Review: Helpers, Sweepers, Data Sources, List Resources

Assume the @maintainer persona. Scope: finders, status, waiters, sweepers, data sources, list resources. Loaded from review-pr.

Finders, status, waiters

• Finder signature: find<Resource>ByID(ctx context.Context, conn *<svc>.Client, id string) (*awstypes.<Type>, error). Variants by ARN/Name use the same shape.
• On *awstypes.ResourceNotFoundException, return smarterr.NewError(&retry.NotFoundError{LastError: err}).
• Status function reuses the finder and returns retry.StateRefreshFunc. Design status so one function powers create, update, and delete waiters.
• Waiters use retry.StateChangeConf. Created/updated waiters typically set NotFoundChecks: 20 and ContinuousTargetOccurence: 2. Deleted waiters use empty Target with Pending covering deletion-in-progress states.
• Prefer SDK-provided status constants (e.g. awstypes.StatusInProgress) over package-level string consts.
• Finders and ResName<Name> constants referenced in tests must be re-exported via exports_test.go.

Flag finders that return raw errors (must wrap with smarterr.NewError), status that duplicates finder logic, or hand-rolled polling loops in place of retry.StateChangeConf.

Sweepers

Each new resource needs a sweeper. Iterate the SDK paginator, build via framework.NewSweepResource(new<Resource>Resource, client, framework.NewAttribute(names.AttrID, aws.ToString(v.<Thing>Id))) (where framework is internal/sweep/framework), and register in the package's sweep.go with awsv2.Register("aws_<svc>_<thing>", sweep<Resource>s, ...optionalDeps). Pass multiple framework.NewAttribute(...) arguments for composite identity.

Flag new resources without a sweeper, sweepers that don't propagate paginator errors via smarterr.NewError, and sweepers using import aliases other than framework for internal/sweep/framework.

Data sources

Data sources have only a Read method.

• Use framework.DataSourceWithModel[T].
• Schema attributes are Required or Optional for search criteria; everything else is Computed.
• Attributes that are Required on the corresponding resource are typically Computed on the data source unless they form lookup criteria.
• No configurable timeouts.
• Tagged data sources expose a single computed tags attribute (no tags_all).

List resources

Framework path embeds the corresponding underlying resource. SDKv2 path uses framework.ListResourceWithSDKv2Resource.

The List method:

1. Get the client.
2. Fetch the config (only when the list takes query attributes such as a parent ID).
3. Stream results from a paginated AWS List API.
4. Set logging fields per item (typically the ARN) via tflog.SetField(ctx, logging.ResourceAttributeKey(...), ...).
5. Set identifying attributes for each result.
6. Set result.DisplayName to a human-readable identifier (typically the resource name).

The listing helper uses an iterator over the SDK paginator:

func list<Thing>s(ctx context.Context, conn *<svc>.Client, input *<svc>.List<Thing>sInput) iter.Seq2[awstypes.<Thing>, error]

The flatten function shared by Read and List lives in the resource file (r.flatten for Framework, resource<Name>Flatten for SDKv2).

Flag list resources that don't set DisplayName, that re-implement flatten logic instead of sharing with Read, or that omit the tflog.SetField per-item logging hook.