On 17/05/2017 00:42, Michael Torrie wrote:
On 05/16/2017 05:01 PM, bartc wrote:
It should be a piece of cake, yes?
If TCC implements the standard sufficiently, then yes it's possible.
However it won't be easy because the Python build tools are geared
towards the dominant compilers (GCC and VS), so you'd have to rework all
the build batch files first. Building an executable involves more than
just compiling a C file. There is dependency calculation, optional
compilation of features, linking, etc.
Thank you, this is what I suggested several hours ago, that I was picked
up on:
"I can't test with Python because it's too complicated to compile,
especially on Windows.
But if I use tcc to compile /my/..." (implying I'd rather have used tcc
to compile CPython, if it had been trivial).
And actually my comments were also about building Python 'outside the
box' if that makes sense. Just manually going through the tasks because
there are certain to be issues that come up and I would need to know
where I am.
Although it seems the information needed to do this is either in the
18000 lines of configure script, or 7000 lines of VS project info, both
incomprehensible, so I doubt I would get far.
Trial and error might work, and tcc seems capable of compiling a few
sample modules I tried. But I don't have the inclination to battle all
the obstacles when all I wanted was a ball-park figure of how much
slower tcc would be than gcc at running CPython.
My much simpler test, on a similar application, suggested it might be
half the speed at worst.
--
bartc
--
https://mail.python.org/mailman/listinfo/python-list
