Fix channel generation and dashboard reconciliation

Author: bruno-espino

Makefile

@@ -12,10 +12,10 @@ TERRAFORM_VERSION_VALID := $(shell [ "$(TERRAFORM_VERSION)" = "`printf "$(TERRAF
 
 export TERRAFORM_PROVIDER_SOURCE ?= pydantic/logfire
 export TERRAFORM_PROVIDER_REPO ?= https://github.com/pydantic/terraform-provider-logfire
-export TERRAFORM_PROVIDER_VERSION ?= 0.1.4
+export TERRAFORM_PROVIDER_VERSION ?= 0.1.5
 export TERRAFORM_PROVIDER_DOWNLOAD_NAME ?= terraform-provider-logfire
 export TERRAFORM_PROVIDER_DOWNLOAD_URL_PREFIX ?= https://github.com/pydantic/terraform-provider-logfire/releases/download/v$(TERRAFORM_PROVIDER_VERSION)
-export TERRAFORM_NATIVE_PROVIDER_BINARY ?= terraform-provider-logfire_v0.1.4
+export TERRAFORM_NATIVE_PROVIDER_BINARY ?= terraform-provider-logfire_v0.1.5
 export TERRAFORM_DOCS_PATH ?= docs/resources
 
 

UPSTREAM_UPJET_NULL_ID_ISSUE.md

@@ -0,0 +1,171 @@
+## Upstream Upjet issue draft: file-backed workspace should treat `id: null` as empty state
+
+### Summary
+
+`provider-upjet-logfire` has a local dashboard connector wrapper because Upjet's
+file-backed Terraform workspace flow does not handle a Terraform state whose
+resource attributes contain `id: null`.
+
+For `IdentifierFromProvider` resources, Upjet seeds Terraform with an empty ID
+before the provider has created the external resource. A later
+`terraform apply -refresh-only` can convert that empty-ID state into a persisted
+null-state file. On the next reconcile, Upjet reopens that file and fails with:
+
+```text
+cannot work with a non-string id: <nil>
+```
+
+The correct upstream behavior is to treat `id: null` the same way it treats an
+empty or missing ID: the resource should be considered absent, and the workspace
+should proceed with a clean create flow.
+
+### Why this is an Upjet bug
+
+- Direct Terraform succeeds against the same provider and backend.
+- Direct Pulumi succeeds against the same provider and backend.
+- The failure only appears in Upjet's file-backed workspace lifecycle.
+- The Terraform provider's dashboard `Read` behavior is valid for normal
+  Terraform usage: if the state ID is empty, it removes the resource from state.
+
+Relevant local files:
+
+- `terraform-provider-logfire/internal/provider/dashboard_resource.go`
+- `provider-upjet-logfire/config/external_name.go`
+- `provider-upjet-logfire/internal/controller/dashboardconnector/connector.go`
+
+### Environment used to reproduce
+
+- Upjet: `github.com/crossplane/upjet/v2 v2.2.0`
+- `provider-upjet-logfire`: local controller build with current repo changes
+- Terraform provider framework: `github.com/hashicorp/terraform-plugin-framework v1.17.0`
+- Reproduced with a real Logfire API key and live backend resources
+- Reproduced both in a Crossplane cluster and with a standalone local Terraform
+  state file
+
+### Minimal reproduction shape
+
+This is easiest to hit with a resource that uses `config.IdentifierFromProvider`
+and whose Terraform provider removes state when `Read` receives an empty ID.
+
+In `provider-upjet-logfire`, `logfire_dashboard` is one such resource:
+
+```go
+var ExternalNameConfigs = map[string]config.ExternalName{
+    "logfire_dashboard": config.IdentifierFromProvider,
+}
+```
+
+And the Terraform provider's `Read` path does this:
+
+```go
+if state.ID.IsNull() || state.ID.IsUnknown() || state.ID.ValueString() == "" {
+    resp.State.RemoveResource(ctx)
+    return
+}
+```
+
+### Concrete reproduction
+
+1. Seed a Terraform state file for a provider-assigned-ID resource with
+   `attributes.id == ""`.
+2. Run `terraform apply -refresh-only`.
+3. Terraform rewrites the state file so the resource still exists structurally,
+   but its attributes become null, including `id: null`.
+4. Let Upjet reopen that file-backed workspace.
+5. Upjet fails reopening/reconciling the resource because it expects `id` to be
+   a string, not null.
+
+Observed state after `terraform apply -refresh-only`:
+
+```json
+{
+  "version": 4,
+  "terraform_version": "1.14.8",
+  "resources": [
+    {
+      "mode": "managed",
+      "type": "logfire_dashboard",
+      "name": "test",
+      "instances": [
+        {
+          "schema_version": 0,
+          "attributes": {
+            "definition": null,
+            "id": null,
+            "name": null,
+            "project_id": null,
+            "slug": null
+          }
+        }
+      ]
+    }
+  ]
+}
+```
+
+### Observed runtime behavior
+
+In Crossplane/Upjet, the resource does not cleanly transition to "does not
+exist". Instead, a later reconcile trips over the persisted null-state file and
+fails before the normal create path can run again.
+
+The provider-local workaround in
+`provider-upjet-logfire/internal/controller/dashboardconnector/connector.go`
+does this:
+
+- inspect `/tmp/<uid>/terraform.tfstate`
+- detect `attributes.id == null`
+- delete that state file
+- treat the observe cycle as `ResourceExists=false`
+
+With that wrapper in place, live Crossplane tests succeeded:
+
+- `Project` reached `Ready=True`, `Synced=True`
+- `Dashboard` reached `Ready=True`, `Synced=True`
+
+### Expected upstream behavior
+
+Upjet should treat a file-backed Terraform state with `id: null` as empty state.
+
+That likely means one or both of these behaviors need to change upstream:
+
+1. State-empty detection should consider `id: null` equivalent to no resource.
+2. Workspace refresh/existence logic should not report the resource as present
+   when the decoded state only contains null attributes.
+
+Likely upstream inspection points:
+
+- `pkg/terraform/files.go`
+- `pkg/terraform/workspace.go`
+
+The important rule is:
+
+> If the decoded resource state has a missing or null identifier, Upjet should
+> treat it as absent and recover to a clean create/observe path instead of
+> failing on type assertions or state reopening.
+
+### Local workaround kept in `provider-upjet-logfire`
+
+Until Upjet is fixed upstream, `provider-upjet-logfire` keeps a narrow wrapper
+for dashboards:
+
+- `provider-upjet-logfire/internal/controller/dashboardconnector/connector.go`
+- `provider-upjet-logfire/internal/controller/dashboardconnector/connector_test.go`
+
+This is a containment fix, not the preferred long-term solution.
+
+### Suggested upstream regression test
+
+Add a test around the file-backed workspace/state-empty path that feeds Upjet a
+state whose resource attributes include `id: null`, then assert:
+
+- no error is returned while reopening/refreshing the workspace
+- the resource is treated as absent
+- a subsequent create flow is allowed to proceed normally
+
+### Notes on unrelated generation behavior
+
+`provider-upjet-logfire` will keep `make generate` as the standard generation
+entry point. Other standard Upjet providers also rely on Makefile-exported
+variables for the scraper and docs sync rather than supporting bare
+`go generate` as the primary path.