Repository navigation
Land the webrtc framer transport crate (D2) - #721
Closed
MattJackson wants to merge 28 commits into
Closed
MattJackson wants to merge 28 commits into
MattJackson wants to merge 28 commits into
Conversation
…atagram door), unbuilt (WEBRTC D2)
…dropped-in door into the crate, add its tests The framer builds against str0m 0.24.1 (media data is an Arc<[u8]>). Every name of the media stack's DTLS interface moves into src/shim.rs (provider, certificate, profile mapping); the shim refuses any keying profile but AEAD-AES-GCM. The drain routes what the stack queued for sealing on every poll, the last one included, so SCTP reaches the far end. Address types come from core::net. The dropped-in door is this crate's cdylib under a dropped-in feature, as stdio's is, so the crate's closure stays busbar-contract plus its framing libraries. Tests: AES_CM refused (cipher and keying item), the aes/cipher/inout dependency closure, no DTLS type outside the shim, keys zeroised on drop (a watching allocator), and both doors through the loader's transport table, each with its RED arm.
…encies (manifest-allowlist)
MattJackson
enabled auto-merge
October 10, 2026 04:44
promote into
|
… drain its name from core Clippy (-D warnings): the stun-success arms in framing.rs and the audio MediaAdded arm fold their `if` into the match guard (collapsible_match); read_remote's return type is a `Remote` alias (type_complexity); door.rs's piece loop is a `while let` (while_let_loop). kind-isolation:transport-registration: the webrtc wire was composed in no place. The root links it as a `transport-webrtc` row on the transport and transport-door axes, behind an optional, off-by-default feature of the same name (the crate edge and the row together, as transport-ws). Its linked entry gains COMPOSES_OVER (none) and SESSION (true, its claim row's session) beside KEY and door. The shipped fold and session fixtures carry its row (both count rows a build leaves off); qa/kind-isolation.toml names the busbar -> busbar-transport-webrtc edge. kind-isolation:law0 (18 -> 20): the new crate made `webrtc` transport-instance vocabulary, so core's existing mentions became hits (busbar-core-connector 68 -> 76, busbar-contract 226 -> 230). Drained, not re-ceilinged: the connector's docs say "media association", the contract's composition tests use a made-up `dgram` wire over `udp`. Proved on Latchkey: clippy -p busbar-transport-webrtc --all-targets -D warnings; check -p busbar --features transport-webrtc --tests; test -p busbar body_chunk_bounds and transport_composed_over_consistency; gate kind-isolation and kind-isolation-ship (transport-registration PASS, law0 back to 18 findings, the rest at base); selftest kind-isolation and kind-isolation-ship; xtask plant_tests a_wire_composed_in_no_place_is_red_and_the_real_tree_has_none and a_manifest_row_is_the_wires_one_registration. The per-crate clippy rows (core-duplex-ws, loader-pack, ...) and the workspace selftest_runs_every_registered_gates_red_proof row share these causes and are left to CI.
MattJackson
added this pull request to the merge queue
Oct 10, 2026
MattJackson
removed this pull request from the merge queue due to a manual request
Oct 10, 2026
…rtc census id added Landing busbar-transport-webrtc teaches the census the transport id `webrtc`, so every `webrtc` in the composition root (a core-neutral crate) became a busbar x transport hit: 12 in tests/plane_transport_neutrality.rs's own source and 3 fixture rows (transport_fold.txt, transport_sessions.txt, session_transport_nouns.txt), figure 11 -> 26. The [[law0]] ceiling does not rise, so the cell is drained back to it: - plane_transport_neutrality.rs no longer spells the noun in its own source (its fixture already holds the list "so the scanner's own source spells no word it bans"): the self-test plants `mulaw` / `dtmf_digit_count` for the bare-word and snake_case cases instead, which proves the same two rules; the noun list and the detector are unchanged (-12). - The three fixture rows stay: the fold and session rows are read by the all-features build that links the wire, and the noun row mirrors the xtask twin's NOUNS. They are offset by three prose-only names of the line carrier's transport in the root's docs (serve_lines.rs x2, serve.rs x1), reworded to "line carrier"/"line transport" (-3). Not run locally: the Latchkey run budget (RUN_CAP) is spent; the count is reasoned from the matrix scanner (every non-Cargo.toml file under the crate, occurrences per line, `webrtc`/`stdio` unmasked) and the PR's CI is the proof for gate:kind-isolation and gate:kind-isolation-ship.
The red row (nextest busbar-kernel-egress::swrr_allocation:: a_warm_pools_selection_allocates_nothing) was a first-try FAIL that passed on retry: the run reported failed 0, and the turnstile reads a TRY 1 FAIL as a red row. Nothing in this PR touches the egress unit (src and test are byte-identical to predev). The witness was racy: its counting flag was process-global, so libtest's main thread, still filing the test it had just spawned, could allocate inside the measuring window on a loaded shard. Ruling: the witness measures the op, not the process. THE DESIGN A.8 (docs/design/BUSBAR-1.6.0.md l.3561: "A counting-allocator witness is 0 for each request-path op"; l.1285 "no allocation and no blocking on the hot path"). The flag is now thread-local. select_weighted runs wholly on the calling thread, so every allocation it makes is still counted and the assertion (== 0) is unchanged.
