[cli/import] - Handle panic during import codegen

[komalali]

Description

This PR is to handle a panic that is caused by a failure during go codegen when using the pulumi import command. Unfortunately, import is impossible to write tests for since it is dependent on a resource existing in a provider. The original PR to add the import command included no tests.

I have also been unable to repro this particular failure manually since I do not have the correct access to use the PushRestrictions feature for BranchProtection in Github as I do not have an organization using the team or enterprise tiers.

I tested this change by manually adding a panic to the generateImportedDefinitions function and observing the resulting output.

Related to: #7112

Checklist

• I have added tests that prove my fix is effective or that my feature works

• I have updated the CHANGELOG-PENDING file with my change

• Yes, there are changes in this PR that warrants bumping the Pulumi Service API version

[lukehoban]

handle a panic that is caused by a failure during go codegen

Is there a reason we can't have codegen return a real error instead of panicng? It wasn't entirely clear here what call in the code here is panicing, and why that is expected such that we need to add a panic handler at this layer?

Assuming there is a good need to still be panicing in the codegen, this change certainly looks like it helps improve the UX for pulumi import.

Unfortunately, import is impossible to write tests for since it is dependent on a resource existing in a provider. The original PR to add the import command included no tests.

Technically I believe we could import a resource like random.RandomString which doesn't have any backing cloud provider state (or have a "dummy" resource provider that we test against). But I agree we can tackle test coverage here independently of this PR.

[komalali]

It wasn't entirely clear here what call in the code here is panicing, and why that is expected such that we need to add a panic handler at this layer?

So, this is the line that panics:

pulumi/pkg/codegen/go/gen_program.go

Line 92 in 3e3e4f2

panic(errors.Errorf("invalid Go source code:\n\n%s", index.String()))

You're right though, it should just return an error. The reason that I didn't fix it at that layer is because docgen has existing code to work around this, and I wanted to avoid working in multiple layers and potentially alter the behavior of existing tooling to get to resolution here.

The reason we're getting to this panic, in this case, is not at the codegen layer, it's something that happens higher up the chain (possibly with the readResource call and the state generated) but I have been unable to investigate that since I don't have the correct access to this resource to observe the panic myself.

Technically I believe we could import a resource like random.RandomString which doesn't have any backing cloud provider state (or have a "dummy" resource provider that we test against). But I agree we can tackle test coverage here independently of this PR.

Yes, but there is no guarantee that any of those imports would lead to the error that we see here, which I can't get to a root cause for because of lack of access to the resource.

In any case, I was trying to resolve the panic without diving into multiple other layers when I can't even repro the error.