[Python-Dev] Proposing "Argument Clinic", a new way of specifying arguments to builtins for CPython

[Chris Angelico]

On Wed, Dec 5, 2012 at 9:17 AM, Larry Hastings <larry at hastings.org> wrote:
> Here's my blue-sky probably-overengineered proposal: we (and when I say "we"
> I mean "I") write a cross-platform C program that could be harmlessly but
> usefully integrated into the build process.  First, we add a checksum for
> the *input* into the Clinic output.  Next, when you run this program, you
> give it a C file as an argument.  First it tries to find a working Python on
> your path.  If it finds one, it uses that Python to run Clinic on the file,
> propagating any error code outward.  If it doesn't find one, it understands
> enough of the Clinic format to scan the C file looking for Clinic blocks.
> If it finds one where the checksum doesn't match (for input or output!) it
> complains loudly and exits with an error code, hopefully bringing the build
> to a screeching halt.  This would integrate Clinic into the build process
> without making the build reliant on having a Python interpreter available.

That would probably work, but it implies having two places that
understand Clinic blocks (the main Python script, and the C binary),
with the potential for one of them to have a bug. Is it possible,
instead, to divide the build process in half, and actually use the
newly-built Python to run all Clinic code? That would put some (maybe
a lot of) restrictions on what functionality the Clinic parser is
allowed to use, but if it can work, it'd be clean. (The main code of
Clinic could still demand a fully-working Python if that's easier; I'm
just suggesting making the "check the checksums" part of the same
Python script as does the real work.)

ChrisA