1. Terraform makes a distinction between "modules" (reusable groups of resources) and "root modules" (things you might actually deploy). But Terraform doesn't really, uh, support that notion https://github.com/hashicorp/terraform/issues/28490 . Do we want to make that distinction too, and implement a new

terraform_deployment

target (we'd probably want this as well for

experimental-deploy

)?

2. Terraform doesn't really have a notion of a "Resolve" the way Pants does. Every root module effectively has its own universe of dependencies, and there isn't a way of joining multiple modules' requirements. We could do something weird like synthesise a Terraform file that tries to depend on multiple modules, or manually combine dependencies ourselves, but that sounds bad. Would it be fine to just synthesise a resolve per "root module" with the name being the path to the module or something based off it?

Re 2. Yes, I think doing everything per root module makes sense. I don't think we should get in the way of TFs idioms or try and jam pants-isms into them, unless it adds substantial value. In this case I think people are used to the top-level thing being a root module.

But why do we need to synthesize a resolve at all? What would that represent?

I didn't realize we were tying this in to generate-lockfiles, but makes sense

so maybe the "resolve" name should be, as you say, the relpath from the repo root to the root folder of the resource

./pants generate-lockfiles --resolve="src/terraform/us-east-1/network"

or whatever