Keep manual edits to a song across rebuilds

The groundwork for an in-app editor. Edits live in edits.json in the work
dir, which no stage produces, so no rebuild removes them. An edit is a
whole lyric line as the user left it, in seconds, so a rebuild with a new
BPM re-snaps it. grid lays the edits over the freshly generated notes
(neighbouring lines give way to an edited one), writes the result to
notes_final.json with warnings brought up to date, and charts that. When
the lyrics change, edits follow their lines by text and order; an edit
whose line's words changed is set aside as stale and reported, not applied.

Notes gain a kind, so golden, freestyle and rap notes can be charted. The
review endpoint returns the merged notes with their syllables, kinds and
edit state, plus the chart's beat grid. PUT/DELETE
/api/jobs/{id}/lines/{n} save and revert a line, DELETE
/api/jobs/{id}/stale/{n} discards a stale edit; each rewrites the work and
packaged charts in place with the tempo already chosen, so a save
re-encodes nothing.

src/songtrace/edits.py

@@ -0,0 +1,173 @@
+"""Manual edits to a song, kept apart from what the pipeline generates.
+
+Edits live in work/<slug>/edits.json, which no stage produces, so no rebuild
+removes them; grid lays them over the freshly generated notes every time. An
+edit is a whole lyric line as the user left it, timed in seconds so that a
+rebuild with a different BPM re-snaps it rather than breaking it.
+"""
+import difflib
+import json
+import time
+from pathlib import Path
+
+from .notes import MIN_NOTE_SEC, TimedNote, low_confidence_warnings
+
+EDITS_FILE = "edits.json"
+FINAL_FILE = "notes_final.json"
+KINDS = {":", "*", "F", "R", "G"}
+
+
+def _empty(lyrics: list[list[str]]) -> dict:
+    return {"lyrics": lyrics, "lines": {}, "stale": []}
+
+
+def load(work: Path, lyrics: list[list[str]]) -> dict:
+    """The song's edits, keyed to `lyrics`.
+
+    Edits are stored with the lyrics they were made against. When those
+    differ, each edited line is matched to its line in the new lyrics by
+    text and order; an edit whose line's words changed is set aside as stale
+    rather than applied to words it was not made for.
+    """
+    path = work / EDITS_FILE
+    if not path.exists():
+        return _empty(lyrics)
+    data = json.loads(path.read_text())
+    if data["lyrics"] != lyrics:
+        data = _remap(data, lyrics)
+        save(work, data)
+    return data
+
+
+def save(work: Path, data: dict) -> None:
+    path = work / EDITS_FILE
+    if not data["lines"] and not data["stale"]:
+        path.unlink(missing_ok=True)
+        return
+    tmp = path.with_suffix(".tmp")
+    tmp.write_text(json.dumps(data))
+    tmp.replace(path)
+
+
+def _remap(data: dict, lyrics: list[list[str]]) -> dict:
+    old = [tuple(line) for line in data["lyrics"]]
+    new = [tuple(line) for line in lyrics]
+    moved = {}
+    for tag, i1, i2, j1, _ in difflib.SequenceMatcher(
+            a=old, b=new, autojunk=False).get_opcodes():
+        if tag == "equal":
+            moved.update({i1 + k: j1 + k for k in range(i2 - i1)})
+    lines, stale = {}, list(data["stale"])
+    for key, edit in data["lines"].items():
+        to = moved.get(int(key))
+        if to is None:
+            stale.append(edit)
+        else:
+            lines[str(to)] = {**edit, "line_index": to}
+    return {"lyrics": lyrics, "lines": lines, "stale": stale}
+
+
+def validate_notes(notes: list[dict]) -> list[dict]:
+    """The notes of a line edit, checked and reduced to what is stored."""
+    clean = []
+    for n in notes:
+        try:
+            note = {"start": float(n["start"]), "duration": float(n["duration"]),
+                    "pitch": int(n["pitch"]), "text": str(n["text"]),
+                    "kind": str(n.get("kind", ":"))}
+        except (KeyError, TypeError, ValueError):
+            raise ValueError(f"malformed note {n!r}")
+        if note["start"] < 0 or note["duration"] <= 0:
+            raise ValueError(f"a note needs start >= 0 and duration > 0: {n!r}")
