[codeasier]

Relationship to #269

This is a maintainer-owned integration candidate for #269, not an update to its author's fork. It preserves #269's three original commits unchanged and merges current main (3f167ec, including #271 onboarding and the descriptor-scoped fstat regression fix) through a two-parent integration commit. The original PR remains open and untouched.

Behavior carried forward

• Sharded native/msinsight frame trees retain ordered caller paths and conserve inclusive/self bytes, requested bytes and block counts; legacy text-only evidence remains explicitly incomplete.
• Latest-request-wins frame-tree cancellation covers Python and SQLite work, queued obsolete requests, analyzer close/cache invalidation and independent worker connection ownership.
• Compatible parser-v3 import reuse checks source identity, complete member inventory, required schema and indexed endpoints, not arbitrary row content, DB hashes or full frame scans. It does not seed query-validation caches or receipts. Native content validation remains intact.
• Parser-v2 compatible caches are preserved and require a new output directory. Existing compatible targets are never replaced, including with --force.

Integration fixes

1. Keep docs: streamline first-use reports and Agent Skills onboarding #271's complete first-use/onboarding and skill recovery changes while retaining feat: add sharded frame-tree analysis and lightweight cache reuse #269's bilingual frame-tree/cache/cancellation guidance.
2. Repair the stale interop assertion that expected arbitrary edited row values to fail lightweight import. Existing feat: add sharded frame-tree analysis and lightweight cache reuse #269 tests and docs already deliberately permit that reuse. The regression now proves edited bytes are not overwritten, and new owned/external member/schema/endpoint failures prove every remaining file is preserved, with and without force.
3. Independent static review found that JSON-encodable frame metadata deeper than 128 could make optional evidence abort native text import. Validate normalized JSON depth before any stack rows are inserted; omit invalid optional evidence as a whole. New native/compatible boundary cases check depth 128/129, preserved text, following-event coverage, no partial frame rows and read-only reuse/query.

Validation and remaining gates

• Independent review of the complete candidate diff found no remaining confirmed correctness blocker after the optional-frame fix.
• Python AST parsing and git diff --check passed. Project unit tests/builds are intentionally delegated to the existing CI rather than duplicated locally.
• CI for this exact head is pending. Python 3.10–3.13, lint/type/format, fixture provenance, required Chromium interactions, clean-wheel and installed-wheel dataset acceptance must be evaluated from their actual results; old feat: add sharded frame-tree analysis and lightweight cache reuse #269 failure/skip outcomes are not a pass for this candidate.
• Prior real-data and GUI evidence recorded by feat: add sharded frame-tree analysis and lightweight cache reuse #269 predates final cancellation/parser-v3 changes. This candidate does not claim fresh upstream GUI, real provider, release or deployment acceptance.
• No tags, releases, deployments, credentials or workflow gates are changed. Draft remains until exact-head CI and acceptance/review requirements are satisfied.

Snapshot Runtime Provenance

• Snapshot provenance evaluation completed.
  Snapshot provenance decision: updated

[codeasier]

Current candidate: 5fdc8375cfdfbaeb71e1746249d0d8c3853c33fe (normal fast-forward; latest main 666c9b3 and #272 badges preserved).

Independent source review found a known interoperability blocker, not just missing GUI evidence: pinned 101f65b hardcodes parser-v2, and checked current upstream master 5c722719e1d288106f8d18588b0a33d979a4065b still computes parser-v2. Current parser-v3 exports cannot be a matching no-reparse cache for these consumers. The old pinned parser's mismatch path removes manifest/log and starts the producer. Do not open the only copy of a v3 export in that GUI.

This commit corrects bilingual current-target claims and documents an isolated-copy retest plan. targetRevision is retained as a historical schema/cache-identity anchor: it does not select a consumer or prove GUI support. Runtime parser-v3 choice, historical fixtures/receipts and strict hash gates are unchanged. Product direction (existing v2-consumer reuse versus v3 awaiting upstream adaptation) remains unresolved; no consumer target is invented and this PR must not merge on a green unit suite alone.

Ordinary exact-head CI is being rerun. The installed-wheel dataset/200-seed acceptance job only runs on workflow_dispatch or schedule, so its PR-event skip is not a pass. Once the head is final, run the existing Test workflow on codex/integrate-frame-tree-269 (no inputs) and verify the resulting head_sha; retain dataset/differential artifacts. No workflow gate is changed, no real GUI run is claimed, and no tag/release/deployment is performed.

