Sans-I/O TDS core (4b/N): invert PLP/LOB column streaming to the sync core

[Saurabh Singh (saurabh500)]

Summary

Layer 4b of the sans-I/O restructure. Inverts the eager PLP/MAX column collect path to run synchronously over the in-memory PacketBuffer, reusing the existing PlpChunkStreamReader resumable state — no new state machine. The async decode_into PLP branch and the buffer-driven sync driver converge on one shared framing body (two refill drivers differing only by .await).

Base: dev/saurabh/sans-io-l4a-fixed-width-step (frozen tip 8bddc4fc).

Related Issues

N/A — internal sans-I/O refactor, no tracking issue (consistent with #189/#191/#194/#195).

What changed

• Factor byte-source-agnostic framing leaves out of PlpChunkStreamReader (decoder.rs): classify_length, on_chunk_header, note_read, chunk_remaining, reached_end, total_read. The async begin/ensure_active_chunk/read_into are rewritten to call these leaves — single-copy framing, no drift.
• Sync PLP driver (sync_decoder.rs): plp_collect_step + PlpProgress step the PlpChunkStreamReader over an owned PacketBuffer one chunk at a time — ensure(4) for a chunk header or a slice of the current chunk body, never the whole value — so wire residency stays bounded to ≲ one packet regardless of total value size. NeedMore carries the absolute header/body width so refill always makes forward progress across packet-straddling prefixes.
• Reader seam (packet_reader.rs): collect_plp_bytes — async framing default for non-buffer readers, sync driver override on PacketReader and NetworkTransport. All GenericDecoder::collect_plp call sites rewired through the seam; value arms unchanged → byte-identical.
• Always-Encrypted over varbinary(max) (token_stream.rs): collect ciphertext via the sync seam then decrypt_cipher_value, folding collect+decrypt into the sync path (mirrors L4a's non-PLP AE fold).

Scope boundary

text/ntext/image are is_plp() == false legacy text-pointer LOBs (not the PLP chunk protocol) and stay on the legacy async path, unchanged and byte-identical — consistent with L4a's boundary. (Coordinator-tracked as a separate p4d follow-up.)

Tests (buffer-driven PacketReader)

• plp_single_chunk_mid_chunk_resume_is_byte_identical_across_refill_boundary — refill boundary swept across every interior offset incl. inside the 8-byte PLP header and 4-byte chunk-length prefix.
• plp_multi_chunk_across_refills_is_byte_identical_with_bounded_residency — multi-chunk to the zero-length terminator + a large value exceeding the 8192-byte buffer capacity decoded under packet-sized refills, proving chunk-bounded residency by construction.
• fully_sync_mixed_row_with_plp_is_byte_identical_across_refill_boundary — [int4][varchar][varbinary(max)] with every cell on the sync core, exact values asserted across every refill offset.

Validation

• cargo btest (nextest) terminates; FAIL name-set byte-for-byte identical to base (the 7 known cert fixtures).
• cargo bfmt + cargo bclippy clean; scripts/bfmt.ps1 + scripts/bclippy.ps1 clean (mssql-py-core builds).

No PLP framing logic duplicated; no new resumable machine introduced.

Co-authored-by: Copilot App 223556219+Copilot@users.noreply.github.com
Copilot-Session: e2c378f8-3ba1-4b48-9ebe-5a4ea2bd2761

[Copilot]

Pull request overview

Moves eager PLP/LOB collection into the synchronous sans-I/O core while preserving shared framing state.

Changes:

• Adds synchronous chunk-at-a-time PLP collection.
• Routes buffer-backed readers through the sync PLP path.
• Integrates PLP Always Encrypted decoding and boundary tests.

Reviewed changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
mssql-tds/src/datatypes/decoder.rs	Exposes shared PLP framing operations.
mssql-tds/src/datatypes/sync_decoder.rs	Adds synchronous PLP collection steps.
mssql-tds/src/io/packet_reader.rs	Adds the PLP collection seam and test-reader override.
mssql-tds/src/connection/transport/network_transport.rs	Implements production buffer-driven PLP collection.
mssql-tds/src/io/token_stream.rs	Routes PLP decryption and adds refill-boundary tests.

Suppressed comments (1)

mssql-tds/src/io/token_stream.rs:1937

• This is another copy of the PLP row-decoding helper already added at lines 1656-1675 and 1820-1839. Reuse a module-level test helper instead of adding a third implementation so setup and protocol assertions cannot drift between boundary tests.

        async fn decode(read_data: Vec<u8>, columns: &[ColumnMetadata]) -> Vec<ColumnValues> {

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[Saurabh Singh (saurabh500)]

Closing: the sans-I/O restructuring is not going to be productionized.

The row-decode performance work that motivated much of this has been re-scoped around #247, where benchmarked spikes show the dominant win comes from a far smaller change — converting TdsPacketReader to RPITIT to remove per-read boxing (#252, measured at −61.6% decode time) — rather than from inverting the core to a sync step() driver. Given that, this stack is a large amount of surface area for a win that is already available more cheaply.

This is cleanup rather than a rejection of the analysis. The branch is deliberately not deleted, so the work remains recoverable if the direction is revisited.