Report storage interfaces

The next major report-server API separates report bytes from the run index. Existing IBDTestDataStore implementations work for one migration release through LegacyDataStoreAdapter. New stores should implement IReportStore and IRunIndexStore in BDTest.Razor.Reports.Storage.

1. Move report JSON to IReportStore. Return a caller-owned stream or null; consume writes without closing the input. Propagate CancellationToken and overwrite existing keys.
2. Replace TestRunSummary in persistent APIs with RunIndexEntry and RunIndexCounts. Their JSON property names match existing summaries. ReportJsonContext.Default.RunIndexEntry supplies generated serialization metadata.
3. Implement paged ListAsync with newest start time first, ordinal ID tie-breaking, continuation tokens and optional exact branch/environment/tag filters. Upsert replaces an existing entry.
4. Register both IReportStore and IRunIndexStore with DI before building the host. Explicit new stores take precedence over IBDTestDataStore; registering only one throws.

For a staged migration:

Illustrative fragment

var adapter = new LegacyDataStoreAdapter(existingStore);
builder.Services.AddSingleton<IReportStore>(adapter);
builder.Services.AddSingleton<IRunIndexStore>(adapter);
builder.Services.AddBDTestReports().BindConfiguration("BDTest");

Registering builder.Services.AddSingleton<IBDTestDataStore>(existingStore) performs the same adaptation automatically. See Report options migration for the service-registration changes. Legacy operations without cancellation tokens can only be checked before and after they finish. Legacy listing still loads all summaries, and legacy raw downloads still serialize a loaded model. Native implementations remove those limitations.

Hosts implementing IDataRepository directly must update StoreData to (string reportId, BDTestReport model, CancellationToken cancellationToken), add tokens to history listing and deletion, and implement GetDataForBulkRead, OpenReadAsync and ListAsync. The old default interface implementation for bulk reads is removed. OpenReadAsync transfers stream ownership to its caller. HTTP raw downloads transfer ownership to MVC.

Hosted models now use BDTest.Contracts; the reader accepts schema version 1 and historical unversioned JSON. Raw downloads now preserve stored JSON exactly, including formatting and historical metadata. Report-list URLs remain unchanged. See Persistence for contracts and a blob-store example.

Nullable storage results

IBDTestDataStore.GetTestData now declares Task<BDTestReport?> for missing reports. Its StoreTestData parameter is BDTestReport?: the legacy adapter continues forwarding a JSON null root to the store. MemoryCacheBdTestDataStore preserves that behavior and reads the null model back as null. Update custom store signatures without adding null-forgiving operators at read sites.

IDataRepository.GetData and GetDataForBulkRead, and report/attachment OpenReadAsync implementations, also declare nullable results. Test the result before accessing report members or disposing/reading a returned stream.

Stored TestRunSummary and RunIndexEntry values preserve explicit JSON nulls for RecordId, Counts and optional metadata. Omitted counts retain their empty-object default. Conversion between summary and index models treats missing counts as zero. Storage writes require a non-null record ID even though deserialization may produce a missing one. These annotations do not change JSON names or historical formats.

Report deletion

Replace POST or DELETE /bdtest/report/delete/{id} with DELETE /bdtest/reports/{id} (replace /bdtest with the prefix selected by MapBDTestReports and include any host path base). The old route is removed. The new endpoint returns 204 No Content for both deleted and missing reports, rather than a 302 redirect. Other methods at the new endpoint return 405. Invalid IDs return 400; supported IDs contain 1–128 ASCII letters, digits, dots, hyphens or underscores, excluding . and ...

Custom browser controls must send the antiforgery request token and its cookie. Use the host's configured antiforgery header (RequestVerificationToken by default) and same-origin credentials, then refresh the list on 204. The named AdminAuthorizationPolicy controls permission to delete, in addition to read access; the removed AdminAuthorizer option is no longer supported. See Security and Custom delete controls.

Custom stores must support repeated deletes and return whether anything was deleted. The built-in repository uses that result to log missing reports through the host's ILogger services at Debug level. It deletes the persistent index before attachment content and report JSON, then attempts both cache invalidations even after failure or cancellation. It preserves the first error; a storage failure returns 500 and callers can retry. A failed index deletion must not delete report data. A later failure may leave unlisted data until a retry succeeds. Custom IDataRepository implementations should preserve the same ordering and coordinate history loading and cache-filling reads with deletion.

Compact Razor page models

Update overrides of Summary, TestTimesSummary, TopDefects and Warnings: their @model is now SummaryViewModel, TimingsViewModel, TopDefectsViewModel and WarningsViewModel, respectively, in BDTest.Razor.Reports.Models.ViewModels. ReportLayoutViewModel.Report now holds compact metadata, and its Scenarios property is replaced by aggregate fields including ScenarioCount. Construct layouts with ReportLayoutViewModel.Create(report) when rendering full-data views yourself.

Custom IDataRepository implementations can keep working through the new GetReportPages default implementation; it builds projections without caching. Override that method for caching, and invalidate projections on deletion/replacement. Existing custom sidebar providers still receive the full report through request-local data, without adding it to the shared view models. Report URLs and query filters are unchanged. Model writes use schema version 1; historical stored JSON remains readable.

See the view-by-view migration table and configure the separate positive ReportPageCacheSizeLimitBytes budget in Caching and limits.