[Tracking] jsfeat parity, refactor and modernization roadmap

[kalwalt]

Roadmap

Tracking issue for the jsfeatNext <-> jsfeat parity, refactoring and modernization effort.
Full plan: docs/jsfeat-parity-and-refactor-audit.md

Work items

• Phase 0 gate - test harness: Test harness: Vitest characterization tests vs jsfeat (Phase 0 gate) #39
• Feature parity audit (Axis 1): Feature parity audit: function & parameter level vs jsfeat (Axis 1) #45
• API/behavioral parity (Axis 2): API/behavioral parity: static-vs-instance, double namespace, per-instance cache (Axis 2) #41
• De-duplicate stubs/monolith (Axis 3/4): Refactor: de-duplicate type-only stubs and the jsfeatNext.ts monolith (Axis 3/4) #47
• Target architecture (Axis 4): Refactor: target architecture (thin aggregator, one class per module, shared cache, static facade) #40
• Build modernization -> Vite (Axis 5): Build modernization: migrate webpack to Vite + dependency cleanup (Axis 5) #42
• Port haar (Appendix A) — deferred to the Someday milestone: Port haar module (Haar cascade object/face detection) #43
• Port bbf (Appendix A) — deferred to the Someday milestone: Port bbf module (BBF face detection) #44

Status (2026-08-18): five of eight done, two deferred, one live.

The checklist above had fallen out of date — #39, #41, #47 and #42 had been
closed for weeks without being ticked, and #40 closed on 2026-08-17 — so this
issue was showing the effort as barely started when it is in fact nearly
finished.

#43 and #44 (haar, bbf) have been deliberately deferred to the Someday
milestone. They are face/object detection cascades: they serve no part of the
planar natural-feature WebAR pipeline that #96 and #97 define as the project's
direction, they block nothing, and the README already declares them
unimplemented, so no promise is broken by not having them. They are deferred
rather than closed because the evaluation is worth preserving — should anyone
ever want them, note that haar needs the tilted integral-image branch, which
carries the same first-column defect as #131.

#45 is the only live item. docs/jsfeat-parity-and-refactor-audit.md §1.1
has the module-level map, but §1.2 only describes the method for the
function-and-parameter enumeration; that enumeration has never been carried out,
and tests/api-shape.test.ts asserts the 0.9.0 calling convention rather than
per-symbol coverage. Nothing today would notice a missing jsfeat method.

This issue closes when #45 lands, with haar/bbf continuing in Someday —
otherwise it stays open indefinitely, held hostage by two items that have been
decided against for now.

Work that has moved on since this roadmap was written is tracked separately: the
1.0.0 milestone (correctness fixes and the CvBackend contract) and the 1.1.0
milestone (features2d expansion — see docs/features2d-expansion-plan.md).

[kalwalt]

Updating the checklist — 3 of 8 items shipped in 0.8.0:

• Phase 0 gate - test harness: Test harness: Vitest characterization tests vs jsfeat (Phase 0 gate) #39
• Feature parity audit (Axis 1): Feature parity audit: function & parameter level vs jsfeat (Axis 1) #45 (substantial progress, not exhaustive — see comment there)
• API/behavioral parity (Axis 2): API/behavioral parity: static-vs-instance, double namespace, per-instance cache (Axis 2) #41
• De-duplicate stubs/monolith (Axis 3/4): Refactor: de-duplicate type-only stubs and the jsfeatNext.ts monolith (Axis 3/4) #47
• Target architecture (Axis 4): Refactor: target architecture (thin aggregator, one class per module, shared cache, static facade) #40 (partial — see comment there)
• Build modernization -> Vite (Axis 5): Build modernization: migrate webpack to Vite + dependency cleanup (Axis 5) #42
• Port haar (Appendix A): Port haar module (Haar cascade object/face detection) #43
• Port bbf (Appendix A): Port bbf module (BBF face detection) #44