[Bf-committers] CVS, Make & MSVC

[Niels Provos]

Some general comments.

I would suggest to keep Makefiles for Unix and Windows separate for
now.  None of the tools (cmake nor qmake) seem to be versatile enough
to deal with auto-configuration.

An overview of different build tools:
  http://www.a-a-p.org/tools_build.html

For Unix, I would still suggest autoconf.  There are many different
Unix platforms and we probably would like to support all of them.  Too
many #ifdef become very complicated to manage.  There was a good paper
about the benefits and drawbacks of autoconf at this years USENIX
technical conference:

Overhauling Amd for the '00s: A Case Study of GNU Autotools
  http://www.cs.columbia.edu/~ezk/research/autotools/autotools.pdf

I do not think that perfection can be reached in one step.  It might
be sensible to get the autoconf stuff and separate Makefiles for
Windows committed.  And then improve on that.

> 5. We need a disciplined set of CVS commit rules... :-)
Peer review is working well for other projects.  Before you commit, you
need to get the approval of someone else (ideally, someone who is
knowledgeable in the area you plan to commit to).

In my experience, forcing people into a very strict environment like
suggested earlier turns people away from a project.  However, having
some form of unit and regression test framework is very useful, but
should not be tied into the commit process.

Niels.