Skip to content
mdfu

Roadmap

Active plan. Completed Tracks A–H / M1–M4 live in History.

Constraints (locked)

Section titled “Constraints (locked)”
  • Interactive TUI (default) + positional-query non-interactive filter (fzf-compatible).
  • Single-box syntax: bare words fuzzy-match; key:value hard-filter; AND-combined.
  • Unified normalized model, not strict per-format keys.
  • No persistent index, no daemon, no cgo by default.
  • Budgets: 5–10k files, cold scan+parse <500ms, per-keystroke query+rank <50ms.

Now (P0 — indexless perf, must hold budgets)

Section titled “Now (P0 — indexless perf, must hold budgets)”
  • T1: Benchmark + profile current path. 5k synthetic-file corpus; measure cold scan+parse and per-keystroke query+rank; CPU profile search.scoreDoc (fuzzy.Find per doc per word vs whole SearchBlob). Acceptance: numbers + p50/p95 in PR; bottleneck confirmed or refuted.
  • T2: Indexless rank optimization. Cache lowered SearchBlob/tokens at parse time; avoid per-doc filter-map rebuilds; early-exit bare-word AND; reuse query parse per keystroke. Acceptance: T1 corpus meets <50ms/keystroke with no behavior change (go test ./... green).
  • T3: Facet pre-index (roaring bitmaps). Pre-build value → bitmap for tag/type/status + hot custom keys; MatchesDoc becomes bitmap And/AndNot before fuzzy scoring. Acceptance: filter-only queries scale with result size, not corpus size; generic key:value still falls back to lookupField.

Next (P1 — only pay if T1/T2 miss)

Section titled “Next (P1 — only pay if T1/T2 miss)”
  • T4: Fuzzy-core bake-off. Compare sahilm/fuzzy (current) vs lithammer/fuzzysearch vs sajari/fuzzy (SymDelete) on recall + latency over real titles/bodies. Acceptance: matrix + keep/swap decision; no swap without TUI blind-test win.
  • T5: Persistent-index spike (gated). Prototype behind a flag: (a) bleve NewMemOnly dynamic mapping, (b) bluge, (c) SQLite FTS5 (modernc vs cgo). Compare binary size, cold-start, query latency, fuzzy+facet parity per Backend Evaluation. Acceptance: decision record + throwaway branches only; no new default dep.
  • T6: Scan hardening. Implement RespectGitignore (currently ignored, internal/scan/scan.go:31) + parallel file load. Acceptance: gitignored files excluded unless --no-ignore; 5k cold scan still <500ms.

Later (v2 — language + display)

Section titled “Later (v2 — language + display)”
  • T7: Query v2. | alternation within a token group, negation beyond tags, per-field boosting. Acceptance: Query Syntax updated + table tests.
  • T8: Facet counts + highlighting. Expose facet counts to TUI/--format json and match-fragment snippets. Acceptance: works indexless; aligns with T5 API if an index lands.

Delivered

Section titled “Delivered”
  • T9: mdtask-compatible task CLI + TUI. mdfu task (list, view/show, open, move, set, ids, archive, validate, install-skills) plus mdfu tasks for an interactive TUI, matching the external mdtask format (checkbox headers, #tag/!priority/@key:value metadata, @blocked_by, .mdtaskrc/.mdfurc config) and command surface; see Tasks and Tasks TUI. Acceptance: engine, CLI, and TUI tests green (go test ./...); mdfu tasks falls back to mdfu task list when stdout is not a TTY.

Working agreements

Section titled “Working agreements”
  • internal/model is the shared contract — changes need cross-package sign-off.
  • Each task: go build ./... && go test ./... green before merge.