Ergonomic, hermetic, user-friendly software build system for Python, Go, Java, Kotlin, Protobuf, Scala, Shell. Pants lets you fearlessly scale up your codebase!

Pants

and `pex` uses that to parse version strings

hm yeah, it’d probably just surface as a resolve failure vs a ‘i found this thing that sorta looked like it may be resolvable but ignored it because it’s version didnt parse’ type message

not sure there’s a great precedent for warning about things like that

would need an incremental-ish type parser

so if it can’t completely parse, at least get the name, see if it’s similar to whats being resolved, warn, etc

and i’m not familiar with `managed_dependencies` - what’s that do?

it overrides the versions resolved by ivy for transitive dependencies, regardless of what versions of those transitive deps were required by the applicable pom.xml's

basically it lets the project owner specify exactly what shows up on the classpath

versus this fun: ```Unsatisfiable: Could not satisfy all requirements for setuptools==30.0.0:
    setuptools==30.0.0, setuptools&lt;20.11,&gt;=2.2(from: apache.aurora.client==0.17.0-&gt;pex==1.1.14)
```

I set `setuptools==30.0.0` since that is what pants v1.3.0 wants

I could isolate the aurora dep to a separate `python_requirements` target

assuming that there's no interaction with the main `python_requirements` target we use

ah, no.. currently we rely on floating transitive reqs (e.g. ranges or no version specified) for compatibility within a resolve context

well, `python_requirements` nets you only addressables

so conflicting versions can all be defined in the same requirements.txt file - then depending on what addresses a given target depends on, it’ll resolve only those

by "rely", do you mean that pants relies on the maintainers of each module to use floating reqs?

and modules like aurora which violate that assumption cause issues like this?

funnily enough m.ateo tried to upstream a fix to aurora to get them to fix their reqs

I don't believe they ever took the patch :disappointed:

so in general most pypi publishers specify either no requirements (`thing`) or the minimal constraint necessary to work (`thing&gt;0.1.0`)

but I suspect the whole ‘strict pinning’ culture at Twitter spilled over into aurora

yeah so in general there's no real call for "transitive dependency mgmt" in python world as there is in JVM world

I'll just make a custom aurora wheel with the reqs set the way I want

and bang my head against the proverbial wall

does aurora actually publish these wheels anywhere? I can’t seem to find on pypi

just reread my runbook, I had generated them many months ago