I'd like to do something that could make things more difficult for typosquatters, what's the best way to commit the code?

[andrew-shvets]

Hey there! I'd like to do something to make life more difficult for typosquatters. Should I just create a remote branch and push my code changes there?

[uranusjr]

Start with decribing what you have in mind first.

[andrew-shvets]

Hi, sure thing.

The idea was to have a list (an XML file) of approved or preferred packages. That way if someone enters an incorrect package name by mistake, that will be flagged as an issue. How this issue is handled could a setting (prevent further installation, display a warning, etc.)

The main goal here is to get rid of typosquatting mistyping problems. For example, you want to install numpy, but you typed in numpu by accident and numpu could be a malicious package just waiting to be installed and executed. And if I make numpy a dependency of numpu, then the original user will get the functionality that they wanted and could miss entirely that they installed something that is undesirable. A warning before the installation happens could prevent this from occurring.

This is an attempt to foil one vector of attack, the idea is to add friction and make this approach more difficult.

[sepehr-rs]

Hi @andrew-shvets, thank you for filing this issue.

I think this is indeed an important security measure, but there isn't a consensus on how it should be implemented yet. There have been many discussions exploring different approaches, such as delaying package visibility, namespace support, and others. Similar discussions have also taken place in uv, but ultimately pip is a package installer, not a package registry, so I think this is better discussed with the PyPI maintainers. FWIW, PyPI has implemented name similarity checks that prevent certain lookalike package names (for example, if "py-foo" exists, you can't upload "py-fo0"). If you have another approach in mind, I'd encourage you to share it with the PyPI maintainers as well.

[andrew-shvets]

First, thank you to the person that attached the different labels to this issue.

Second, that's a good feature that PyPi has (although I have my doubts about the effectiveness pf this approach), but my view is that it should be done right before the package is installed and it should be a warning for the person doing the installation. People often make silly mistakes and they could be substantial in some cases.

How about this, I create an MVP and then we can discuss this further.

[sepehr-rs]

but my view is that it should be done right before the package is installed and it should be a warning for the person doing the installation

Yes, but this isn't really something that pip should support per se. This is better handled by the package registry or index you're installing from, since pip doesn't have access to the full set of packages available in an index. Keeping an allowlist of approved package names, as you suggested above, also doesn't seem practical, the list could grow very quickly and would be difficult to maintain.

How about this, I create an MVP and then we can discuss this further.

You're welcome to do so, but unless there's consensus among the maintainers that this belongs in pip, and your proposal addresses the concerns that have already been raised, I doubt it would be accepted.

[uranusjr]

From what I can tell, this needs the server to first be able to compile a list of valid names, before pip can do anything. This should therefore likely be raised in warehouse instead (the project that backs pypi.org) or on discuss.python.org before pip can do anything useful.