Alias the StreamOne payload column so Select() projections keep their ETag

[jeremydmiller]

Fixes #5158. Also closes out the last untested claim behind #5120.

The defect

WriteSingle / StreamOne<T> threw at runtime whenever the queryable carried a Select() projection and emitETag was on (the default). Regression from #5027 / PR #5030 (44f9b44a2), shipped in 9.18.0.

Dumping the generated SQL for three shapes of the same query showed two distinct failure modes, not the one reported:

plain   OK      select d.data, d.mt_version as mt_etag_version from ... mt_doc_issue as d where d.id = :p0 LIMIT :p1;
anon    THREW   select  jsonb_build_object('Description', d.data ->> 'Description') , d.mt_version as mt_etag_version from ...
scalar  THREW   (no SQL — threw before the command was built)

Mode 1 — the lost data alias. VersionSelectClause.Apply rebuilds the select list from Inner.SelectFields() instead of delegating to the inner clause's own Apply, so any aliasing Apply would have emitted is dropped. DataSelectClause's field is the literal d.data, so Postgres names the column data anyway and the plain path worked by luck. SelectDataSelectClause's field is a jsonb_build_object(...) expression, so rebuilt without the alias the column came back named jsonb_build_object and the reader's GetOrdinal("data") threw IndexOutOfRangeException.

Mode 2 — scalar projections. Select(x => x.Description) makes T a primitive, and FindMapping(typeof(string)) throws ArgumentOutOfRangeException: This type cannot be used as a Marten document before any SQL is built.

The fix

1. VersionSelectClause aliases the payload explicitly to data, making the name intentional for every inner clause rather than incidental for one. DocumentTable.SelectColumns guarantees the payload is the first selected field — "the order of the selection is data, id, everything else" — so only field 0 is aliased and the remaining fields pass through untouched, keeping their own names. That matters: a revisioned document's storage already selects d.data, d.mt_version, and aliasing more than the payload would have broken it.
2. MartenLinqQueryProvider.StreamOneWithVersion<T> resolves the mapping from the source document type (parser.DocumentTypes().First()) rather than from T. Under a projection T is the projected type, and asking StorageFeatures for a mapping of it either invents one whose version metadata defaults to enabled or throws for a primitive. The version being read is the source document's either way.

Net effect: projections keep their ETag rather than merely not crashing. The projection is a pure function of the document, so it validates against the same version — a client can cache either representation off the same tag.

Tests

• stream_one_over_a_select_projection_serves_the_projection_and_the_document_etag — Alba e2e over a new /minimal/issue/{id}/summary endpoint: the projected body is served, its ETag equals the full document's, and that tag drives a 304.
• stream_one_over_an_anonymous_type_projection_emits_the_document_etag — the exact shape from the report. Cannot go through an endpoint (the result type must be nameable), so it runs against WriteSingle directly.
• stream_one_over_a_scalar_projection_emits_the_document_etag — mode 2.
• stream_one_over_a_select_projection_of_a_revisioned_document_emits_the_revision — the projection path on the numeric-revision flavor.
• stream_one_over_a_select_projection_of_a_versionless_document_emits_no_etag — the source type still decides, so a projection over a versionless document must not pick up a default from the projected type.

Closing out #5120

#5120's proposal shipped in full in #5121, but one load-bearing claim in it was asserted in a code comment and never tested:

For a SingleStreamProjection target the resulting ETag is semantically identical to StreamAggregate's (both are the stream version), so the two read styles stay cache-coherent.

That is the entire justification for the feature. stream_one_and_stream_aggregate_serve_the_same_etag_for_the_same_stream now hits both endpoints for the same stream, compares their ETags at version 1 and version 2 (so a coincidental match at "1" cannot pass), and confirms a tag minted by either style is honored by the other with a 304. With that pinned, #5120 can be closed.

Verified locally (net10.0)

Marten.AspNetCore.Testing · LinqTests 1421/1421 · DocumentDbTests 1089 passed (1 pre-existing skip) · CompiledQueryTests 9/9 · EventSourcingTests.

Worth noting one course-correction during development: a first attempt guarded with CanCarryVersion (skip the version column for multi-field select clauses) and silently dropped the ETag for IRevisioned/ILongVersioned documents, whose storage selects data and mt_version. Two existing tests caught it, and it was replaced with the narrower first-field aliasing above.

🤖 Generated with Claude Code