Port _guess_toolchain_kwargs_from_python_config away from distutils #28
Conversation
| object_names = [ | ||
| oname for oname in make_vars["OBJECT_OBJS"].split() | ||
| if "(" not in oname and ")" not in oname] | ||
|
|
||
| object_suffix = ".{}".format(object_names[0].split(".")[1]) |
There was a problem hiding this comment.
This just tries to get the extension for objects on the system, right?
It seems sysconfig has a dumber parse_makefile that doesn't parse multiline variables. This switched to check MODOBJS, which seems to be the only variable in the makefile that's defined on a single line :\
There was a problem hiding this comment.
The new code isn't particularly worse than the old code, and it seems to do the job at least on Linux. So, close enough, to me.
codepy/toolchain.py
Outdated
| cflags.append(lib) | ||
|
|
||
| ld, *ldflags = config_vars["LDSHARED"].split() | ||
| so_ext = os.path.splitext(config_vars["EXT_SUFFIX"])[-1] |
There was a problem hiding this comment.
The previous version seemed to assume that make_vars["SO"] returned an extension, but at least on my system it doesn't exist at all. config_vars["SO"] exists though (from sysconfig), but that returns the cpython-[arch].so.
It also says that "SO" is deprecated in favor of "EXT_SUFFIX".
There was a problem hiding this comment.
Actually, it's .cpython-[arch].so.. was that the intended full suffix here?
There was a problem hiding this comment.
I think that's OK. As long as CPython looks for xyz.cpython-[arch].so for import xyz, that's all that's needed.
alexfikl
force-pushed
the
port-distutils
branch
from
32a453b to
3137b14
Compare
|
My biggest worry here is Mac compatibility. Could you add a CI that tries that? (Maybe on |
alexfikl
force-pushed
the
port-distutils
branch
from
3137b14 to
d3bff9e
Compare
2 participants
At least on my system, all these variables seemed to return the same things, so hopefully it doesn't break too much. I also ran the
loopytests and they still pass.