Why does llvm depend on python 2.7?

Chris Jones jonesc at hep.phy.cam.ac.uk
Wed Feb 5 14:33:03 UTC 2020


On 05/02/2020 2:18 pm, Ken Cunningham wrote:
> doing that doesn't really take us in the right directiion, though chris.
> reproducible builds and all that.
> we go from a fully spec'd proper python 

Actually no. That is not what happens at the moment. Check the clang-9.0 
config on darwin 11+ for yourself. The macports python version is 
currently *not* used at all. The lines

        configure.args-append \

which configure the clang-X build to use the macports python (2.7) are 
*only* used on darwin < 11. On everything else the defaults are used, 
which means the cmake build finds and uses /usr/bin/python.

to a randomly-found one, in a setting where afaik python 3 is known not 
to work.

We are already, for the clang-X builds (note I am specifically *not* 
talking about llvm-X or lldb-X here) relying on whatever the build 
systems decides to used based on its default configuration, for darwin 
 >= 11.

Now, if you want to *always* spec. the clang-X builds to use *always* 
use macports provided python 2.7, that is indeed an option. But that 
also requires some fixes in the port file w.r.t. what is currently done. 
i.e. remove the darwin  < 11 check at


so the configuration above is applied to *all* OSes, not just darwin < 11.

> and I am not sure at present how to even test all the lldb etc integration they mention.

the changes I am proposing do not touch the lldb-X builds. They already 
fully spec. the builds to use macports python, on all OSes.



> stripping the python spec is trivial, I could do it in 5 minutes, but it doesn't sound like what we should do here...
> ?
> K

More information about the macports-users mailing list