Skip to content
Open
7 changes: 5 additions & 2 deletions Doc/library/subprocess.rst
Original file line number Diff line number Diff line change
Expand Up @@ -648,8 +648,11 @@ functions.
.. note::

If specified, *env* must provide any variables required for the program to
execute. On Windows, in order to run a `side-by-side assembly`_ the
specified *env* **must** include a valid :envvar:`SystemRoot`.
execute. On Windows, in order to run a `side-by-side assembly`_, or a
Python program using the :mod:`socket` module (or another module that
depends on it, such as :mod:`asyncio`), the specified *env* **must**
include a valid :envvar:`!SystemRoot`; omitting it may emit a
:exc:`RuntimeWarning`.

.. _side-by-side assembly: https://en.wikipedia.org/wiki/Side-by-Side_Assembly

Expand Down
3 changes: 3 additions & 0 deletions Lib/subprocess.py
Original file line number Diff line number Diff line change
Expand Up @@ -1545,6 +1545,9 @@ def _execute_child(self, args, executable, preexec_fn, close_fds,
if cwd is not None:
cwd = os.fsdecode(cwd)

if env is not None and not env.get('SystemRoot'):
warnings.warn("env lacks a valid 'SystemRoot'.", RuntimeWarning)
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I dislike this RuntimeWarning. It's ok to run programs without SystemRoot env var, or even in an empty environment. It just works fine currently for some programs. Example:

import subprocess, sys
cmd = [sys.executable, '-c', 'print("Hello World")']
subprocess.run(cmd, env={})

This program just works, it displays "Hello World" as expected.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I'm not sure how to synthesize this feedback with #134363 (comment).

How do we prevent the problem, if not warning about it, and not setting it in the subprocess? Should we implicitly override the caller's choices and clone SystemRoot into the subprocess env?

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I don't think that we have to prevent the problem. We can document it and explain how to avoid it.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Well, it works since Windows decided that using PATH to locate runtime dependencies (like an LD_LIBRARY_PATH equivalent) was such a terrible idea that it bypasses it for most OS stuff, but that "fix" upsets people as well.

I'm not particularly strong either way on this fix. It bites people all the time, and it's practically always a pointless thing to do (pass an insufficient environment). At the same time, there are 5-6 other variables that also ought to be included or things will break weirdly (e.g. TEMP), and this change doesn't check for those. But I don't think we have any good way to only warn when things break.

Is there a silent-by-default-enabled-with-X-dev warning that could be used instead? Would that be okay, @vstinner?

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

If we add a warning, it may be better to emit the warning in the socket module, no?

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Is there a silent-by-default-enabled-with-X-dev warning that could be used instead? Would that be okay, @vstinner?

A warning only emitted in the Python Developer Mode (-X dev) sounds more acceptable to me, yes.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

it may be better to emit the warning in the socket module, no?

The user's system could be configured to avoid that error, even without the environment variable, while plenty of other parts could also break if they are reconfigured.

Launching an application with an empty environment is the bad idea, so that's where the warning should be. I've diagnosed that as the root cause of a lot of issues.

Do we have an existing warning? Or do we need to invent a new one? This isn't a deprecation, and that's the only warning I'm aware of.


sys.audit("subprocess.Popen", executable, args, cwd, env)

# Start the process
Expand Down
Original file line number Diff line number Diff line change
@@ -0,0 +1,4 @@
Document that :envvar:`!SystemRoot` is also required for a Python subprocess
that uses :mod:`socket` or :mod:`asyncio` on Windows, and add a
:exc:`RuntimeWarning` if an *env* supplied to :class:`subprocess.Popen`
lacks it. Patch by John Keith Hohm.
Loading