You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
This consolidates reported symptoms; their causes and applicability to Zoo-Code have not yet been confirmed.
Problem
Indexing may fail to start, stall, complete only partially, or report success while search remains broken. Users need a predictable process: a correct index, actionable errors, and recovery without restarting the editor or reindexing the entire workspace.
Context
Roo-Code reports cover Windows/macOS/Linux, local and remote workspaces, large projects, multiple languages, Ollama/OpenAI-compatible/Gemini providers, and local/cloud Qdrant deployments.
Reported symptom groups — with references for validation:
Startup and settings: start/save buttons do not work; settings reset or export incorrectly; users report problems saving keys and selecting custom models — #4919, #5091, #5035, #12366.
Progress and recovery: stalls at zero or partway through, progress percentages decrease, false success states, reindexing fails after clearing the index, and no convenient resume after failure — #12159, #5819, #11214, #4582, #4875, #5803.
Compatibility: available Ollama models are rejected; Qdrant URL/port issues, HTTP 400/405 errors, batch/token limits, and HTTP 429 quota errors; vector dimensions mismatch the collection, including after switching models — #10863, #11428, #4092, #5088, #5123, #5713, #10991, #5616.
Index contents and quality: excluded files enter the index, exclusion rules are lost after clearing it; documentation/.NET code is skipped; Python code is chunked incorrectly, Go snippets are duplicated; search fails even after successful indexing — #11797, #5656, #4660, #5238, #5247, #5367, #5089.
Isolation and coverage: per-project settings are requested; users report keys being shared across profiles. Related limitations include multi-root workspaces, symlinks, file search in large projects, and searching JAR dependencies — #11947, #5442, #8041, #4654, #5721, #5183.
Desired behavior
Indexing runs in the background, respects exclusions and project settings, reports outcomes accurately, explains failures, and allows users to stop/resume processing. Search returns up-to-date, permitted, meaningful snippets.
Constraints / preferences
Keep the editor responsive; bound memory and API usage; do not expose secrets in diagnostics; avoid unnecessary full reindexing. Agree separately on the scope of related coverage and isolation requests.
Request checklist
Complete the duplicate search: issues have been partially reviewed; Discussions have not been checked.
Confirm symptoms in Zoo-Code and agree on the enhancement scope.
Zoo Code Task Links
No public task link available.
Acceptance criteria
Settings persist and are restored; the selected model and dimensions are used consistently.
Progress distinguishes active processing, partial success, completion, and failure; the failure reason and next step are visible.
Stopping, restarting, clearing, and resuming after failure work without restarting VS Code.
Exclusions are respected during initial indexing, updates, clearing/reindexing, and search; deleted/modified files are accurately reflected in the index.
Provider limits are handled without endless retries; large projects do not freeze the UI or crash the extension host.
Index completeness, snippet quality, and search functionality are verified for an agreed set of languages.
Proposed approach
First reproduce the reports and assess them against the current Zoo-Code version; then split confirmed problems into subtasks and add regression checks. Use this request as an umbrella tracker.
Trade-offs / risks
The scope is too broad for a single PR; some reports are closed or concern external service configuration. Index format fixes may require rebuilding the index; reducing concurrency improves stability but slows processing. None of the referenced reports has yet been confirmed as a current Zoo-Code bug.
This consolidates reported symptoms; their causes and applicability to Zoo-Code have not yet been confirmed.
Problem
Indexing may fail to start, stall, complete only partially, or report success while search remains broken. Users need a predictable process: a correct index, actionable errors, and recovery without restarting the editor or reindexing the entire workspace.
Context
Roo-Code reports cover Windows/macOS/Linux, local and remote workspaces, large projects, multiple languages, Ollama/OpenAI-compatible/Gemini providers, and local/cloud Qdrant deployments.
Reported symptom groups — with references for validation:
Desired behavior
Indexing runs in the background, respects exclusions and project settings, reports outcomes accurately, explains failures, and allows users to stop/resume processing. Search returns up-to-date, permitted, meaningful snippets.
Constraints / preferences
Keep the editor responsive; bound memory and API usage; do not expose secrets in diagnostics; avoid unnecessary full reindexing. Agree separately on the scope of related coverage and isolation requests.
Request checklist
Zoo Code Task Links
No public task link available.
Acceptance criteria
Proposed approach
First reproduce the reports and assess them against the current Zoo-Code version; then split confirmed problems into subtasks and add regression checks. Use this request as an umbrella tracker.
Trade-offs / risks
The scope is too broad for a single PR; some reports are closed or concern external service configuration. Index format fixes may require rebuilding the index; reducing concurrency improves stability but slows processing. None of the referenced reports has yet been confirmed as a current Zoo-Code bug.