fix: bound diagnostics and stabilize large-history refreshes by Carl723000 · Pull Request #124 · ClaudeCodeUsage/ClaudeCodeUsage

Carl723000 · 2026-10-02T22:46:44Z

Summary

Bound unknown-model warnings, diagnostics, pricing downloads, aggregation labels and background result queues; add explicit prices for Opus 5.5, Sonnet 5.5 and current GPT-6 models.
Reuse unchanged Claude/Codex dashboard data and accepted sharing previews. Today attribution has a single calendar-day cache: minute polls within an hour avoid repeated history scans, while hourly Content attribution still recomputes once.
Reject retired asynchronous results after source changes, price refresh, clear or disposal. Temporary failures retain verified data only for the same source.
Improve preview/export validation, paused refresh, work progress, directory recovery and ordinary defaults reset. Upgrades preserve legacy BYOK protocol semantics; ordinary defaults reset does not delete a key.
Preserve provider-specific token, cache, price and quota meanings. Collection remains read-only, with no new permissions or automatic AI traffic.

Related to #122 and #99. Builds on the merged #119, #120 and #121 contributions; thanks @rsyuzyov.

Validation

Node: 1,255 passed, 1 filesystem-fixture skip, 0 failures.
Full Chromium Webview suite: 194 passed.
Remote CI: Node Tests, UI Test and Release Package all pass on the exact PR head.
120,000 synthetic records with default-on content analysis: minute polls read no log bodies, repeat no attribution, rebuild no hidden history and replace no document; accepted previews remain usable. The hour-crossing regression verifies one Content week computation, then reuse on the following minute.
The frozen candidate passes package safety checks. All 89 production JS files match the tested output and the runtime independently reviewed in Claude Desktop, byte for byte.
Failure-driven regressions cover source/pricing races, legacy-format migration and cache release.

Acceptance and limits

Native macOS installed-VSIX interaction and actual F5 Extension Development Host acceptance are still pending, so this PR starts as a draft.
Synthetic scale checks, Chromium and module tests do not establish that every instance of the reporter's native Windows OOM is eliminated.
Merge only after remote CI and native acceptance pass. This PR does not publish a release; the maintainer reviews the release draft and clicks Publish.

🤖 Generated with OpenAI Codex

`getModelPricing` is called for every usage record, and a model missing from the table printed its fallback warning on every call. On a live history with a model newer than the table (`claude-opus-5-5`) that was ~91k console lines for one full index build and ~27k for a refresh that re-read a few large files. In the extension host the console write is the expensive part: in a CPU profile of refreshes it cost more than the pricing lookup itself. The fallback is deterministic per model name, so the first line says everything the repeats did. A runtime override that later adds the model is checked before the fallback is reached, so nothing needs resetting. Test: `an unknown model is reported once, not once per priced record` — fails on the parent commit (6 warnings instead of 2). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> (cherry picked from commit 0ff1d30)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> (cherry picked from commit 4e8ade5)

…99) Measured on a 979-file history with several agents appending: the append fast path never ran. Every refresh reported 53 analysis payload rebases, and any payload rebase disables `fastAppendAnalysis`, so each tick of a few kilobytes paid the slow path — rebuilding the seen-UUID set over the whole history. The 53 files are the ones whose every event is older than the window. On each refresh the cutoff moves, `rebaseFileAnalysisCutoff` takes the "last event is before the cutoff" branch and returns a fresh `newAnalysisAcc`. The payload is empty before and after, but `sameAnalysisPayload` compares collections by identity (by design: a metadata-only rebase keeps them), so every such file reads as changed on every refresh, for good. That branch is now taken only when the last event has just left the window (`last >= prior.analysis.cutoffMs`). A file that was already past the previous cutoff is empty already — `analyzeLine` ignores lines older than the cutoff — and falls through to the metadata-only rebase, keeping its collections. On the same history: `payloadRebases` 53 -> 0 on every refresh. Test: `a file already outside the window keeps its empty contribution as the window drifts` — it fails on the parent commit ("Values have same structure but are not reference-equal") and asserts the analysis equals the full loader's. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> (cherry picked from commit 2eccff1)