[codeasier]

Follow-up at 6cf8a152adca5f2d450a0c6522b37dd2a3b380f5

The preceding head passed the full Python 3.13 test suite (3,142 tests), coverage floors, wheel build and source audit, then exposed a stale clean-wheel comparison: dataset event now includes seven ordered-frame evidence fields absent from the standalone default event row. This head preserves exact parity for all nine base fields and independently queries the installed standalone event_frames command to check ordered filename/line/name values, duplicates, counts, event/position ownership and completeness. Dataset source coverage and full raw-array identity/namespace are checked separately; arbitrary raw extra fields are not claimed independently compared by that public standalone surface. Empty boundary stacks require additional nonempty frame evidence, not a vacuous pass. Added positive/refusal regressions and both-format orchestration coverage; no workflow/gate weakening.

Also merged current main f876665 as a second parent, preserving PR #274's complete bilingual README updates alongside this PR's cache/frame/cancellation additions. Independent static review found no blocker in these changes; source/test static checks passed, but they do not replace actual installed-wheel execution.

Fresh exact-head validation is running: Test, Platform compatibility. The parser-v3/current-msinsight incompatibility and pending product decision described above remain unchanged. Manual/scheduled dataset acceptance and applicable real-GUI acceptance remain outstanding. This PR remains Draft and must not be merged on ordinary CI alone.

[codeasier]

Final ordinary-CI receipt — 6cf8a152adca5f2d450a0c6522b37dd2a3b380f5

Both exact-head workflows are now terminal success:

• Test / 41224192339: Python 3.13 full suite 3,234 passed, 14 skipped, 7 deselected, 4 subtests; 93% total coverage and all critical floors passed. Python 3.10–3.12, lint/format/type checking, provenance and required Chromium interactions also passed.
• The same run built/audited the wheel and passed the actual isolated clean-wheel CLI/skills acceptance. Native and compatibility exports each covered 8,180 real events in two shards, with peak/base-event parity, complete ordered-frame field/identity checks and nonempty frame evidence. This is offline CLI acceptance, not a live-agent or GUI claim.
• Platform compatibility / 41224192340: all four Windows partitions and macOS passed (macOS 3,218 passed). The tested merge tree matches the candidate tree dff8133f06e5853d7ba6c8f11466e560d13c8687.

Final recheck: actual main remains f876665, included as a parent; the candidate is clean/mergeable, independently reviewed, but remains Draft / not eligible to merge.

Outstanding gates:

1. Resolve the product decision: current parser-v3 exports are incompatible with the checked parser-v2 msinsight cache consumers. The historical targetRevision is not a verified v3 consumer target.
2. On the final chosen head, manually run Test, select branch codex/integrate-frame-tree-269, leave inputs unchanged (this workflow defines none), and verify its run head_sha exactly matches the chosen head. The Installed-wheel dataset acceptance job must actually execute/pass its required installed-wheel dataset checks and 200-seed paired differential acceptance, with both evidence artifacts present. Its PR-event skip here is not acceptance.
3. After a matching fixed consumer/build is established, run the documented GUI acceptance using a separately copied trusted pickle and adjacent candidate export in a new isolated directory, never the only copy or a parser-v2 cache. Record build/wheel/source identities, cache UP_TO_DATE and no-producer-invocation evidence, cross-shard curves/blocks, repeated navigation/cancellation, and before/after inventories. The checked v2 consumers cannot establish v3 reuse; their reparse path may modify the copied cache. Historical GUI receipts do not cover this head.

Original #269 and its foreign-fork branch remain unchanged. No tag, release, deployment or merge was performed.

[codeasier]

Closing this parallel integration candidate to consolidate follow-up work in #269. This is not a rejection of the feature and does not mean #269 has absorbed every change here. The branch codex/integrate-frame-tree-269 is retained at 6cf8a152; no branch or commit is being deleted.

Comparison with #269 at 39d0793 confirms that its core optional-frame depth preflight, stale event-comparator error and interop content-edit expectation have been repaired. It also fixes the allocation-ID membership O(N²) scan still present in this candidate. The following remaining patch themes should be considered in #269 rather than lost with this PR:

1. Bilingual compatibility documentation: historical 101f65b versus current v3, targetRevision meaning, old-consumer reparse/cache-mutation warning, and isolated-copy GUI acceptance plan. See the corrected protocol and safe retest.
2. Stronger clean-wheel evidence: independent installed standalone event_frames comparison, strict base-field types and complete references, source identity/coverage, and nonempty frame evidence. See comparator and CLI proof queries.
3. Compatible-cache preservation regressions for owned/external caches, missing member/schema/endpoint damage, both force modes and unchanged focus: tests.
4. Depth-boundary regressions: depth128 accepted/depth129 omitted, a following valid event still covered, and unchanged bytes through reuse/read-only queries: tests. The alternative shared-helper layout in feat: add sharded frame-tree analysis and lightweight cache reuse #269 need not be mechanically replaced.

Items 2–4 identify remaining verification coverage, not a claim that the original fixed product defects still reproduce. #269 still needs current-main conflict resolution, review and CI for its final commit, plus the applicable installed-wheel/differential and coordinated-consumer acceptance. This PR's historical passing runs do not transfer to #269.