Fix to restore library_path as absolute path

[maiself]

Fix for possible issue with library_path being set as a local resource path rather than an absolute file system path. May have broken the Python bindings, and possibly any other extension needing an absolute path. Needs a little more investigation, marking as draft for now. @dsnopek, could you look this over too?

Related:

• Change GDExtension's library_path back to an absolute path #84620
• Add renaming of PDB files to avoid blocking them #87117
• GDExtension: Fix missing library path breaking hot reloading #90961

[dsnopek]

Ack! Thanks for catching this. As you point out (via linking PR #84620), this isn't the first time we've accidentally broken this. I think we should probably add a test case to the godot-cpp tests to make sure the path is an absolute path, so we don't break it again.

The change in your PR looks fine to me, although, I haven't done any testing.

[maiself]

I think we should probably add a test case to the godot-cpp tests to make sure the path is an absolute path, so we don't break it again.

I was thinking last night about adding something like a ERR_FAIL_COND_V_MSG to this function to catch this kind of mistake earlier if it should happen again. And possibly rename some variables for clarity, but that's getting more involved.

I'm not sure what a godot-cpp test case for this world look like, if you think that's the best way to address this can you put that together?

[dsnopek]

I just posted PR godotengine/godot-cpp#1520 which adds a godot-cpp test.

I was thinking last night about adding something like a ERR_FAIL_COND_V_MSG to this function to catch this kind of mistake earlier if it should happen again.

I'm not sure that makes sense in this case. Those error macros are generally used for errors in input to a function, whereas this is an error of output. We also won't necessarily see that error during development (ie it wouldn't fail CI on PRs that break this), but if godot-cpp starts failing tests, then we know there's an issue to look into.

[dsnopek]

Oh, and in working on that test, I did test that it fails without this PR and succeeds with it, so I have now tested this PR as well!

Can you take it out of draft?

[maiself]

All sounds good to me. Good point about the error messages and input vs output.

[akien-mga]

Thanks!