I've successfully built and installed a copy of Python3.6.8 (replacing a
probably buggy installation on my RHEL system, different story). Also I
set up a virtualenv by doing "$ /usr/local/bin/python3.6dm -m venv
/usr/local/pyenv36"
In my activated virtualenv, I try "$ pip install numpy" but it fails
with lots of errors about something that is valid only in C99. Obviously
numpy relies on -std=c99
(pyenv36)$ /usr/local/bin/python3.6dm-config --cflags
-I/usr/local/include/python3.6dm -I/usr/local/include/python3.6dm
-Wno-unused-result -Wsign-compare -g -Og -Wall
(pyenv36)$
This surprises me, because "my" /usr/local/bin/python3.6m itself was
built -std=c99, as seen in these sysconfig entries:
'CONFIGURE_CFLAGS_NODIST': '-std=c99 -Wextra -Wno-unused-result '
'PY_CFLAGS_NODIST': '-std=c99 -Wextra -Wno-unused-result '
'PY_CORE_CFLAGS': '-Wno-unused-result -Wsign-compare -g -Og -Wall
-std=c99 '
I can install numpy on the system Python3, and not surprisingly, there
is -std=c99 among tons of other CFLAGS:
$ /usr/bin/python3.6m-config --cflags
-I/usr/include/python3.6m -I/usr/include/python3.6m -Wno-unused-result
-Wsign-compare -O2 -g -pipe -Wall -Wp,-D_FORTIFY_SOURCE=2 -fexceptions
-fstack-protector-strong --param=ssp-buffer-size=4 -grecord-gcc-switches
-m64 -mtune=generic -D_GNU_SOURCE -fPIC -fwrapv
-DDYNAMIC_ANNOTATIONS_ENABLED=1 -DNDEBUG -O2 -g -pipe -Wall
-Wp,-D_FORTIFY_SOURCE=2 -fexceptions -fstack-protector-strong
--param=ssp-buffer-size=4 -grecord-gcc-switches -m64 -mtune=generic
-D_GNU_SOURCE -fPIC -fwrapv
$
My questions are:
1) Why does the output of /usr/local/bin/python3.6m-config --cflags
differ from the CFLAGS in sysconfig.get_config_vars() ? I've confirmed
that the /usr/local/bin/python3.6m binary and
/usr/local/bin/python3.6m-config are really from the same build.
2) How does one activate the necessary CFLAGs for extension building?
3) Where do all the extra flags in the system /usr/bin/python3.6m-config
--cflags come from?
4) Why isn't the config script installed into the /bin of a virtualenv?
This is super annoying when building extensions on a multi-version
system because it may lead to version mix-ups
A bit of background on this: I've written a C extension that leaks
memory like a sieve on my production RHEL7's system python3 but neither
on my Debian development system nor on my "self-built" Python on the
production server. So I'd like to install the self-built Python on the
production server, but for that I need a bunch of other packages to work
as well, including numpy.
Thanks!
--
https://mail.python.org/mailman/listinfo/python-list
