PYTHONDONTWRITEBYTECODE documentation is incorrect

[mikekaganski]

Documentation

PYTHONDONTWRITEBYTECODE is documented in Doc/using/cmdline.rst as:

If this is set to a non-empty string, Python won't try to write .pyc files on the import of source modules.

This is incorrect. The variable is obtained in Python/initconfig.c as:

    int dont_write_bytecode = 0;
    _Py_get_env_flag(use_env, &dont_write_bytecode, "PYTHONDONTWRITEBYTECODE");
    if (dont_write_bytecode) {
        config->write_bytecode = 0;
    }

and _Py_get_env_flag reads the environment variable value as integer, treating non-integer string as 1. Specifically, this means that PYTHONDONTWRITEBYTECODE=0 will not suppress writing .pyc files, even though the environment variable is assigned a non-empty string "0".

Linked PRs

• gh-143134: Docs: clarify PYTHONDONTWRITEBYTECODE behavior #143144

[StanFromIreland]

The same can be said for several others:

$ PYTHONDONTWRITEBYTECODE=0 ./python -c "import sys; print(sys.flags.dont_write_bytecode)"
0
$ PYTHONINSPECT=0 ./python -c "import sys; print(sys.flags.inspect)"
0
>>> exit
$ PYTHONNOUSERSITE=0 ./python -c "import sys; print(sys.flags.no_user_site)"
0

[Javed-Shekh777]

Hi, I’d like to work on this documentation fix and submit a PR.

[serhiy-storchaka]

The documented behavior is the initial behavior. Since the change is not documented, it was not intentional change.

It seems that it was changed in a7368ac (bpo-32030). Since it was so long ago, we cannot longer consider it a behavior bug, but a documentation bug. The documentation should be made reflecting the current behavior, and the versionchanged directives should be added.

cc @vstinner

[vstinner]

Other environment variables are impacted (they also use _Py_get_env_flag()):

• PYTHONDEBUG
• PYTHONDONTWRITEBYTECODE
• PYTHONINSPECT
• PYTHONLEGACYWINDOWSSTDIO
• PYTHONNOUSERSITE
• PYTHONOPTIMIZE
• PYTHONUNBUFFERED
• PYTHONVERBOSE

[StanFromIreland]

PYTHONINSPECT

IIUC Serhiy correct (I'm not quite sure why) the behaviour of this one is correct #143135 (comment)

[serhiy-storchaka]

For all of these variables check whether setting the variable to "0" actually changes the behavior or only sys.flags. If it only changes the latter, then this is the bug in the code, otherwise the bug is in the documentation.