I recently learned that pip
does not resolve transitive dependencies in any sane fashion. This means after installing a package, some of its dependencies' dependencies may not actually be satisfied.
Is there a way to programmatically check that all installed packages' dependencies are actually satisfied after installing?
Specifically, I would like to be able to create a virtualenv, install some packages into it, and then verify that all installed packages actually have the required dependencies.
Edit: Here is an example to better illustrate the problem. Suppose package A
depends on packages B
and C
, both of which depend on package D
, possibly with different version ranges. When installing A
, pip will arbitrarily choose one of the specified version ranges for D
to satisfy. For example, B
's dependency on D
might be satisfied, but C
's required version for D
may not be satisfied. I want to detect whether such problems exist.
This answer shows how to get all locally installed packages. This one shows a way to verify that a given package is installed, which also recursively checks the package's dependencies. Combining these:
import pip
import pkg_resources
pkg_resources.require(str(dep.as_requirement())
for dep in pip.get_installed_distributions())
This will raise a ResolutionError
if any of the installed packages has an unsatisfied requirement.
Aside: This seems to be what the new pip check
command is intended to do, but it doesn't catch some dependency issues that the above script catches, such as extras-related issues.
Update: Recent versions of pip
do not provide a get_installed_distributions()
function. Per https://github.com/pypa/pip/issues/5243, pkg_resources.working_set
can be used to get the dependency list:
import pkg_resources
pkg_resources.require(str(dep.as_requirement())
for dep in pkg_resources.working_set)