Error parsing Os about pyflow

[David-OConnor]

Should be fixed

[David-OConnor]

Fixed in latest commit, for the case of BSD. Like you suggest, I'm going to parse pep 508 and handle all cases, so this doesn't come up for a different OS - that PEP's exactly what we need to prevent this from coming up again.

[David-OConnor]

It looks like neither PEP 508 nor the sys module docs have a comprehensive list. this SO post is the best I can find. Also using some tricks like if the name contains bsd, linux etc. Let me know if you have a more comprehensive/official list.

[nuno-andre]

The point is that these requirements are abstract, and were devised to be interpreted by the application according its context. So there will be nothing like an exhaustive list of values for those requirements.

And I think this is a great design because, to my mind, there's no reliable way to model such an ever changing domain.

I usually find them easier to handle by reversing the evaluation in a lazy and best-effort approach, rather than attempting to model every single dependency; since you know how to evaluate the markers of the systems you support, but not many others which actually do not matter. In this way, this issue is not a bug, but an enhancement ;)

The flow in naïve-Python would be roughly like:

requirements_that_matter = list()

for requirement in install_requires:
    for marker, value in requirement.markers.items():
        if value not in this_specific_system.get_values_for(marker):
            break
    requirements_that_matter.append(requirement)

I think this approach makes the dependency management easier, especially the transitive ones.

It's great to hear about your commit. I'm going to try the build again and I'll give you feedback. Thanks!