+        if note["kind"] not in KINDS:
+            raise ValueError(f"unknown note kind {note['kind']!r}")
+        if not note["text"].strip():
+            raise ValueError("a note needs text")
+        clean.append(note)
+    return sorted(clean, key=lambda n: n["start"])
+
+
+def set_line(data: dict, index: int, notes: list[dict]) -> None:
+    data["lines"][str(index)] = {
+        "line_index": index, "words": data["lyrics"][index],
+        "notes": validate_notes(notes), "saved_at": time.time()}
+
+
+def clear_line(data: dict, index: int) -> bool:
+    return data["lines"].pop(str(index), None) is not None
+
+
+def merge(notes: list[dict], breaks: list[float], data: dict
+          ) -> tuple[list[dict], list[float]]:
+    """Generated notes and breaks with every edited line laid over them.
+
+    An edited line replaces the generated notes of that line, and the user's
+    timing wins: generated notes of earlier lines are cut back to end where it
+    starts, and those of later lines to start where it ends. Line breaks next
+    to an edited line move to the middle of the gap it leaves.
+    """
+    edited = {int(k): e["notes"] for k, e in data["lines"].items()}
+    if not edited:
+        return notes, breaks
+    spans = {li: (ns[0]["start"], ns[-1]["start"] + ns[-1]["duration"])
+             for li, ns in edited.items() if ns}
+    kept = []
+    for n in notes:
+        li = n["line_index"]
+        if li in edited:
+            continue
+        start, end = n["start"], n["start"] + n["duration"]
+        for eli, (a, b) in spans.items():
+            if li < eli:
+                end = min(end, a)
+            elif li > eli:
+                start = max(start, b)
+        if end - start >= MIN_NOTE_SEC:
+            kept.append({**n, "start": start, "duration": end - start})
+    for li, ns in edited.items():
+        kept += [{**n, "line_index": li, "low_confidence": False} for n in ns]
+    kept.sort(key=lambda n: (n["start"], n["line_index"]))
+
+    def line_after(t: float) -> int | None:
+        return next((n["line_index"] for n in notes if n["start"] >= t), None)
+
+    first = {}
+    last = {}
+    for n in kept:
+        first.setdefault(n["line_index"], n)
+        last[n["line_index"]] = n
+    order = sorted(first, key=lambda li: first[li]["start"])
+    out_breaks = []
+    for prev, nxt in zip(order, order[1:]):
+        if prev not in edited and nxt not in edited:
+            # the generated break between these two lines, untouched
+            gen = [t for t in breaks if line_after(t) == nxt]
+            if gen:
+                out_breaks.append(gen[0])
+                continue
+        end = last[prev]["start"] + last[prev]["duration"]
+        out_breaks.append((end + first[nxt]["start"]) / 2)
+    return kept, out_breaks
+
+
+def finalize(work: Path, lyrics: list[list[str]]) -> dict:
+    """The song as it is to be charted: generated notes with the edits laid
+    over them, warnings brought up to date, written to notes_final.json."""
+    generated = json.loads((work / "notes_sec.json").read_text())
+    data = load(work, lyrics)
+    notes, breaks = merge(generated["notes"], generated["breaks"], data)
+    warnings = generated.get("warnings", [])
+    if data["lines"]:
+        # low-confidence warnings must describe the merged notes, so a line
+        # the user fixed stops being reported; the others are about lyrics
+        warnings = (low_confidence_warnings([TimedNote(**n) for n in notes])
+                    + [w for w in warnings if "low-confidence" not in w])
+    if data["stale"]:
+        n = len(data["stale"])
+        warnings.append(f"{n} edited line{' no longer matches' if n == 1 else 's no longer match'}"
+                        " the lyrics; set aside in the editor")
+    final = {"notes": notes, "breaks": breaks, "warnings": warnings,
+             "edited_lines": sorted(int(k) for k in data["lines"]),
+             "stale": data["stale"]}
+    (work / FINAL_FILE).write_text(json.dumps(final))
+    return final

