possible fix for develop attribute with export-pkg

[memsharded]

Changelog: BugFix: Set the self.develop=True attribute for recipes when they are used with conan export-pkg, in all methods, it was previously only setting it for the package() method.
Docs: Omit

Fix #6583

This is a exploratory PR, to see if something breaks. Not guaranteed it will be merged, need to understand better if it could break something.

[jgsogo]

I see it is a small change, but it modifies the graph that is created: from a consumer one to a test_package one. I can imagine an scenario where expanding the graph to the test_package will download more recipes or even fail if using a graphlock.

conan/conans/client/graph/graph_manager.py

Line 135 in 9940169

if create_reference: # Test_package -> tested reference

... nevertheless, I would say it has low risk because affects only to one command and can be merged, but I think we should improve the "tricky, but works" part, can we?

[jgsogo]

If we are merging this, I think we cannot trust in this trick anymore. Now, with the test_package the root has two neighbours and I don't know if the nodes[0] will be always the one we want (I cannot run Conan now, and having a look to the sources in the browser is a bit harder...)

[memsharded]

I see it is a small change, but it modifies the graph that is created: from a consumer one to a test_package one. I can imagine an scenario where expanding the graph to the test_package will download more recipes or even fail if using a graphlock.

No, it is not true. The graph is not modified at all, only the loader.dev_reference is assigned, that is the only effective change.

[jgsogo]

What I see is that, from the _load_root_node we start to call

return self._load_root_test_package(path, create_reference, graph_lock, profile)

instead of

root_node, ref = self._load_root_consumer(path, graph_lock, profile, graph_info.root)

but yes, having a look to the sources it looks like there is not much difference between those two functions.

Ok then

[memsharded]

What I see is that, from the _load_root_node we start to call

Nop. Check the conditional:

 if isinstance(reference, ConanFileReference):
            return self._load_root_direct_reference(reference, graph_lock, profile)

[jgsogo]

OMG, there is another if before to call a different _load_xxx! 😰

Now I see, this is almost changing nothing 👍