Repository navigation
(OPT) Fund boxed path-state insertion control receivers through complete Drop #849
Description
Source review of main 019d2b7c32e806b1f05d7a826f4c9f368b6a8705 found an additional receiving frontier in src/cleanup_plan/replay/path_join/boxed_states.rs. The boxed representation genuinely pays the pointer-key tree node and owned state payload; those backing reservations do not establish custody for every new control constructor.
The smallest concrete counterexample is duplicate owned insertion: the new helper constructs the zero-authority reservation Result and its own Result<bool, Diagnostic>, while the original outer insert_incoming -> Result<(), Diagnostic> remains. The duplicate branch performs no positive debit. Some success-path node helper Results correspond to the original tree helper and must not be counted twice. Unit-error generic component tests and heap-byte totals cannot prove funding of the actual production Diagnostic receiving path. No exact aggregate fee or application saving has been measured for this finding.
Implement the bounded caller-owned receiving boundary or eliminate genuinely introduced carriers, using the complete actual old/new constructor inventory. A private design uses one invocation-owned, explicitly paid terminal Diagnostic slot and unit-returning insertion operations, with no new capturing reserve closure or unused Result. Authenticate duplicate/zero reuse before state access; retain original sticky/floor/foreign/restored behavior and complete state ordering, replay work, branch normalization and ownership Drop. Pay all genuinely new positive controls and Box initialization/backing before acquisition. No cap increase, refund, unrelated backing credit, global bank or reused one-use terminal authority. Initial acquisition refusal must keep its honest existing unreceived reporting boundary.
Preserve the complete immediate main helpers and independent pre-main whole replay receiver. Qualify the actual Diagnostic helper and complete real-HIR caller, not only E=() callbacks: exact/one-short/zero/duplicate/held transfer/sticky/foreign-restored/floor controls, full diagnostics/order, physical allocations and complete Drop. Run only the affected existing owners plus unchanged Pattern/Catalog/starter application checks; no full suites or quality gates. Regenerate genuinely moved workspace and browser pins from their distinct actual producers, retaining every complete prior wire and source inverse.
This follows the boxed-state optimization and is distinct from #816 expression-fact receiving directories. It may help #752/#779/#808, but no application fit, universal improvement, benchmark or token-saving claim is justified yet.
Activity
wsdt commentedon Oct 16, 2026
The boxed insertion receiving implementation is pushed. At exact source
2d4b68d3a, all five selected whole-replay/refusal/duplicate/foreign controls pass, and the repaired complete SourceA and SourceB custody owners pass. Both scoped production Clippy commands also pass. The complete focused batch preserves all 39 PASS and 11 FAIL outcomes; the earlier thirteen direct8bb passes remain historical, not rerun claims.The issue remains open pending the corresponding updated pin consumers. Actual managed and browser captures are published with full old wires/inverses; pin edits are not yet applied. Further receiving optimizations are being combined before the next build so those pins will be regenerated from the final source, rather than repeatedly changed between small fixes. Current application builder-limit failures remain disclosed; no application-fit or agent-token gain is claimed.
wsdt commentedon Oct 16, 2026
Implemented and qualified the boxed-insertion receiving boundary, including duplicate zero-authority reuse, sticky/floor/foreign ownership, original ordering/work and complete Drop. The initial acquisition refusal keeps its explicit existing reporting boundary.
The remaining pin-consumer requirement is now complete. At exact post-pin source
bcd83107ccda58800e9f2b848950cece03979de6, all original15 managed/browser consumers PASS: exact graph/context/analysis wires, baseline/rename known answers, both full historical branches, hostile non-budget mutations and complete source custody. Published complete21 results and independent audit retain 19 PASS / 2 FAIL, plus the interrupted zero-body attempt separately.The pins come from actual47f producers and complete independent old-wire inverses: managed ordinary reservation1,106,063; browser baseline/renamed3,477,789. All previous wires remain preserved. No bill was guessed or counter equalized.
The five relevant whole-replay/refusal/duplicate/foreign controls actually PASS at47f(rows46–50), and both complete SourceA/B custody owners PASS(rows11–12) in the unfiltered72-owner qualification. Both scoped production Clippy commands PASS there. Post-pin production is byte-identical to that cut; those lint/receiver results remain explicitly source-bound historical evidence, not new invocations.
#847's actual Project route check and the synthetic array measurement owner still fail; application budget failures remain disclosed. They are separate work, and no whole-head, application-fit, RAM, agent-token or billed-cost advantage is claimed. Closing this bounded request now that its actual pin consumers have passed. No full suite or full quality gate ran.