src/songtrace/grid.py

@@ -3,7 +3,8 @@ from pathlib import Path
 
 import numpy as np
 
-from .ultrastar import LineBreak, Note, UltraStarSong, format_bpm
+from .ultrastar import (LineBreak, Note, UltraStarSong, format_bpm, get_header,
+                        parse_bpm, parse_ultrastar, serialize_ultrastar)
 
 
 def detect_bpm(instrumental_path: Path) -> float:
@@ -49,10 +50,24 @@ def to_ultrastar(notes: list[dict], breaks: list[float], file_bpm: float,
         else:
             b = max(beat(t), cursor)
             length = max(1, beat(t + n["duration"]) - beat(t))
-            body.append(Note(kind=":", beat=b, length=length, pitch=n["pitch"],
-                             text=n["text"]))
+            body.append(Note(kind=n.get("kind", ":"), beat=b, length=length,
+                             pitch=n["pitch"], text=n["text"]))
             cursor = b + length
     out_headers = list(headers)
     out_headers.append(("BPM", format_bpm(file_bpm)))
     out_headers.append(("GAP", str(gap_ms)))
     return UltraStarSong(headers=out_headers, body=body)
+
+
+def rechart(txt_path: Path, notes: list[dict], breaks: list[float]) -> None:
+    """Replace a chart's notes, keeping its headers and its BPM.
+
+    For an edit saved from the editor: the tempo was settled when the song
+    was built, and the packaged chart's headers may differ from the work
+    copy's (a re-encoded #VIDEO, a dropped #COVER), so each keeps its own.
+    """
+    song = parse_ultrastar(txt_path.read_text(encoding="utf-8"))
+    bpm = parse_bpm(get_header(song, "BPM"))
+    headers = [(k, v) for k, v in song.headers if k.upper() not in ("BPM", "GAP")]
+    txt_path.write_text(serialize_ultrastar(to_ultrastar(notes, breaks, bpm, headers)),
+                        encoding="utf-8")

src/songtrace/notes.py

@@ -34,6 +34,7 @@ class TimedNote:
     text: str             # display text incl. leading space / "~"
     low_confidence: bool
     line_index: int
+    kind: str = ":"       # UltraStar note type; generated notes are all normal
 
 
 def hz_to_midi_float(hz: float) -> float:
@@ -170,8 +171,7 @@ def _fmt(t: float) -> str:
     return f"{int(t // 60)}:{int(t % 60):02d}"
 
 
-def _warnings(notes: list[TimedNote], dropped_lines=(), missing_spans=(),
-             ref_lines=()) -> list[str]:
+def low_confidence_warnings(notes: list[TimedNote]) -> list[str]:
     runs: list[list[TimedNote]] = []
     for n in notes:
         if not n.low_confidence:
@@ -181,10 +181,15 @@ def _warnings(notes: list[TimedNote], dropped_lines=(), missing_spans=(),
             runs[-1].append(n)
         else:
             runs.append([n])
-    warnings = [
+    return [
         f"{len(run)} low-confidence note{'' if len(run) == 1 else 's'} "
         f"near {_fmt(run[0].start)}–{_fmt(run[-1].start + run[-1].duration)}"
         for run in runs]
+
+
+def _warnings(notes: list[TimedNote], dropped_lines=(), missing_spans=(),
+             ref_lines=()) -> list[str]:
+    warnings = low_confidence_warnings(notes)
     for group in _consecutive_runs(dropped_lines):
         n = len(group)
         text = " ".join(ref_lines[group[0]]) if group[0] < len(ref_lines) else ""

src/songtrace/pipeline.py

@@ -47,7 +47,11 @@ def song_slug(inp: SongInput) -> str:
     return name.replace("/", "").replace("\0", "")
 
 def read_warnings(work: Path) -> list[str]:
-    notes_file = work / "notes_sec.json"
+    # grid's copy accounts for the user's edits; notes' is all a song built
+    # before edits existed has
+    notes_file = work / "notes_final.json"
+    if not notes_file.exists():
+        notes_file = work / "notes_sec.json"
     if not notes_file.exists():
         return []
     return json.loads(notes_file.read_text()).get("warnings", [])

