Re: A proposal for a modern build-system

From: Christian Mazakas (christian.mazakas_at_[hidden])
Date: 2025-05-06 19:08:11

On Tue, May 6, 2025 at 7:46 AM Peter Dimov via Boost <boost_at_[hidden]>
wrote:

> Christian Mazakas wrote:
> > You can invoke it similarly to b2 today:
> >
> >     ./c2.py \
> >       hash2 test \
> >       --cxxstd 11,20 \
> >       --variant debug,release \
> >       --address-model 64 \
> >       --toolset clang-19,gcc-14 \
> >       --asan --ubsan \
> >       --ctestflags="--output-on-failure -R 'xxh3'" \
>
> --output-on-failure (and --no_tests=error) should be the default, really.
>

Great call! I can update the script to do that trivially.

> >       --no-cmake \
>
> What does --no-cmake do?
>

I should update the name but it makes the script skip the CMake
configuration phase.

Right now, the script uses a very simple and naive mangling scheme for
creating the names of build directories. A goal of mine was to have them be
quasi-human readable so that way a frustrated user could intuitively
inspect a specific build configuration so like b2, I don't include any
passed cxxflags in the mangled name. To make sure that an accurate
configuration is always used, I delete the existing CMakeCache.txt if it
was there, which means a full re-configuration by CMake. `--no-cmake` is a
flag used to skip this part and instead `make -C build_c2py` is invoked
directly.

> >       -j20
>

Ha, in the ideal world I suppose we'd check for the available number of
CPUs and default to the max available. This is the value passed down to the
Make invocation.

- Christian