chore: bump IronRDP pin 879ffed8 → a5d1c682 (−2 vendored forks, −1 divergence) - #178
Merged
Conversation
…etire rdpeusb [WIP] Bump all 15 git-pinned ironrdp crates 879ffed8 -> a5d1c682 (upstream master 2026-08-03) and the [dependencies] version constraints to match (ironrdp 0.14->0.17, server 0.10->0.13, acceptor 0.8->0.10, pdu 0.7->0.9, core 0.1->0.2, rdpsnd 0.7->0.9, rdpdr 0.5->0.7, egfx 0.1->0.3, dvc 0.5->0.8, svc 0.6->0.8, cliprdr 0.5->0.7, displaycontrol 0.5->0.8, connector/tokio 0.8->0.10). Retire the ironrdp-rdpeusb fork: all four divergences merged upstream (#1418/#1513/#1420/#1321), so its [patch] entry becomes a git-pin (upstream is publish=false) and vendor/ironrdp-rdpeusb/ can be deleted. WIP: the kept vendored forks (server/acceptor/rdpdr/dvc/rdpeudp) still declare the old crate versions + APIs, so the tree does not resolve/build yet — the per-fork re-vendor + src re-wire follows in subsequent commits.
…d1c682 - rdpeudp (macrdp-authored, no upstream twin): bump its ironrdp-core dep 0.1->0.2 + its standalone [patch] rev. Builds clean, all 53 crate tests pass (the additive-core keystone holds). - rdpdr: ZERO upstream source churn 879ffed..a5d1c682 (only Cargo.toml/CHANGELOG), so the fork source was already current; bump package version 0.5->0.7 + dep versions (core 0.2/error 0.2/pdu 0.9/svc 0.8) + rust-version 1.94. Added a standalone [patch] for isolated validation. Builds clean. - dvc: real re-vendor (upstream churned lib/client/server). Refreshed those three files verbatim from a5d1c682, kept pdu.rs/complete_data.rs (upstream-unchanged, carry div1). Re-applied div1 (Soft-Sync: SoftSyncResponse arm + get_channel_id_by_name in server.rs against the new DynamicChannelAllocator, SoftSyncRequest arm in client.rs). RETIRED div2 (the processor.close hook) — upstream #1302 now owns it via Drop + remove_by_channel_id. Bump version 0.5->0.8 + deps, drop unused slab, add standalone [patch]. Builds clean. Each validated by a standalone 'cargo build' against the a5d1c682 git deps. Remaining: acceptor (M3c re-apply) + server (the bulk) + src/ + workspace build.
Refresh connection.rs verbatim from a5d1c682 and re-apply only the surviving divergences (Approach B — the authoritative upstream base + macrdp-only additions), because upstream absorbed far more than the dry-run mapped: RETIRED (now upstream in a5d1c682, NOT re-applied): - div(1) honor-size incl. the #1404 operator max -> unified honor_client_desktop_size: Option<DesktopSize> (#1373+#1404) - div(2) keyboard_layout (#1397) - div(3) read-side multitransport_flags (#1453) - the MCS message-channel allocation + SC_MCS_MSGCHANNEL grant + join - advertise_extended_client_data (upstream now always advertises EXTENDED_CLIENT_DATA) KEPT (macrdp-only, re-applied as pure additions): - div(3) M3c OFFER half: multitransport_offer/offered + setter, MultitransportOffer type + lib.rs re-export, SC_MULTITRANSPORT advertise in create_gcc_blocks (re-gated on multitransport_offer.is_some()), the Server-Initiate-Multitransport emit in LicensingExchange reconciled onto upstream's message_channel_id + multitransport_flags, and the CapabilitiesWaitConfirm channel-skip - div(4) client fingerprint fields (name/version/build) QA: connection.rs vs a5d1c682 = 142 pure additions + 2 intended modifications (let mut written; multi_transport_channel populate); lib.rs = only the re-export; finalization.rs = only the channel-skip (26/0); channel_connection/credssp/util byte-identical. Standalone cargo build + clippy clean, my additions fmt-clean. Downstream TODO for the server fork + main.rs: adopt the set_honor_client_desktop_size(Option<DesktopSize>) signature (bool -> Option).
