More helpful error when no matching version

[jtrakk]

Sometimes I try to install something and it says

Could not find a version that satisfies the requirement foobar (from versions: none)

when there is a package by that name but some constraint isn't satisfied, such as python_requires or --no-binary. This situation can be confusing.

The error message could explain what happened:

Found a package named foobar, but it specifies requirements that were not satisfied:
- python_requires: 3.7
- platform: Darwin

[chrahunt]

I can see this being useful for complex packages that publish many different versions. Here are a few options:

1. Put the information in an error message, as shown above. One potential problem is that the actual error might get lost if displaying information for several packages (see e.g. tensorflow)
2. A separate pip command that we could reference in the error text which, when run, would provide output similar to the above
3. A separate utility outside of pip - at the very least we could point to the parts of pip that would be helpful examples for implementing something like this.

Since we're planning to swap out the dependency resolver in pip at some point, I think it's worthwhile to ask whether this kind of checking is a use case that's been considered.

@pradyunsg, any thoughts?

[pradyunsg]

This specific case is probably covered in #5003. I'll go ahead and close this issue as a duplicate of that issue. ;)

---

A separate pip command that we could reference in the error text which, when run, would provide output similar to the above

I touched upon this in #6119. I like the idea -- we just need to create the software infrastructure around our logging to make it much easier to do. Or someone needs to figure out a better way than I outlined there, which I'm sure exists, given our trade-offs of developer time.

---

Error reporting for dependency resolvers, is a lot more complicated than it seems on it's surface. For context, this is the best current work that I could find, and that's not as informative as I'd hoped. And that's in a model that's simpler than Python packaging's dependency model. :(

I've sunken a lots of hours thinking and scribbling about this. I have some ideas; however, lack of time and mental peace needed to tackle a purely theoretical problem like this isn't helping. :)

[pradyunsg]

Also, thanks for filing this @jtrakk! Much appreciated! ^>^