feat: search UX — auto-retry, per-file truncation, query WAL, skip dirs#194
feat: search UX — auto-retry, per-file truncation, query WAL, skip dirs#194
Conversation
Auto-retry broadening: codedb_find strips delimiters (_-.) and retries when initial query returns 0 results. Per-file match truncation: search output limits to 5 matches per file with truncation indicator. Header shows "(N shown)" count. Query tracking WAL: search/find/word queries logged to queries.log as NDJSON with JSON-escaped query strings. Foundation for combo-boost. Skip dirs: added .swc, .terraform, .serverless, elm-stuff, .stack-work, .cabal-sandbox, .cargo, bower_components. Codex review fixes: removed unused var, JSON escaping for special chars, accurate shown-count header. Generated with Claude Code Co-Authored-By: Claude <claude@anthropic.com>
Benchmark Regression ReportThreshold: 10.00%
|
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: d6ec0503e9
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| var file_counts = std.StringHashMap(u8).init(alloc); | ||
| defer file_counts.deinit(); | ||
| const max_per_file: u8 = 5; |
There was a problem hiding this comment.
In handleSearch, the per-file match count is stored as u8, then incremented for every hit. Queries that return more than 255 matches from one file will overflow at gop.value_ptr.* += 1; in safety-checked builds this can panic the MCP request, and in non-safe builds it can wrap and break truncation behavior. This makes high-hit searches unstable and should use a wider integer type (usize/u16) for counting.
Useful? React with 👍 / 👎.
| if (c == '"') { escaped[elen] = '\''; elen += 1; } | ||
| else if (c == '\\') { if (elen + 1 < escaped.len) { escaped[elen] = '\\'; escaped[elen + 1] = '\\'; elen += 2; } } | ||
| else if (c == '\n' or c == '\r' or c == '\t') { escaped[elen] = ' '; elen += 1; } | ||
| else { escaped[elen] = c; elen += 1; } |
There was a problem hiding this comment.
The query WAL writer only sanitizes ", \\, \n, \r, and \t, but leaves other ASCII control bytes untouched. A client can send a query containing escaped control chars (for example \u0001), which gets written raw into queries.log, producing invalid JSON/NDJSON lines and breaking downstream parsing for ranking features. Use a full JSON string escaper for all 0x00..0x1F characters.
Useful? React with 👍 / 👎.
1 participant
Summary
Auto-retry with query broadening
When
codedb_findreturns 0 results, automatically strips delimiters (_,-,.) and retries.auth_middleware→authmiddlewarestill finds the file.Per-file match truncation
Search output limits to 5 matches per file with
... (more matches truncated)indicator. Footer shows(N shown, M truncated)when results are cut. Header stays as first line (MCP parser compatible).Query tracking WAL
Search/find/word queries logged to
~/.codedb/projects/<hash>/queries.logas NDJSON with JSON-escaped query strings. Foundation for future combo-boost ranking.Additional skip dirs
Added
.swc,.terraform,.serverless,elm-stuff,.stack-work,.cabal-sandbox,.cargo,bower_components.Memory check
100-query burst test: 73.6MB → 74.4MB (+0.8MB). No leaks.
zig build testzero leaks.Test plan