[LLVMdev] 2.7 Pre-release1 available for testing

Óscar Fuentes ofv at wanadoo.es
Sat Mar 20 12:54:31 PDT 2010


Russell Wallace <russell.wallace at gmail.com> writes:

> Ah, thanks... curious, that set of instructions recommends running
> make in place (in the combined llvm directory), instead of in a
> separate directory as when building llvm without clang.

It doesn't *recommend* running cmake in-source. It is just that the
instructions are written that way. When generating VS solution files,
building in-source is not a problem. It is when generating makefiles, as
the generated makefiles will overwrite those provided with LLVM's
sources.

> Anyway, following those instructions (substituting unpacking the tar
> files for the svn checkout) gets me a .sln file (the resulting
> makefile doesn't work, but the .sln does), in which the build all
> command appears to work, and reports complete success.

cmake will generate makefiles if you wish. Just use this command:

cmake -G "NMake Makefiles"

but then you should use a different directory for the build.

> However, while it does generate a bunch of the usual ll*.exe's,
> clang.exe is still not there. Am I still missing something?

Suppossing that you have the LLVM source code in c:/llvm, do you have

c:/llvm/tools/clang/CMakeLists.txt

?

If that is not present, most likely your clang setup is wrong. The LLVM
cmake build test for the presence of the above file and, if found,
automatically builds clang. If the test fails, clang is ignored.

Once you have clang on the right place, reinvoke cmake from the top LLVM
source directory again so it notices the presence of clang. Use the same
command line you used for the first cmake invocation.




More information about the llvm-dev mailing list