[cfe-dev] is configure+make dead yet?

[David Röthlisberger]

On 26 Jun 2012, at 20:26, James K. Lowden wrote:
> I used autoconf to build Clang not because I'm "stuck" on a
> system without Cmake, but because I have expertise in autoconf and none
> with Cmake.  I've never found Cmake compelling enough to justify
> learning a new feature-test and dependency syntax.  

Before discussing the technical merits of CMake vs autoconf, we need to
answer the question: Is having my favourite build system available a
good enough reason to impose the maintenance burden of *two* build
systems on the rest of the llvm & clang community?

As a *user* of the build systems (i.e. only using them to compile, not
actually editing the Makefiles/CMakeLists.txt) I already found it very
confusing having two build systems. There is no documentation (except
buried in the mailing lists) as to which projects / features /
architectures are supported by which build system. And having two
separate build systems almost guarantees that there will always be
something that can only be built with one system or the other.

That was as a user. I can only imagine how demoralising it must feel to
have to add a new project or feature to two different build systems.

If the following statement is true, then which build system to choose
is a no-brainer:

> cmake, while ugly, can be made to support all of our use cases. There
> are some use cases that autoconf+make can't support
-- Chandler Carruth, On 21 Jun 2012, at 01:19

It seems to me from this thread† that the community has agreed that,
once the CMake build system supports every project/architecture/etc that
the autoconf build system can, the CMake system will become the official
build system.

It also seems clear from Doug's list‡ that the CMake system isn't ready
yet. Is it really true that the CMake system doesn't run the LLVM test
suite, or am I misreading Doug?

I hate to think of the person-hours that have been spent on this
discussion over the years, so it would be nice to achieve *something*
from this round of the discussion. I propose that this decision be
clearly documented: The long term goal is to support only the CMake
build system; until then, new projects must support the CMake build
system.

I'm not sure where best to document it, though. So far I have found:

* "LLVMBuild Documentation" at "Development Process Documentation"
  (http://llvm.org/docs/development_process.html)

* "The LLVM Getting Started Guide" and "LLVM CMake guide" at
  "User Guides" (http://llvm.org/docs/userguides.html)

* "Getting Started: Building and Running Clang"
  (http://clang.llvm.org/get_started.html)


† Quotes from this thread:

> I consider it a worthy goal.
Douglas Gregor, On 21 Jun 2012, at 18:21

> I'd rather we just pick cmake
Chandler Carruth, On 21 Jun 2012, at 01:19

‡ From Douglas Gregor, On 21 Jun 2012, at 18:21:
> Some things that CMake needs to do well for it to become the only way to
> build LLVM/Clang:
>   - Optionally build and install compiler-rt 
>   - Optionally build and install libc++
>   - Ease-to-use cross-compilation support
>   - Documentation to make it easy to understand how to do the above
>   - LLDB?
>   - LLVM testsuite support
> 
> And some value-add that might make CMake motivating for others:
>   - Easy bootstrap
>   - Build packages/installers