src/songtrace/stages/grid.py

@@ -1,13 +1,16 @@
 import json
 from pathlib import Path
 
-from .. import grid
+from .. import edits, grid
+from ..lyrics import normalize_lyrics
 from ..pipeline import Ctx, PipelineError, Stage, song_slug
 from ..ultrastar import serialize_ultrastar
 
 
 def _run(ctx: Ctx) -> None:
-    data = json.loads((ctx.work / "notes_sec.json").read_text())
+    data = edits.finalize(ctx.work, normalize_lyrics(ctx.inp.lyrics_text).lines)
+    if data["edited_lines"]:
+        ctx.log(f"[grid] kept {len(data['edited_lines'])} edited line(s)")
     meta = json.loads((ctx.work / "meta.json").read_text())
     if ctx.inp.bpm_override is not None:
         bpm = ctx.inp.bpm_override
@@ -36,5 +39,6 @@ def _run(ctx: Ctx) -> None:
 
 def grid_stage() -> Stage:
     return Stage(name="grid",
-                 outputs=lambda ctx: [ctx.work / "song.txt"],
+                 outputs=lambda ctx: [ctx.work / "song.txt",
+                                      ctx.work / edits.FINAL_FILE],
                  run=_run)

src/songtrace/web.py

@@ -9,11 +9,12 @@ import uuid
 from dataclasses import asdict, dataclass, field
 from pathlib import Path
 
-from fastapi import FastAPI, File, Form, HTTPException, UploadFile
+from fastapi import Body, FastAPI, File, Form, HTTPException, UploadFile
 from fastapi.concurrency import run_in_threadpool
 from fastapi.responses import FileResponse
 from fastapi.staticfiles import StaticFiles
 
+from . import edits, grid
 from . import rebuild as rebuild_policy
 from .config import Config
 from .lyrics import normalize_lyrics
@@ -132,6 +133,7 @@ def create_app(cfg: Config, runner=run_pipeline) -> FastAPI:
     jobs: dict[str, Job] = {}
     order: list[str] = []
     lock = threading.Lock()
+    edit_lock = threading.Lock()
     todo: queue.Queue[str] = queue.Queue()
 
