Repository navigation
rewrite(phase1): port request/sequencer/scheduler from c-RCP's RC5-conformant reference - #132
Merged
SoundMatt merged 1 commit intoAug 21, 2026
Conversation
…nformant reference Part of the v3.0.0 ground-up rewrite (cpp-RCP #129, ROADMAP.md Phase 17), branch rewrite/v3-from-c-rcp. Third and final Phase 1 module: the conditional-request layer (compound, compound-wait, triggered, chained, timed, cancellation) plus the shared sequencer/scheduler state c-RCP keeps as 3 separate modules -- kept in cpp-RCP's existing single-module unification (request.hpp already did this better than c-RCP's own current split). Real content deltas found and fixed against c-RCP's request.c/ request_sequencer.c/scheduler.c: - Fixed cpp-RCP issue #58 (cs-bit polarity inverted in compound_wait_check_of()/should_execute_chained()) as part of this port, confirmed against c-RCP's rcp_chained_advance(): was 'cs || !predecessor_errored' (cs=true never aborted anything regardless of outcome), now '!cs || !predecessor_errored', matching RCP_CHAINED_CS_CONTINUE_ON_ERROR(0)/ABORT_ON_ERROR(1). - compound_wait_check_of()/CompoundWaitCheck had no basis in c-RCP -- traced through rcp_compound_encode_request()/_decode_request(), cs is hardcoded to 0 and never surfaced for compound/compound-wait. Removed; the real comparison-mode selector is the ACF header's evt[2:0] (acf::compound_wait_evt_valid/compound_wait_match, from the acf/avtp port), now exposed via CompoundRequest's evt_op field. - cancel_single() collapsed two distinct c-RCP outcomes (NOT_FOUND vs NOT_CANCELLABLE) into one. Added RequestErrc::request_not_cancellable, split the cases. - SequencerTable::try_advance's '+1 wrapping' model was wrong -- replaced with CompoundStep + advance_guard()/start_condition_met()/ apply_next_state()/tick()/wait_tick(), matching c-RCP's explicit next_state field and the guard/start-condition distinction (rcp_compound_advance_guard() vs rcp_compound_start_condition_met()). Also newly ported (behavior c-RCP has that cpp-RCP never had): structured wire sub-fields + encode/decode for every request kind, reserved-octet and evt/hs/cs wire validation, TriggeredRuntime, wraparound-safe timed admission mapped onto acf::WireErrorCode, and frame_timing_consistent. Fixed-capacity: RequestLedger::records_ (unconditional-push_back std::vector) -> BoundedVector<RequestRecord, 64>. c-RCP has no ledger to port a bound from (its request store is a pure-function library) -- this is this port's own decision. Chose 64, not 256 (the full transaction_num range): at 256/256 every possible id is already a collision, making the capacity check unreachable. Mutation-tested (disabled the full() guard, confirmed the new overflow test failed, restored). Not ported, by design (noted in-header, not a follow-up TODO): c-RCP's positional (has_predecessor/chain_position) chain model -- cpp-RCP's existing predecessor/successor graph in RequestLedger is strictly more capable and stays as the one chained-cascade model. Left as // TODO(phase1-followup): REQ-SEQ-013/014's per-sequencer owner + fail-closed access control (needs regmap.hpp storage, out of this pass's request.hpp-only scope); rcp_compound_peek_request_type()/ rcp_timed_encode_request_tscf() convenience wrappers (composable today from existing acf/avtp primitives). 12 TEST_CASEs (REQ-SEQ-only) -> 135 TEST_CASEs / 610 assertions, covering REQ-CMP/TRIG/CHAIN/TIMED/CANCEL/SEQ/SCHED. Verified: full tree rebuilds clean (221/221 targets, incl. the just- merged fragment/respqueue), 55/55 ctest suites pass (100%) -- independently re-verified from a fresh build, not just the porting agent's own report. No call sites outside request.hpp/test_request.cpp required changes (e2e.hpp/watchdog.hpp/authz.hpp all kept their exact prior signatures against the new API). Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Signed-off-by: Matt <47545907+SoundMatt@users.noreply.github.com>
This was referenced Aug 21, 2026
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.
Part of the v3.0.0 ground-up rewrite (#129, ROADMAP.md Phase 17). Third and final Phase 1 module: the conditional-request layer (compound, compound-wait, triggered, chained, timed, cancellation) plus the shared sequencer/scheduler state c-RCP keeps as 3 separate modules — kept in cpp-RCP's existing single-module unification (
request.hppalready did this better than c-RCP's own current split).Real content deltas found and fixed
rcp_chained_advance(): wascs || !predecessor_errored(cs=true never aborted anything regardless of outcome), now!cs || !predecessor_errored.compound_wait_check_of()/CompoundWaitCheckhad no basis in c-RCP — traced throughrcp_compound_encode_request()/_decode_request(), cs is hardcoded to 0 and never surfaced for compound/compound-wait. Removed; the real comparison-mode selector is the ACF header'sevt[2:0](from the acf/avtp port), now exposed viaCompoundRequest::evt_op.cancel_single()collapsed two distinct c-RCP outcomes (NOT_FOUNDvsNOT_CANCELLABLE) into one — split, addedRequestErrc::request_not_cancellable.SequencerTable::try_advance's '+1 wrapping' model was wrong — replaced with an explicitnext_state+ separate guard/start-condition predicates, matching c-RCP'srcp_compound_advance_guard()/rcp_compound_start_condition_met().Also newly ported: structured wire sub-fields + encode/decode for every request kind, reserved-octet/evt/hs/cs wire validation,
TriggeredRuntime, wraparound-safe timed admission,frame_timing_consistent.Fixed-capacity
RequestLedger::records_: unconditional-push_backstd::vector→BoundedVector<RequestRecord, 64>. c-RCP has no ledger to port a bound from (pure-function library) — this port's own decision. Chose 64, not 256 (the fulltransaction_numrange): at 256/256 every id is already a collision, making the capacity check unreachable. Mutation-tested.Not ported, by design
c-RCP's positional chain model — cpp-RCP's existing predecessor/successor graph is strictly more capable.
Left as TODO(phase1-followup)
REQ-SEQ-013/014's per-sequencer owner + access control (needs
regmap.hppstorage, out of scope); two convenience-wrapper functions (composable today from existing primitives).12 → 135 test cases, 610 assertions.
Verified
Full tree rebuilds clean (221/221 targets, incl. the merged fragment/respqueue), 55/55 ctest suites pass (100%) — independently re-verified from a fresh build. No call sites outside
request.hpp/test_request.cpprequired changes.Tracked at #129 — completes Phase 1.