Files
ARCADIA-Studio/harness/prerelease-integrated-source-20260924.md
T

8.6 KiB

Integrated Windows prerelease candidate

This is the source handoff for Gitea issue37, owned by Codex-956atf. Reserved version: v0.2.0-preview.integrated.20260924. Remote branch: codex/prerelease-integrated-956atf-20260924. This checkpoint does not claim combined compilation, native acceptance, packaging or publication of a release. The exact candidate commit and subsequent validation belong in issue37.

Integrated contributions

  • Base 1006b496ae5db4d850a6b7f3f5b59bf4ca99a3f8 combines the earlier e34 Scenario/Text/attachment/State work with ordinary/State multiple selection and actual Text Apply stale-action tests.
  • Merge c1b78acb9aa8c5024a3178b890102712078fc226 adds Process and Allocation group source 495464e84b5bae8ddad93e9fc69ef353e7f1c884, including its Process and pure planner ancestors. The new Allocation evaluator's Diagram literal explicitly sets the combined schema's layout: None.
  • Merge 62d3ffe9 adds Allocation drawing/Traceability source 301126f6098286bf440c1bd1b1ed685753984723 and picker Escape candidate 3e916f623c7709d6c443552cd9ec95df5918eb00. The latter includes a test-import correction but has no standalone runtime acceptance.

The single textual conflict was menu action dispatch in app.rs. Resolution retains exact group/Process dispatch first, then Traceability's captured-view placement, then State's exact placement and selection reconciliation. Independent source review found no confirmed defect in that resolution, the shared Escape wrapper, or direct Allocation drawing versus group-gesture precedence. This is source review, not a substitute for the combined tests below.

Test-only successor 542b10ac587b932c0404580a98159bd497775ef0 adds real Text Apply followed by captured Process reorder and Allocation Press/Release replay in both OA and SA. It compares complete model/source, selection and primary, history, draft, status and newer gesture ownership. Allocation protects an actual new drag; an idle obsolete Press intentionally reports a stale-view explanation. Direct rustfmt and whitespace checks pass; these new tests remain uncompiled until the combined gate. The independent source review records the inspected merge blobs and remaining discriminating checks.

Peer evidence remains attached to its actual source, not relabeled as this candidate: Allocation group 57aec979f640ba1b8aa6fcbb7d9e26b29cf51596, drawing 45605001f690ad9ee2ed786c3cd77ffbde8e10aa, and e34 Windows gates be7470afca4894c265d61800f9b94d31444e7234. Drawing evidence has a separate ancestry and is linked rather than merged for source. Private mixed-parser work, unfinished Scenario group reordering and aggregate removal are excluded.

Captured event admission successor

Combined code a6b9f6029aad618a84d136c5293353688d36809f cherry-picks the narrow peer code commits 44217307, 72b1bb7, and 6f247af795b7bec1818c588b2dabf203f8f68d8a. The peer's exact source and initial evidence are published at c655c5e. Only app.rs, group-test registration and two new test modules change in this code integration. Historical evidence archives are linked rather than merged.

Captured Connection Endpoint/Restart now require the current route context and a matching live drawing session before shared Allocation retirement. Captured Route Commit/CommitAttachment require the current context and selected edge; selection may change without advancing the context. Stale packets preserve newer Arrange/drag state and feedback. Current route edits require no invented drawing session, and unscoped commands retain their existing behavior.

The three Connection and six Route regressions are in:

  • crates/oa-desktop/src/app/allocation_drag_tests/group/connection_lifetime.rs
  • crates/oa-desktop/src/app/allocation_drag_tests/group/route_lifetime.rs

The real Text Apply cases added at 542b10ac remain in:

  • crates/oa-desktop/src/app/process_multiselect/tests.rs: completed_text_apply_rejects_captured_process_reorder_without_retiring_current_state
  • crates/oa-desktop/src/app/allocation_drag_tests/group.rs: completed_text_apply_rejects_old_allocation_packets_without_retiring_a_new_drag
  • Ordinary/State coverage is in app/multiselect_tests.rs; earlier navigation tests in navigation_union.rs are a separate scope.

Parent read the complete patch and tests, integrated without conflicts, and passed direct rustfmt and whitespace checks. A fresh independent source review found no actionable defect in the stated boundary. These do not substitute for the final combined execution. The vendor patch-description correction is incorporated at 17eaa03fd69741dff746f21d150dc04c341411b3.

Combined gates and package handoff

The first peer compilation exposed E0425 in the new Route fixture: its project_view call lacked the projection:: qualification used by the parent helpers. Combined successor 4deced53b98cd4d7649ba7aa60fa4b9b30d2f3b8 cherry-picks peer e68a83959900e556768d8470640e28a7df7d8cb8 to correct that single test line. Production code is unchanged. This concrete finding supersedes the earlier 9e99618c build handoff; the failed compile ran no tests and remains part of the peer's evidence. Source review and formatting did not catch it.

The first combined Windows attempt on 897d9f04 then found E0433 in the Allocation Text Apply test added at 542b10ac: the test's module lacked iced::widget::text_editor. The current successor incorporates peer 7c2b06b63f8a0fe14b3f001996bdf965b36a4e85, adding that import only. Production code remains unchanged. The failed combined compile ran no tests, lint or product build; its 3,709 source inputs stayed unchanged and formatting passed. Preserved failure evidence records that distinct failure. The corrected successor requires fresh combined execution; no earlier pass is substituted.

Codex-ytr83r owns the exact-candidate Windows build/package on its separate machine. Codex-9e8r6t owns independent source/seam review and a downloaded-package check. Codex-956atf remains the sole integration, tag and release publication owner. Retrieve the published candidate before starting; never build from an assumed matching local branch. Record the complete input membership, raw bytes, compiler, environment, literal command, exit, output and produced artifact hashes.

With the existing locked Windows toolchain/cache, jobs1 and incremental0:

cargo test --locked --offline --workspace --no-fail-fast -- --test-threads=1
cargo clippy --locked --offline --workspace --all-targets -- -D warnings
cargo build --locked --offline -p oa-desktop -p oa-server

Bind the actual test-name inventory and documented ignores to the final test artifacts. Preserve initial failures and corrections. Run format checks and distinguish inherited unrelated formatting from new changes. Validate the actual Text Apply stale-action tests for ordinary/State, Process and Allocation, picker overlay priority, Traceability placement and both Allocation interaction modes.

Use the established packaging recipe from the fixed candidate with the exact selected gnullvm runtime and BuildEvidence. A development-profile prerelease is acceptable when explicitly labeled. Package Studio and server, examples, offline guide and dependency notices; verify all archive entries and imports. Test a fresh extraction from an unrelated working directory, inspect real GUI editing/history/Save/reopen and picker Escape, and check server persistence through restart. Publish artifacts to the owner-created draft release only after its ID is supplied in issue37. Do not silently replace an existing published version.

Known limits for the human preview

Process crossed routes can overlap and obscure endpoints after reordering. Multiline State labels can overlap. Aggregate model removal and remaining Scenario group adapters are unfinished. Existing native evidence omits some held modifier/drag, compact-window and SA-reopen cases; compilation cannot establish those interactions. Picker Escape is a candidate until the combined native gate passes. Full OA/SA feature completeness, broad accessibility/DPI coverage and public source8 support are not claimed. Linux/WSL remains explicitly deferred by the user. See the human checks.