     for job in _discover_jobs(cfg):
@@ -466,7 +468,9 @@ def create_app(cfg: Config, runner=run_pipeline) -> FastAPI:
         if job.status != "done":
             raise HTTPException(404)
         work = cfg.work_dir / song_slug(job.inp)
-        notes_file = work / "notes_sec.json"
+        notes_file = work / edits.FINAL_FILE
+        if not notes_file.exists():
+            notes_file = work / "notes_sec.json"
         if not notes_file.exists():
             raise HTTPException(404)
         notes_data = json.loads(notes_file.read_text())
@@ -486,13 +490,96 @@ def create_app(cfg: Config, runner=run_pipeline) -> FastAPI:
         for n in notes_data["notes"]:
             by_line.setdefault(n["line_index"], []).append({
                 "start": n["start"], "end": n["start"] + n["duration"],
-                "pitch": n["pitch"], "low_confidence": n["low_confidence"]})
+                "pitch": n["pitch"], "low_confidence": n["low_confidence"],
+                "text": n["text"], "kind": n.get("kind", ":")})
+        edited = set(notes_data.get("edited_lines", []))
         lines = [{"index": i, "text": " ".join(words), "dropped": i in dropped,
+                  "edited": i in edited,
                   "notes": by_line.get(i, []), "words": words_by_line.get(i, [])}
                  for i, words in enumerate(ref.lines)]
+        stale = [{"text": " ".join(e["words"]), "notes": e["notes"]}
+                 for e in notes_data.get("stale", [])]
         media = {kind: _resolve_media(job, header) is not None
                  for kind, header in MEDIA_HEADERS.items()}
-        return {"lines": lines, "missing_spans": missing_spans, "media": media}
+        return {"lines": lines, "missing_spans": missing_spans, "media": media,
+                "stale": stale, "grid": _chart_grid(work)}
+
+    def _chart_grid(work: Path) -> dict | None:
+        """The beat grid the chart was built on, for snapping edits to it."""
+        txt = work / "song.txt"
+        if not txt.exists():
+            return None
+        song = parse_ultrastar(txt.read_text(encoding="utf-8"))
+        bpm, gap = get_header(song, "BPM"), get_header(song, "GAP")
+        if bpm is None or gap is None:
+            return None
+        return {"bpm": float(bpm.replace(",", ".")), "gap_ms": float(gap.replace(",", "."))}
+
+    def _editable(job_id: str) -> Job:
+        job = _finished_job(job_id)
+        slug = song_slug(job.inp)
+        with lock:
+            # a build of this song would overwrite the chart being edited
+            if any(j.status in ("queued", "running") and song_slug(j.inp) == slug
+                   for j in jobs.values()):
+                raise HTTPException(409, "this song is being rebuilt")
+        if job.status != "done":
+            raise HTTPException(409, "only a finished song can be edited")
+        return job
+
+    def _edit(job_id: str, change) -> dict:
+        """Apply `change` to the song's edits, then rewrite its charts."""
+        job = _editable(job_id)
+        work = cfg.work_dir / song_slug(job.inp)
+        lyrics = normalize_lyrics(job.inp.lyrics_text).lines
+        with edit_lock:
+            data = edits.load(work, lyrics)
+            try:
+                change(data, lyrics)
+            except ValueError as e:
+                raise HTTPException(400, str(e))
+            edits.save(work, data)
+            final = edits.finalize(work, lyrics)
+            charts = [work / "song.txt"]
+            if job.output_dir:
+                charts.append(Path(job.output_dir) / f"{song_slug(job.inp)}.txt")
+            for txt in charts:
+                if txt.exists():
+                    grid.rechart(txt, final["notes"], final["breaks"])
+        with lock:
+            job.warnings = list(final["warnings"])
+        return get_review(job_id)
+
+    def _line(lyrics, index: int) -> int:
+        if not 0 <= index < len(lyrics):
+            raise HTTPException(404, f"no line {index}")
+        return index
+
+    @app.put("/api/jobs/{job_id}/lines/{index}")
+    def save_line(job_id: str, index: int, body: dict = Body(...)):
+        """Save a line as edited: its full note list, in seconds."""
+        notes = body.get("notes")
+        if not isinstance(notes, list):
+            raise HTTPException(400, "body must be {\"notes\": [...]}")
+        return _edit(job_id, lambda data, lyrics: edits.set_line(
+            data, _line(lyrics, index), notes))
+
+    @app.delete("/api/jobs/{job_id}/lines/{index}")
+    def revert_line(job_id: str, index: int):
+        """Drop a line's edit, so it follows what the pipeline generates."""
+        def change(data, lyrics):
+            if not edits.clear_line(data, _line(lyrics, index)):
+                raise HTTPException(404, f"line {index} has no edit")
+        return _edit(job_id, change)
+
+    @app.delete("/api/jobs/{job_id}/stale/{n}")
+    def discard_stale(job_id: str, n: int):
+        """Throw away an edit whose line's lyrics have since changed."""
+        def change(data, lyrics):
+            if not 0 <= n < len(data["stale"]):
+                raise HTTPException(404, f"no stale edit {n}")
+            data["stale"].pop(n)
+        return _edit(job_id, change)
 
