Since 3.5, optimized Python modules for -O and -OO are installed using different names. The Python build system itself seems to been prepared for this as Python itself is built with both .opt-1.pyc and .opt-2.pyc files built. However, distutils at the moment can only install one of them.

Gentoo is relying on 'setup.py install --compile -O2' doing the right thing. We can change the arguments if necessary but calling it twice would require a lot of conditional changes to how we build packages. Of course, not installing both variants is not an option since it causes Python to attempt to write them at runtime later.

Could we please get some way to build both .opt-1.pyc and .opt-2.pyc in setup.py install? Either through making -O2 build also -O1 implicitly, or providing an ability to specify multiple optimization levels.

Relevant Gentoo bug: https://bugs.gentoo.org/585060

I would suggest using --optimize=1,2 / -O1,2 for this purpose.

Documentation in this part of Lib/distutils/command/build_py.py and Lib/distutils/command/install_lib.py should also be updated:

        ('optimize=', 'O',
         "also compile with optimization: -O1 for \"python -O\", "
         "-O2 for \"python -OO\", and -O0 to disable [default: -O0]"),

Since this is an addition to the changes made for PEP 488 (Issue23731), nosying Brett

So the reason Python does the right thing at install is because compileall is used by the Makefile to do the compilation and it's run twice (it actually doesn't use distutils for this).

My question is what did distutils do before PEP 488 since .pyc and .pyo files had different names before as well? Did distutils build both previously and so this is a regression? Or is this a feature request?

Brett, .pyc was controlled by --compile and .pyo by --optimize (either 1 or 2). Technically only two variants could be used simultaneously, and distutils accounted for that.

Now you have .pyc + .opt-1.pyc + .opt-2.pyc, so three variants instead of two.

But distutils still does the right thing with -O2 and --optimize 2, right? If that's the case then this is an enhancement request to add support to distutils to compile all bytecode levels and not a regression.

Yes, you could put it like this.

FWIW the behaviour in distutils was intended to be controlled by the setup.p options, but in fact also depended on the options of the python command itself (-O and dont-write-bytecode option/envvar).  This was changed in distutils2 but judged too big to fix in distutils: https://hg.python.org/distutils2/rev/7c0a88497b5c

Distutils is now deprecated (see PEP 632) and all tagged issues are being closed. From now until removal, only release blocking issues will be considered for distutils.

If this issue does not relate to distutils, please remove the component and reopen it. If you believe it still requires a fix, most likely the issue should be re-reported at https://github.com/pypa/setuptools