The upstream Grafana TF provider has versioned resources like: https://registry.terraform.io/providers/grafana/grafana/latest/docs/resources/apps_dashboard_dashboard_v2beta1 I want to generate versioned resources like it is done in the Kubernetes provider. Converting the resource above, I am able to set the Pulumi type identifier to

grafana:apps/v2beta1:Dashboard

Via

resources.go

, I configure the language sections with module mappings. For instance, my

schema.json

contains the following for `nodejs`:

"nodejs": {
            ...
            "moduleToPackage": {
                "apps/v2beta1": "apps/v2beta1"
            },
            ...
        },

The generated NodeJS SDK however has file

sdk/nodejs/apps/dashboard.ts

, while I expected

sdk/nodejs/apps/v2beta1/dashboard.ts

. What am I missing?

So I don't think a module name mapping of X => X does anything. per-chance does your schema set a moduleFormat regex?

My attempt so far: https://github.com/pulumiverse/pulumi-grafana/pull/383/changes/babf79108e698f5e1553cbd507837f82624351b2

hmm I think this might be a bridge thing, gimme a min

So I don't think a module name mapping of X => X does anything.

Why not? The codegen behaves differently for

grafana:apps/dashboard:Dashboard

and

grafana:apps/v2beta1:Dashboard

. Without mappings, both generate to

grafana.apps.Dashboard

(NodeJS), while I want the latter to be remapped properly. By accident, the Pulumi module name and the language package name are the same. My expectation is that it would still do its magic.

So looking at the code whenever we take values from the modulesToPackages mapping it's just to replace the module name that was mapped. So if we have X and we look up X in the map and get back X nothing has. changed. I think what's happening with tfbridge though is it defaults the module format regex to

(.*)(?:/[^/]*)

so it's pulling off that /dasboard or /v2beta1 before doing any further module naming

Isn't SDK generation invoking

pulumi package gen-sdk

feeding it

schema.json

? Asking because my

schema.json

contain the correct Pulumi resource types and mappings sections in the language configuration sections.

what's the schema.json say?

https://raw.githubusercontent.com/pulumiverse/pulumi-grafana/babf79108e698f5e1553cbd50783[…]24351b2/provider/cmd/pulumi-resource-grafana/schema.json The mappings are in the language section at the top. Search for

grafana:apps/v2beta1:Dashboard

as the resource type.

Yeh so I think "grafanaapps/v1beta1Dashboard" should be "grafanaapps/v1beta1/dashboardDashboard" to be in alignment with other resources like "grafanaalerting/recordingRuleV0Alpha1RecordingRuleV0Alpha1" note how the resource name is duplicated into the module name but then ignored by the rest of codegen

Leaving the mixed naming setup in this grafana provider aside for a minute, why is the versioned naming working for the Kubernetes provider, even with X=>X mappings like

"policy/v1": "policy/v1",

? https://raw.githubusercontent.com/pulumi/pulumi-kubernetes/refs/heads/master/provider/cmd/pulumi-resource-kubernetes/schema.json My understanding of SDK generation has been so far that one needs to get the proper information into

schema.json

, usually done by the

gen

specific command (

pulumi-gen-kubernetes

,

pulumi-tfgen-grafana

). Once at that point, actual SDK generation is triggered via

pulumi package gen-sdk

. Is there still something specific to SDK generation which is different in these

gen

commands?

because k8s has a different module format set in its schema

Are you referring now to some config option? I tried dotted notation too in the Grafana provider, e.g

grafana:<http://apps.grafana.com/v2beta1:Dashboard|apps.grafana.com/v2beta1:Dashboard>

and matching mappings, but the generated type ended up being

grafana.apps.grafana.com.Dashboard

In the schema meta options there's a "moduleFormat" setting

🤔 I can't find

moduleFormat

in the Kubernetes

schema.json

. I'll try your suggestion, but I had hoped to understand how Pulumi resource type to SDK type generation works. Understanding why it works for the Kubernetes provider would have brought me a step further in understanding.

Ah k8s probably uses the default which is '(.*)'

Ah, so K8s using the default

moduleFormat

, while my bridge provider has

"moduleFormat": "(.*)(?:/[^/]*)"

💡

It's pretty much just a back compat thing for tfbridge, we wouldn't recommend using it for anything now