     return app
 

tests/test_edits.py

@@ -0,0 +1,124 @@
+import json
+
+import pytest
+
+from songtrace import edits
+
+LYRICS = [["one", "two"], ["three", "four"], ["five", "six"]]
+
+
+def gen(start, line, pitch=0, dur=0.4, low=False, text="la"):
+    return {"start": start, "duration": dur, "pitch": pitch, "text": text,
+            "low_confidence": low, "line_index": line}
+
+
+def edited(*lines):
+    data = edits._empty(LYRICS)
+    for index, notes in lines:
+        edits.set_line(data, index, notes)
+    return data
+
+
+def note(start, pitch=5, dur=0.4, text="la", kind=":"):
+    return {"start": start, "duration": dur, "pitch": pitch, "text": text, "kind": kind}
+
+
+GENERATED = [gen(1.0, 0), gen(1.5, 0), gen(4.0, 1), gen(4.5, 1), gen(8.0, 2), gen(8.5, 2)]
+BREAKS = [3.0, 6.0]
+
+
+def test_no_edits_leaves_the_generated_song_alone():
+    assert edits.merge(GENERATED, BREAKS, edits._empty(LYRICS)) == (GENERATED, BREAKS)
+
+
+def test_an_edited_line_replaces_its_generated_notes():
+    notes, _ = edits.merge(GENERATED, BREAKS, edited(
+        (1, [note(4.2, pitch=7, kind="*"), note(4.8, pitch=9)])))
+    line1 = [n for n in notes if n["line_index"] == 1]
+    assert [(n["start"], n["pitch"], n["kind"]) for n in line1] == [
+        (4.2, 7, "*"), (4.8, 9, ":")]
+    assert not any(n["low_confidence"] for n in line1)
+    assert [n["line_index"] for n in notes] == [0, 0, 1, 1, 2, 2]
+
+
+def test_neighbours_give_way_to_an_edited_line():
+    # line 1 edited to run from 1.7s to 8.6s, over its neighbours' notes
+    notes, _ = edits.merge(GENERATED, BREAKS, edited(
+        (1, [note(1.7), note(8.2, dur=0.4)])))
+    by_line = {}
+    for n in notes:
+        by_line.setdefault(n["line_index"], []).append(
+            (round(n["start"], 2), round(n["start"] + n["duration"], 2)))
+    assert by_line[0] == [(1.0, 1.4), (1.5, 1.7)]     # cut back to end at 1.7s
+    assert by_line[2] == [(8.6, 8.9)]                 # 8.0 note gone, 8.5 starts at 8.6
+
+
+def test_breaks_beside_an_edited_line_move_and_the_rest_stay():
+    data = edited((2, [note(9.0), note(9.5)]))
+    _, breaks = edits.merge(GENERATED, BREAKS, data)
+    assert breaks[0] == 3.0                           # lines 0-1: generated
+    assert breaks[1] == pytest.approx((4.9 + 9.0) / 2)  # lines 1-2: re-centred
+
+
+def test_an_edit_follows_its_line_when_lines_are_added_above(tmp_path):
+    data = edited((1, [note(4.0)]))
+    edits.save(tmp_path, data)
+    moved = edits.load(tmp_path, [["zero"]] + LYRICS)
+    assert list(moved["lines"]) == ["2"] and moved["lines"]["2"]["line_index"] == 2
+    assert moved["stale"] == []
+    assert json.loads((tmp_path / edits.EDITS_FILE).read_text())["lyrics"][0] == ["zero"]
+
+
+def test_an_edit_whose_words_changed_is_set_aside(tmp_path):
+    edits.save(tmp_path, edited((1, [note(4.0)]), (2, [note(8.0)])))
+    changed = [LYRICS[0], ["three", "more"], LYRICS[2]]
+    data = edits.load(tmp_path, changed)
+    assert list(data["lines"]) == ["2"]
+    assert [s["words"] for s in data["stale"]] == [["three", "four"]]
+
+
+def test_validate_notes_checks_and_orders():
+    assert [n["start"] for n in edits.validate_notes([note(2.0), note(1.0)])] == [1.0, 2.0]
+    for bad in (note(1.0, kind="X"), note(1.0, dur=0), note(-1.0), note(1.0, text=" "),
+                {"start": 1.0}):
+        with pytest.raises(ValueError):
+            edits.validate_notes([bad])
+
+
+def test_saving_no_edits_removes_the_file(tmp_path):
+    edits.save(tmp_path, edited((0, [note(1.0)])))
+    data = edits.load(tmp_path, LYRICS)
+    edits.clear_line(data, 0)
+    edits.save(tmp_path, data)
+    assert not (tmp_path / edits.EDITS_FILE).exists()
+
+
+def _write_generated(work, notes, warnings):
+    (work / "notes_sec.json").write_text(json.dumps(
+        {"notes": notes, "breaks": BREAKS, "warnings": warnings}))
+
+
+def test_finalize_without_edits_passes_warnings_through(tmp_path):
+    _write_generated(tmp_path, GENERATED, ["anything at all"])
+    final = edits.finalize(tmp_path, LYRICS)
+    assert final["warnings"] == ["anything at all"]
+    assert final["edited_lines"] == [] and final["stale"] == []
+    assert json.loads((tmp_path / edits.FINAL_FILE).read_text()) == final
+
+
+def test_finalize_stops_reporting_a_line_the_user_fixed(tmp_path):
+    notes = [gen(1.0, 0), gen(4.0, 1, low=True), gen(8.0, 2)]
+    _write_generated(tmp_path, notes, ["1 low-confidence note near 0:04–0:04",
+                                       "missing lyrics near 0:20–0:25"])
+    edits.save(tmp_path, edited((1, [note(4.0)])))
+    final = edits.finalize(tmp_path, LYRICS)
+    assert final["warnings"] == ["missing lyrics near 0:20–0:25"]
+    assert final["edited_lines"] == [1]
+
+
+def test_finalize_reports_stale_edits(tmp_path):
+    _write_generated(tmp_path, GENERATED, [])
+    edits.save(tmp_path, edited((1, [note(4.0)])))
+    final = edits.finalize(tmp_path, [LYRICS[0], ["other"], LYRICS[2]])
+    assert final["warnings"] == ["1 edited line no longer matches the lyrics; "
+                                 "set aside in the editor"]