…mp [WIP] Bump the vendored server Cargo.toml to a5d1c682: version 0.10->0.13, rust 1.94, all ironrdp deps to their a5d1c682 versions (async 0.10, ainput 0.8, core 0.2, pdu 0.9, svc 0.8, cliprdr 0.7, displaycontrol 0.8, echo 0.4, dvc 0.8, tokio 0.10, acceptor 0.10, graphics 0.9, rdpsnd 0.9, rdpdr 0.7, egfx 0.3). Add the nscodec feature + ironrdp-nscodec dep (div-6 retires into upstream #1332's crate). Add a standalone [patch] (vendored siblings by path + the rest git-pinned + two-sided dvc) so the server crate builds in isolation for incremental validation — like the other five forks. Untrack the pre-existing vendor/ironrdp-server/Cargo.lock (now covered by vendor/*/Cargo.lock). STANDALONE BUILD ESTABLISHED: resolution + all deps compile; the (unchanged) server source produces a bounded 33-error surface (egfx,nscodec; no multitransport yet) — the reconciliation is an API-break fix-up, NOT a refactor: rdpeusb.rs (26, the evolved ironrdp-rdpeusb PDU API) + server.rs (7, the honor-size/autodetect/egfx external touch points). macrdp's own server.rs (auth/preemption logic) is kept. Source reconciliation is the next step, now with an incremental safety net.
…liprdr) [WIP] - honor-size: reconcile macrdp's (bool, Option<max>) onto the acceptor's unified set_honor_client_desktop_size(Option<DesktopSize>) (Some(max)=honor+clamp, None=off). - autodetect: a5d1c682 moved it out of ShareDataPdu into rdp::autodetect on the MCS message channel. macrdp doesn't use RDP network-autodetect (div-15 kernel RTT), so adapt the (dead) request arm to AutoDetectReqPdu + drop the response arm. - cliprdr: cover the new ClipboardMessage::SendInitiateFileCopy variant (same initiate_file_copy call; macrdp drives copy via its own ServerEvent::ClipboardFileCopy). server.rs down 7->1 error (egfx,nscodec build); the remaining one is the egfx_on_udp_handle multitransport-cfg artifact (resolves under the real config). rdpeusb.rs (26, the evolved ironrdp-rdpeusb PDU API) is next.
…Us [WIP]
Adopt-vs-port decision: PORT now, ADOPT (retire the fork) later. rdpeusb.rs is the
most interop-fragile file in the tree (dozens of live-debugged FreeRDP-lenient/
mstsc-strict fixes) and USB can't be live-re-verified without an entitled build +
hardware, so a mechanical port preserving every fix byte-for-byte beats swapping in
uchouT's #1417 server unverified. a5d1c682 keeps the low-level wire types public +
DvcEncode; only the reshaping changed:
- UrbdrcServerPdu -> UrbdrcServer{Control,Device}Pdu (channel split); ditto the client
enums. Device process() decodes DevicePdu first, falls back to ControlPdu for the
CAPABILITIES-interface caps response (the per-device handshake mstsc REQUIRES).
- Flat TsUrb enum -> two-level TsUrbIn/TsUrbOut { header, kind: TsUrb{In,Out}Kind };
the TS_URB header is hoisted out of each payload struct.
- UrbFunction variants -> URB_FUNCTION_* newtype consts.
- Hand-built RIMCALL_RELEASE SharedMsgHeader -> first-class InterfaceRelease PDU.
- UsbDvcPdu/UsbHeaderMsg -> one generic UsbDvc<T: Encode> (wire types are DvcEncode).
- setup_to_typed_urb returns (UrbFunction, TsUrbInKind); TsUrbOutKind lacking CtlFeatReq
now type-enforces the force_transfer_in rule.
