Prefer importlib.metadata over pkg_resources if available

[sentrivana]

Let's not use pkg_resources and avoid a DeprecationWarning if importlib.metadata is available.

Fixes #2048

[sentrivana]

There's also dist.name (it's just an alias of dist.metadata["Name"]), but it hasn't always been there. dist.metadata["Name"] should work with all py versions.

[antonpirker]

If distributions() yields the same stuff as pkg_resources.working_set this looks good and can be merged.

[sentrivana]

If distributions() yields the same stuff as pkg_resources.working_set this looks good and can be merged.

@antonpirker It does with one difference: distributions() keeps the capitalization, working_set makes everything lowercase, e.g.:

# distributions()
MarkupSafe 2.1.2
matplotlib-inline 0.1.6
mock 5.0.2
packaging 23.1
parso 0.8.3
pexpect 4.8.0
pickleshare 0.7.5
pip 23.1.2
platformdirs 3.5.0
pluggy 0.13.1
prompt-toolkit 3.0.38
ptyprocess 0.7.0
pure-eval 0.2.2
py 1.11.0
Pygments 2.15.1

# working_set
markupsafe 2.1.2
matplotlib-inline 0.1.6
mock 5.0.2
packaging 23.1
parso 0.8.3
pexpect 4.8.0
pickleshare 0.7.5
pip 23.1.2
platformdirs 3.5.0
pluggy 0.13.1
prompt-toolkit 3.0.38
ptyprocess 0.7.0
pure-eval 0.2.2
py 1.11.0
pygments 2.15.1

Do you see this creating issues anywhere? I grepped for where we actually use the modules and it seems like we attach them to the event and check them here but elsewhere we seem to be using sys.modules instead.

[antonpirker]

@sentrivana could you please add a test for testing _generate_installed_modules?

[sentrivana]

@antonpirker Changed the PR a bit and added a test, could you please skim over to see if it still looks ok to you?

[antonpirker]

Why do we need to replace the "-" with a "_"?
Wouldn't we then report a package name "sentry_sdk" that is different than the one the user installs "sentry-sdk"?

[sentrivana]

There's a weird mismatch between importlib and pkg_resources where they report the package typing-extensions differently; importlib reports typing_extensions, while pkg_resources reports it as typing-extensions. This is the only package that is different (if we normalize to lowercase, which we're doing). I'll remove the .replace() and change the test to ignore the one package.

[antonpirker]

I guess if we just try to keep as much the same as it was reported with pkg_resources we are good to go.