an official FESCo-debated position on LLVM use? I do not think
> Fedora has the resources to maintain two critical toolchain pieces. I
> think LLVM is useful, etc. BUT its growing use is concerning.
Don't confuse llvm and clang, llvm has no equivalent in gcc world,
clang is a C compiler like gcc that uses llvm tech.
Right so I wasn't confusing these :) However, we package both together
and for ease of discussion many folks are going to think of it as a gcc
alternative (aside from the specific gfx situations you and ajax have).
My main concern was potential for growing use beyond that. I made an
analogy about glibc to which I accept ajax's response that they're
trying to reconcile with eglibc, but it's more the general concept I was
getting at. Let me avoid a specific example because someone will find a
way to find a hole in it :) Instead, my stance is we want to be very
careful about unsupportable use of LLVM. I've filed a ticket with FESCo
so hopefully there can be some debate as to acceptable use :)
It probably makes sense that one of myself, ajax or glisse help out
llvm, but we aren't the most reliable people in terms of spare time to commit.
Right. You guys have various incentives to care about specific use of
LLVM itself so I'm sure it will always be supported to some level, but
for the other piece - clang+LLVM, etc. - to grow further use in the
distro (in displacement of gcc) I feel we'd need to have actual RH staff
to support it that I don't think we have any plans to have. So I want to
cut this off at the pass before we blink and we have a problem.