Do not allow --target, --user, --root and --prefix together

--target, --user, --root and --prefix are mutually exclusive and simply don’t make sense when used together. Instead of failing for irrelevant errors, the error should explicitly describe why it is not excepted. This is currently implemented in GH-7828. I move the discussion here since the PR to avoid distracting the code review process.

[hexagonrecursion]

target and root

Something is seriously wrong here since (not /tmp/foo, /tmp/bar or /tmp/bar/tmp/foo but /tmp/bar/tmp):

$ pip install --target /tmp/foo --root /tmp/bar appdirs
Collecting appdirs
  Using cached appdirs-1.4.3-py2.py3-none-any.whl (12 kB)
Installing collected packages: appdirs
Successfully installed appdirs-1.4.3
$ ls /tmp/foo
$ tree /tmp/bar
/tmp/bar
└── tmp
    └── pip-target-_apxmir1
        └── lib
            └── python
                ├── appdirs-1.4.3.dist-info
                │   ├── DESCRIPTION.rst
                │   ├── INSTALLER
                │   ├── METADATA
                │   ├── metadata.json
                │   ├── RECORD
                │   ├── top_level.txt
                │   └── WHEEL
                ├── appdirs.py
                └── __pycache__
                    └── appdirs.cpython-37.pyc

It smells like tempfile.mkdtemp(dir=root/'tmp', prefix='pip-target-') is used as a temporary installation location and the files are never moved into target.

[pfmoore]

There are subtleties here, as @pradyunsg said. I appreciate that current behaviour is weird and likely broken in many cases, but see this xkcd comic 🙂

Specific things that should be considered:

• Users can set --user in their global config, and then they don’t have a way of switching it off. Having it ignored rather than erroring out when combined with other flags may be important to such users. This may be sufficient to want to behave differently depending on whether the flag came from config or command line, Which isn’t something we’ve ever done before, I believe.
• Ubuntu/Debian patch pip to make --user the default. This is an even more extreme case of the previous point.
• Some tools I believe force --root or --prefix - they are likely the main users of those flags (I don’t think end users specify them manually very often) so we’d need to understand how they expect things to work, or be prepared to react quickly if we break something for them.

That’s quite a lot of research to do.

With the new resolver work in full swing, and a release coming soon, I doubt many of the @pypa/pip-committers will have a lot of time to look at this issue in the near future. But if you were able to investigate some of the points above, and summarise your findings as you did with the current behaviour, that would help a lot when the discussion does get restarted.

Sorry if this sounds like stalling on the issue - the work you’ve put in here is genuinely appreciated, and I’d love to get this tidied up. Just not right now 😉