Every interop fix (typed control URBs, alt-0 config filter, SHORT_TRANSFER_OK,
decode-tolerance, per-device caps handshake, closed()-race) preserved byte-identical.
Standalone vendored server builds clean (default + egfx,nscodec); rdpeusb
rustfmt/clippy clean. Live USB re-verification deferred to hardware.
Also gate the egfx_on_udp_handle field init with #[cfg(feature = "multitransport")]
to match its gated declaration (a default/egfx build previously failed E0560).
… (div-12) [WIP] The re-vendored acceptor advertises EXTENDED_CLIENT_DATA unconditionally (upstream a5d1c682 always sets ResponseFlags::EXTENDED_CLIENT_DATA_SUPPORTED), so the M3c multitransport offer path no longer needs the set_advertise_extended_client_data(true) call — it was retired from the re-vendored acceptor. The offer alone drives the multitransport negotiation now. Completes the vendored ironrdp-server compilation: builds clean across the full feature matrix (default, egfx, nscodec, multitransport, helper), 0 errors. gfx/handler/sound/ builder API breaks the pin bump anticipated did not materialize (already covered by the earlier server.rs commits).
…1345 egfx) [WIP] - audio.rs: adopt the #1359 RdpsndServerHandler rewrite. The crate now owns format selection (new choose_format) AND the client-list wFormatNo arithmetic, so macrdp's local choose_audio_format is DELETED — a hard-won divergence (the "wFormatNo indexes the CLIENT's list" fix) absorbed upstream. choose_format returns common.first() (the crate orders common by our server preference, AAC-ahead-of-PCM); start now takes the NegotiatedFormat and returns Result, reading use_aac from format.format(). Replaced the 3 wFormatNo unit tests with one guarding the macrdp-side AAC-ordering invariant that choose_format = common.first() relies on. - h264.rs: on_frame_ack gained a 4th param (total_frames_decoded, #1345); accepted but unused — macrdp derives its decode-backlog floor from frame_id. - conn_test.rs: mirror the RdpsndServerHandler change in the NoAudio test double, and add the new ironrdp_connector::Config.connection_type field (Lan, the canonical default used by the client config / testsuite / web). Full workspace builds clean (egfx + multitransport); cargo test 203 passed / 0 failed / 3 ignored. macrdp src/ clippy + fmt clean. nscodec confirmed still unconditional (in-tree encoder, no feature gate) — no regression. Live A/V re-verification deferred to the soak.
The fork's four divergences all merged upstream (div1 #1418 SupportedUsbVer newtype, div2 #1513 usb_device match, div3 #1420 UsbConfigDesc.trailing, div4 #1321 completion guards) and src/rdpeusb.rs is now ported to a5d1c682's split URBDRC PDUs, so ironrdp-rdpeusb resolves purely via a git-rev [patch.crates-io] — the vendored copy was unreferenced. Delete it (18 files) + refresh the now-stale "VENDORED" / "can be deleted" comments. Workspace builds clean; 203 tests pass. First whole vendored fork fully retired by the bump.
…ean the vendored server [WIP] Two finalization steps on the vendored ironrdp-server fork: FMT — the verbatim-from-a5d1c682 files carry IronRDP's style (max_width=120); macrdp's DIVERGENCE files had drifted narrower (and my rdpeusb.rs port was default-100-fmt'd), so a plain `cargo fmt` would have churned the whole crate off upstream style. Add vendor/ironrdp-server/rustfmt.toml (max_width=120 + reorder_imports — the stable-safe subset of IronRDP's repo config; imports_granularity/group_imports are nightly-only) and format the divergence files to match, using STABLE rustfmt so the result is reproducible with the default toolchain. No verbatim upstream file is touched. CLIPPY — the crate had 25 pre-existing warnings, ALL in macrdp's divergence files (none in the rdpeusb.rs port). `cargo clippy --fix` applied 18 machine-safe simplifications (collapsible_if x17 + a redundant async-closure call); the remaining 7 are intentional patterns given targeted #[allow] with reasons (large_enum_variant on two transient select-race enums, too_many_arguments on RdpServer::new, new_ret_no_self on a DVC-message constructor) plus one real manual_clamp fix and a doc_lazy_continuation blank-line fix. macrdp is a single package, so `cargo fmt --all` / clippy from the repo root do NOT reach this vendored crate — check it via --manifest-path. Crate is now fmt-clean (stable) + clippy-clean; full workspace builds; 203 tests pass.