MattJackson
added this pull request to the merge queue
Oct 10, 2026
MattJackson
removed this pull request from the merge queue due to a manual request
Oct 10, 2026
…e-task-row-132b-webrtc-framer
…plane edges The webrtc framer adds six non-test .rs files under crates/, so the population (941 at base) is 947 and the one-file-short selftest plant no longer goes red: blocking-ffi, response-header and settings-leak selftests failed on it. population::FLOOR is a ratchet pinned at the measured count, so it moves to 947. The tracing selftest is fixed by merging predev (13a9ecb, the SCAN_FLOOR plant cuts with Overlay::below_floor). This branch touches qa/kind-isolation.toml, which arms kind-isolation:deps. That row was red for two standing edges with no rows: busbar-llm -> busbar-plane-llm (no [[dep]] row and no [[transitional]] drain row) and busbar-voice -> busbar-plane-streaming (no [[dep]] row). Both are legacy -> plane drain edges at count 1; they get not-allowed [[dep]] rows and the llm edge gets its [[transitional]] row, on the same terms as busbar-a2a -> busbar-plane-a2a. Not proved on Latchkey: busbar-release dev refused every row (run budget spent, RUN_CAP 4000/4000). Reasoned from the CI log of run 38094908998; the PR's CI is the proof.
Conflict in xtask/src/gates/population.rs (FLOOR): keep both sides. predev's computed one_file_short plant and its note stay; FLOOR stays at 947, the census measured with this branch's six busbar-transport-webrtc/src files, because a census floor "pins its ceiling at TODAY'S MEASURED VALUE" (BUSBAR-1.6.0.md:2090-2092). Predev's own floor guarantees >= 941 files there, so the merged tree holds >= 947.
Collaborator
Author
|
Claim released by busbar-release: the |
…door, chosen by rank busbar-transport-webrtc's conformance (2 tests) failed at `stated()`: "the cdylib exports busbar_plugin_door". `choose_cdylib` took `target/debug/libbusbar_transport_webrtc.*`, uplifted by an earlier plain build without the `dropped-in` feature, so that image has no door. The test build's own image, `deps/libbusbar_transport_webrtc.*`, does export it but was never a candidate: cargo names a workspace cdylib with no metadata hash, and a `cargo test` never uplifts a lib's cdylib (the lib is not a root unit), so the uplifted file is only ever another build's. The choice now ranks: the image CDYLIB_ENV names; the lib unit's own unhashed `deps/<lib>`; the uplifted image; the one hashed image. At every rank an image that does not load or exports no `busbar_plugin_door` is passed over, and only doorless images built is a failure naming them. Why this over a dedicated target dir: the suite is published to every plugin repo and runs inside their test binaries; building the image into a dir of its own would mean a nested cargo build from a test (a second cold build of the closure, a lock on the target dir), and `plugin-ci.yml` already names its image in CDYLIB_ENV. Probing the symbol decides by what an image IS, the door the dropped-in leg needs, never by which build ran last; the existing refusal of two different images at one rank stays. Tests: a doorless uplifted image is passed over for the deps image with the door, before and after a later plain build; doorless images only are refused naming them; the real probe tells the example cdylib from a file with no door. Fixed forward with the root cause (BUSBAR-1.6.0 §11.4, Part 2 #2: one table, compiled in or dropped in; Part 6: nothing waived).
Collaborator
Author
MattJackson
added a commit
that referenced
this pull request
Oct 12, 2026
…ows #721 carried; they are not this crate's edges, and the transitional row armed legacy-drain red
MattJackson
added a commit
that referenced
this pull request
Oct 13, 2026
webrtc framer transport crate on predev (row 132, piece 1; carries #721)
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Status row 132, piece D2: the webrtc framer transport crate and its dropped-in door, built and tested on predev. Starts from the WIP commit 185d3b2 (cherry-picked alone; D1 #706 had no overlapping hunks).
What changed after the WIP
Arc<[u8]>.src/shim.rs: the provider, the certificate, and the keying-profile mapping. The shim refuses any profile other than AEAD-AES-GCM. No DTLS engine is named anywhere.core::netinstead ofstd::net. These are address types only; the dep-wall gate bansstd::netin plugin source.*_streamsfields renamed to*_by_stream(the vocab gate read them as the streaming plane's name).dropped-infeature, the same waybusbar-transport-stdiodoes it. The WIP's separatebusbar-transport-webrtc-plugincrate depends on the logic crate, which breaches kind-isolation:closure (deps: bump schemars from 1.2.1 to 1.2.2 #40(a), OWNER-LOCKED: a plugin's workspace closure is busbar-contract alone). plugins.yaml keepscrate: busbar-transport-webrtc-plugin(pending_crate), as stdio's row does.Tests (each with a RED arm), all passing on Latchkey:
cargo test -p busbar-transport-webrtc(24 tests)tests/dependency_closure.rs): aes/cipher/inout are confined to the SRTP key derivation, both in Cargo.lock and in the source. RED arms: a second crate depending on aes, and the framing naming aes, are both refused.tests/zeroise.rs, a watching allocator). RED arm: the same key in a plain Vec is seen freed unzeroed.tests/no_dtls.rs). RED arms: an engine in the shim, and the DTLS interface outside the shim, are both refused.tests/conformance.rs, following export_conformance_tests): the linked and dropped-in doors state the same Statement and give equal transcripts. RED arm: the cdylib admitted against a different Statement is refused.Needs the owner
[package.metadata.busbar.linked]table. Linking a datagram framer into the busbar binary is a composition-root change; it may touch boot and golden fixtures, and the connector can't serve it yet. I left it out of D2. Either register it here or carry it with the connector piece.dimplunconditionally, so a DTLS engine is in the crate's build graph even though the framer never instantiates it. The tests check the framer's source, not its closure.