Four header comments cite tryFireImpl and SubscriberEntry, neither of which exists

[Yaraslaut]

Summary

tryFireImpl and SubscriberEntry are cited in header comments — four sites between them — but exist nowhere in the tree. Same defect as #349's dispatchDecoded, in different files, so #349's fix does not touch them.

Found by generalising the sweep #349 asked for from remote.hpp to every header.

The sites

tryFireImpl — no tryFire-anything exists in any .hpp/.cpp:

include/morph/core/bridge.hpp:2307:   // just as the reactive `set<>` path does via `tryFireImpl`. Without
include/morph/core/registry.hpp:126:  /// (`BridgeHandler::tryFireImpl`) or the type-erased `executeJson` gate
include/morph/core/registry.hpp:413:  // (BridgeHandler::set<>'s tryFireImpl and
tests/test_coverage_gaps.cpp:575:      // ── bridge.hpp: tryFireImpl returns when draft is absent (lines 498-499)

registry.hpp:126 is a Doxygen comment, so as with #349 the dangling name is published API documentation. The tests/ site additionally cites bridge.hpp lines 498-499, which is worth checking independently — the comment names both a function that does not exist and a line range.

SubscriberEntry — no such type; the real state is Bridge::_subscriptions:

include/morph/forms/flows.hpp:189: /// "Subscription semantics" — exactly one `SubscriberEntry` per action type),

Verification status

Verified by whole-tree git grep -w on master 82c0d7bc. git grep -n "tryFire" over *.hpp/*.cpp returns only the comment sites above and nothing else; git grep -n "Subscriber" returns hasSubscribers(), _subscriptions, and the one dangling citation.

Not verified: what each name was renamed to. tryFireImpl is described as the reactive set<> path's fire step, so the successor is probably reachable from BridgeHandler::set<>, but I did not confirm which function now does that work — whoever fixes this should, rather than deleting the reference.

The sweep, and why a naive lint would be noisy

The script classified every backticked identifier in a header comment (2292 citations across 60 headers) as resolving or not, by checking whether it appears on any non-comment line tree-wide. 21 did not resolve. Most are legitimate and a lint must not flag them:

• external symbols — INT_MAX, pthread_create, __LINE__, __PRETTY_FUNCTION__, __FUNCSIG__, __udivti3, SQLITE_TRANSIENT, EMSDK_VERSION, dump_int_error (a Glaze internal, and the comment says so);
• names of things that deliberately do not exist — checkedRound is cited in rational.md precisely as the counterpart morph chose not to provide;
• wire-protocol kind strings rather than C++ identifiers — registerShared, setTimeout;
• non-C++ artifacts — SlotRegistry is a .qml file; fuzz_dispatch_execute, invalid_partial_key, unknown_key, wasm_singlethread are fuzz/CI artifacts;
• static helpers in .cpp files the script's header-only scan did not index — skip_ws.

So the true positives are 2 of 21. A lint on this rule as written would be ~90% false positives, and a lint that noisy gets disabled — which is the failure mode AGENTS.md's "verify rather than assert" section warns about, one level up. Anyone building this check needs an allowlist, or a narrower rule (for instance: only flag Foo::bar-qualified citations, or only names matching a declaration-shaped pattern in the same header).

That is the reason this is filed rather than shipped alongside #349's two-line repoint.

What would close this

Repoint the four sites at whatever now does that work — or delete the references if the concept is gone — after establishing what replaced tryFireImpl. Independently: decide whether a narrowed, allowlisted version of the sweep is worth wiring into scripts/, given that the same class produced #328's dangling docs/findings/ citations, #339's citation lint, and now #349.

[Yaraslaut]

Triage: rescope — the sites are right, but "repoint them at whatever now does that work" presumes a rename, and there wasn't one

The five-minute version: nothing replaced tryFireImpl. The mechanism it named was deleted, and these comments describe a removed mechanism as though it were current — a different and larger defect than a stale identifier.

The archaeology the issue asked for

$ git log --oneline -S tryFireImpl -- include/
7a0d68ac Stateful models: keyed shareable instances, instance subscriptions ... (#20)
d2cb3ae1 Production-hardening + GUI enhancement program (28 items) + 2 security fixes (#19)
92fbc250 Spec-audit remediation ... (#12)
00d480d5 [morph] Init

Last real definition, at d2cb3ae1 — bridge.hpp:776, static void tryFireImpl(const std::shared_ptr<SubscriberState>&, std::string_view): it snapshotted the per-action-type draft out of SubscriberState::entries, ran recomputeAll, checked readiness, and fired, with running/pending coalescing. 7a0d68ac removed it and the whole handler-side draft machinery with it. Today's bridge.hpp has no set<>, no draft, no entries, no SubscriberEntry — git grep -n "set<" include/morph/core/bridge.hpp returns exactly one hit, the dangling comment at :2317.

So the fix is not repointing. There is nothing to point at.

What that changes at each site

• registry.hpp:118-132 (Doxygen, published API): the broken part is the sentence, not the parenthesis. It says these are "the two execution sites an in-progress action reaches without first passing through the reactive set<> gate (BridgeHandler::tryFireImpl) or the type-erased executeJson gate". There is no reactive gate any more, so the arithmetic of "two sites, three gates" no longer describes anything. Same for registry.hpp:413.

• flows.hpp:189 is wrong about the section it cites, independently of the dead name. It attributes to bridge.md's "Subscription semantics" the claim "exactly one SubscriberEntry per action type". That section (docs/spec/core/bridge.md:663-670) says the opposite axis: "subscribe<R>(cb) is keyed on the result/state type, not on an action."

• A fifth site the sweep did not find, and the most misleading of them. FlowSession::set<>'s own @brief (flows.hpp:240-241) says it "forwards it to the handler's ordinary set<>". It does not — it writes std::get<A>(_drafts) and fires the step itself, and a comment eight lines below already says so: "The readiness gate that used to live in the handler's draft machinery lives here now" (:262). The AllDistinct<Steps...> static_assert (:198) is likewise required by that std::tuple<Steps...>, not by "one slot of the handler's per-action-type draft" — so a live constraint is currently justified by a mechanism that no longer exists.

Corrected scope

Rewrite these five to the mechanism that exists — flow-owned drafts in FlowSession, subscribe<R> keyed on result type — rather than repointing names. Also re-check tests/test_coverage_gaps.cpp:575's cited bridge.hpp lines 498-499 while there: that range is now mid-doc-comment on attachHandlerAsync, so the line numbers are stale too.

On the lint half

Unaffected, and its 2-of-21 measurement stands — the noise analysis is the right reason not to have shipped it alongside #349. One addition: a name-resolution lint would not have caught the fifth site at all. set<>, draft and handler all resolve tree-wide; the comment is wrong about behaviour, not about a symbol's existence. So the class here is "comments describing removed mechanisms", which no identifier-existence check can see, and that is worth knowing before anyone invests in the allowlisted narrow rule.

Labels: triage: rescope (keeping documentation, area: core).