…0.9.5 main advanced to v0.9.4 (the pre-auth framing DoS hotfix) while this branch was in flight. Reconcile: - keep the branch's a5d1c682 pins (Cargo.toml/Cargo.lock) - DROP vendor/ironrdp-async: v0.9.4 vendored it to guard read_by_hint, but a5d1c682 already carries upstream #1515 (find_size hardening), so the pin now fixes the DoS upstream — the vendor dir is redundant - version 0.9.3 -> 0.9.5 (past the published 0.9.4) - docs (CLAUDE/README/release-history) merged cleanly from v0.9.4 macrdp fully green after the merge: build --all-targets, 203 tests, clippy, fmt (+ vendored server build/fmt). A pre-existing vendored-server clippy let_unit_value drift (newer clippy 1.97, unrelated to this merge) is fixed next.
…erver) clippy 1.97's let_unit_value now flags `let gfx_handle = attach_channels_impl(..)` and the `let _ = gfx_handle;` consumers when the egfx feature is off (the fn returns () there). Pre-existing under the newer toolchain, NOT from the v0.9.4 merge (the lines are unchanged since 2918040). Targeted #[allow]+reason at the four sites (server.rs 963/1385/1408/1449), matching the existing divergence clippy pattern — no logic change (security-critical divergence-23 preemption code). Vendored server clippy + fmt clean; macrdp unaffected (builds with egfx, so the allows are inert there).
…5+#1341 a5d1c682 has the real QOI fix (server-side always-Rgb #1335 + client-side Rgba decode #1341), so macrdp's --qoi-force-rgb workaround is redundant. Converge qoi_encode to upstream's exact always-*x form; delete the QOI_FORCE_RGB static + set_qoi_force_rgb setter + lib.rs re-export (vendored server), and the --qoi-force-rgb CLI flag + wiring + configuration.md entry (macrdp). Always-Rgb is now the behavior, matching upstream. Divergence log (7) marked harvested. QA green: build --all-targets, 203 tests, clippy -D warnings, fmt — macrdp AND the vendored server (default features incl. qoi).
5 tasks
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
1 participant
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.
Bumps the IronRDP dependency pin
879ffed8→a5d1c682(133 commits) and shrinks vendored divergence. No user-facing behavior change — the default runtime path is functionally unchanged.What lands
[patch]wiring.ironrdp-rdpeusbfork retired — its lenient USB-3 caps decode landed upstream; the USB URBDRC server (src/rdpeusb.rs) was ported to a5d1c682's split control/device PDUs (not blindly adopted — the mstsc-strict interop fixes were preserved).ironrdp-asyncfork dropped — v0.9.4 vendored it to guardread_by_hintagainst the pre-auth framing DoS, but a5d1c682 already carries #1515'sfind_sizehardening, so the pin fixes the DoS upstream.src/: honor-client-desktop-size, rdpsnd format selection (#1359 — the crate now owns it, so macrdp'schoose_audio_formatdivergence is gone), egfx frame-ack (#1345).--qoi-force-rgbworkaround — a5d1c682 has #1335+#1341, soqoi_encodeconverges to upstream's always-Rgb form). docs(h264): reconnect-blank is two stacked mstsc bugs (full xrdp mirror tried) #9 (honor-size) assessed + deferred (documented in the vendored-server divergence log).Verification
build --all-targets+ 203 tests + clippy-D warnings+ fmt — macrdp AND the vendored server.find_sizeDoS — clean-rejects on the bump).Net: −2 vendored forks, −1 divergence, current with upstream.