conda-feedstock broken?

[drewgilliam]

I've haven't gotten any response over at conda-forge, so I'm hoping someone here can help. The conda-forge feedstock has failed for almost all 2022 pipenv releases (latest pipenv version on conda-forge is 2022.1.8). Would anyone on the pipenv team be able to help fix the conda-forge feedstock?

conda-forge/pipenv-feedstock#24
https://anaconda.org/conda-forge/pipenv/files

[matteius]

@drewgilliam Can you kick off some new builds over there? All of the old builds are too old to have the log files still available in Github Actions.

[drewgilliam]

Thanks for the quick response! I don't think I have permissions to do anything in pipenv-feedstock - I'm just a user of conda-forge not an admin of the pipenv-feedstock. There seem to only be two human contributors in the history of pipenv-feedstock., @roccqqck and @beckermr, maybe they could kick something off?

[beckermr]

I don't recall making any non-admin changes to the feedstock. (I did accidentally make commits to most feedstocks on conda forge a while back.)

Feel free to push a fix.

[matteius]

@beckermr Well for one, I don't know what the pipenv-feedstock project is about, and the actual problem has been lost to the logs being no longer available, so unless you can kick off a run that demonstrates the failure within the logs, then I won't be spending time on it.

[drewgilliam]

So I believe I figured out the problem, which boils down to an environment variable.
conda-forge/pipenv-feedstock#25

PR #4944 changed the way PIPENV_IS_CI is calculated

PIPENV_IS_CI = env_to_bool(os.environ.get('CI') or os.environ.get('TF_BUILD') or False)

In my understanding, this code expects/requires CI and TF_BUILD to be unset, empty, or boolean-ish (truthy/falsey strings). However, in pipenv_feedstock CI=azure and all pipenv commands fail. I can replicate the problem just by setting CI=anything on pretty much any machine..

$ CI=dummy pipenv --help
Traceback (most recent call last):
  File "/usr/local/bin/pipenv", line 5, in <module>
    from pipenv import cli
  File "/usr/local/pipenv/lib/python3.9/site-packages/pipenv/__init__.py", line 57, in <module>
    from .cli import cli
  File "/usr/local/pipenv/lib/python3.9/site-packages/pipenv/cli/__init__.py", line 1, in <module>
    from .command import cli  # noqa
  File "/usr/local/pipenv/lib/python3.9/site-packages/pipenv/cli/command.py", line 4, in <module>
    from pipenv import environments
  File "/usr/local/pipenv/lib/python3.9/site-packages/pipenv/environments.py", line 100, in <module>
    PIPENV_IS_CI = env_to_bool(os.environ.get("CI") or os.environ.get("TF_BUILD") or False)
  File "/usr/local/pipenv/lib/python3.9/site-packages/pipenv/environments.py", line 34, in env_to_bool
    raise ValueError(f"Value is not a valid boolean-like: {val}")
ValueError: Value is not a valid boolean-like: dummy

Is this expected behavior?

[matteius]

Sorry for the delay in responding @drewgilliam, yes I believe this is expected that CI should be set to 1 or unset or 0. However at first glance, that code doesn't look correct. Shouldn't this line:
PIPENV_IS_CI = env_to_bool(os.environ.get("CI") or os.environ.get("TF_BUILD") or False)
Actually be more correct as this:
PIPENV_IS_CI = env_to_bool(os.environ.get("CI")) or env_to_bool(os.environ.get("TF_BUILD"))

cc: @oz123

[matteius]

@drewgilliam Clarifying question -- does your environment need CI to be set to azure for other reasons in the build pipeline besides pipenv?

[drewgilliam]

I don't know too much about the conda-forge feedstock environment - in my limited understanding, its a fairly automated environment where a lot of stuff is autogenerated for each feedstock package by the "conda smithy". The CI variable appears to be set in one of these auto-generated files, I assume/hope for good reason:
https://github.com/conda-forge/pipenv-feedstock/blob/main/.azure-pipelines/azure-pipelines-linux.yml#L32

Even with the suggested change, env_to_bool is going to throw an error if CI or TF_BUILD are anything but truthy/falsey.

pipenv/pipenv/environments.py

Lines 17 to 35 in d2365bb

	_false_values = ("0", "false", "no", "off")
	_true_values = ("1", "true", "yes", "on")
	def env_to_bool(val):
	"""
	Convert **val** to boolean, returning True if truthy or False if falsey
	:param Any val: The value to convert
	:return: False if Falsey, True if truthy
	:rtype: bool
	"""
	if isinstance(val, bool):
	return val
	if val.lower() in _false_values:
	return False
	if val.lower() in _true_values:
	return True
	raise ValueError(f"Value is not a valid boolean-like: {val}")

Assuming you're just trying to check if you're in a CI=1 type environment, maybe it would be better if env_to_bool returned false by default?

_true_values = ("1", "true", "yes", "on")

def env_to_bool(val):
    """
    Convert **val** to boolean, returning True only if truthy
    :param Any val: The value to convert
    :return: True if truthy, otherwise False
    :rtype: bool
    """
    if isinstance(val, bool):
        return val
    else:
        return val.lower() in _true_values

[matteius]

@drewgilliam Is it safe to close this ticket at this point seeing as conda-feedstock has been fixed?

[drewgilliam]

@matteius - before you close this issue, is there an answer to my comment above? It seems unexpected that pipenv will fail for any user/system that happens to have an environment variable CI which does not translate to a boolean variable.

[matteius]

@drewgilliam I opened a PR to improve handling of the CI variable: #5222