[LLVMdev] RFC: Timeline for deprecating the autoconf build system?

Renato Golin renato.golin at linaro.org
Tue Nov 4 13:59:50 PST 2014


On 31 October 2014 22:08, Tom Stellard <tom at stellard.net> wrote:
> - Is there any technical reason why the remaining autoconf users can't switch
>   to CMake?

Hi Tom,

One reason that wasn't mentioned is the buildbot infrastructure. Many
bots still build on autoconf and they all need to be moved into a
CMake builder. I have moved some ARM bots to my new ClangCMakeBuilder
but that's far from enough. Compiler-RT and libc++ (and I believe lld)
do use CMake, but the test suite is still dependent on autoconf and
the Windows build uses a different CMake builder. We were trying to
merge both, but didn't go very far.


> - What is a reasonable timeframe to allow the remaining autoconf users
>   a chance to migrate to CMake?

I'd say a few months and many engineer-hours from all over the place.
We'd need a task force like we did for the C++11 move: someone with
the will (you :) and representatives from the major groups (Linux,
BSD, Darwin / ARM, Intel, MIPS, PPC, etc) moving the bots, fixing the
issues, etc.

The big unknown is what to do with bots to which no one has access to,
or to which cannot be built with CMake. For example an old
architecture that hasn't been migrated to CMake. Do we deprecate the
bot? The architecture? Strongly encourage someone to migrate it? What
are the costs of this unknown migration, if that's even possible?

We may try to aim for the next release and see how far do we get...

cheers,
-renato



More information about the llvm-dev mailing list