Bug 1697100: Don't let environment affect subprocess module search path

[mdboom]

@cpeterso: Can you confirm this addresses your issue by copying _process_dispatcher.py on this branch to /Users/chris/Code/mozilla/firefox/obj-x86_64-apple-darwin20.3.0-clang-mozbuild/_virtualenvs/init_py3/lib/python3.8/site-packages/glean/?

[badboy]

Awaiting confirmation that it solves the issue.

[cpeterso]

@cpeterso: Can you confirm this addresses your issue by copying _process_dispatcher.py on this branch to /Users/chris/Code/mozilla/firefox/obj-x86_64-apple-darwin20.3.0-clang-mozbuild/_virtualenvs/init_py3/lib/python3.8/site-packages/glean/?

I copied your new https://github.com/mozilla/glean/blob/95fb6783762608800d70c2e071b665a8c094e8a6/glean-core/python/glean/_process_dispatcher.py to my objdir obj-x86_64-apple-darwin20.3.0-clang-mozbuild/_virtualenvs/init_py3/lib/python3.8/site-packages/glean/_process_dispatcher.py, but I still hit the same Python error:

firefox % mach mozregression
...
 0:00.78 INFO: Running mozilla-central build for 2020-03-09
Traceback (most recent call last):
  File "/Users/chris/Code/mozilla/firefox/obj-x86_64-apple-darwin20.3.0-clang-mozbuild/_virtualenvs/init_py3/lib/python3.8/site-packages/glean/_subprocess/_process_dispatcher_helper.py", line 36, in <module>
    success = func(*args)
TypeError: _process() missing 1 required positional argument: 'throttle_backoff_time_s'

[cpeterso]

Looks like I just needed to re-run mach bootstrap. So you probably close this issue as WONTFIX. Sorry for the wild goose chase!

I don't have a PYTHONPATH environment variable set, but it looks like mach is picking up glean.py from my mach bootstrap directory here: $HOME/.mozbuild/_virtualenvs/mach/lib/python3.8/site-packages/glean/glean.py. That copy of glean.py still refers to throttle_backoff_time_s.

I reran mach bootstrap and then mach mozregression worked without hitting the throttle_backoff_time_s error.

[cpeterso]

It would be helpful if mach had a clobber status so it could warn users when they need to re-run mach bootstrap.

[badboy]

Looks like I just needed to re-run mach bootstrap. So you probably close this issue as WONTFIX. Sorry for the wild goose chase!

I don't have a PYTHONPATH environment variable set, but it looks like mach is picking up glean.py from my mach bootstrap directory here: $HOME/.mozbuild/_virtualenvs/mach/lib/python3.8/site-packages/glean/glean.py. That copy of glean.py still refers to throttle_backoff_time_s.

I reran mach bootstrap and then mach mozregression worked without hitting the throttle_backoff_time_s error.

Hmpf, sounds like bad interaction between the 2 python programs.

It would be helpful if mach had a clobber status so it could warn users when they need to re-run mach bootstrap.

IMO worth reporting on bugzilla. Would also play well with ac_add_options --enable-bootstrap then I guess.
Though I'm not sure it would have helped here because probably no one would have flagged this as clobber-worthy.

[mdboom]

Thanks @cpeterso. It feels like it should at least theoretically be possible to ensure that the code in the subprocess is the same as in the parent process somehow (even if this shot-in-the-dark didn't work). Then we wouldn't have to worry about clobbering issues etc. So, I'll close this PR as WONTFIX, but might keep the parent bug open for a while to investigate further.

[kvark]

I was having a similar issue on NixOS, and grabbing the patch appears to fix it.
@mdboom can we land this please?

[mdboom]

I think this is still fine. We had originally chosen to pass on it because we didn't full understand the root cause (and we still don't). But it's definitely helping the minority that falls into this -- we should keep an eye out on any regressions when this lands in m-c, though.l