tests/test_grid.py

@@ -169,7 +169,7 @@ def test_grid_stage_writes_song_txt(tmp_path, monkeypatch):
     monkeypatch.setattr(grid_stage_mod.grid, "detect_bpm", lambda p: 120.0)
     ctx = make_ctx(work, language="English", genre="Pop", year=1999)
     stage = grid_stage_mod.grid_stage()
-    assert stage.outputs(ctx) == [work / "song.txt"]
+    assert stage.outputs(ctx) == [work / "song.txt", work / "notes_final.json"]
     stage.run(ctx)
     song = parse_ultrastar((work / "song.txt").read_text(encoding="utf-8"))
     assert [k for k, _ in song.headers] == [
@@ -257,3 +257,45 @@ def test_grid_stage_reports_no_notes_as_pipeline_error(tmp_path, monkeypatch):
 def test_grid_stage_registered():
     names = [s.name for s in all_stages()]
     assert names[-1] == "package" and names[-2] == "grid"
+
+
+def _regenerate(work, pitch):
+    """What a rebuild leaves: fresh notes for both lines, at `pitch`."""
+    (work / "notes_sec.json").write_text(json.dumps({
+        "notes": [{**note(1.0, 0.5, pitch, "one"), "line_index": 0},
+                  {**note(3.0, 0.5, pitch, "three"), "line_index": 1}],
+        "breaks": [2.0], "warnings": []}))
+
+
+def test_an_edited_line_survives_a_rebuild(tmp_path, monkeypatch):
+    from songtrace import edits
+    work = tmp_path / "w"
+    write_work(work)
+    monkeypatch.setattr(grid_stage_mod.grid, "detect_bpm", lambda p: 120.0)
+    ctx = make_ctx(work)
+    ctx.inp.lyrics_text = "one\nthree"
+    data = edits.load(work, [["one"], ["three"]])
+    edits.set_line(data, 1, [{"start": 3.2, "duration": 0.6, "pitch": 9,
+                              "text": "three", "kind": "*"}])
+    edits.save(work, data)
+    for regenerated_pitch in (0, 4):
+        _regenerate(work, regenerated_pitch)
+        grid_stage_mod.grid_stage().run(ctx)
+        song = parse_ultrastar((work / "song.txt").read_text(encoding="utf-8"))
+        notes = [n for n in song.body if isinstance(n, Note)]
+        assert [(n.kind, n.pitch, n.text) for n in notes] == [
+            (":", regenerated_pitch, "one"), ("*", 9, "three")]
+
+
+def test_rechart_replaces_the_notes_and_keeps_headers_and_tempo(tmp_path):
+    from songtrace.grid import rechart
+    from songtrace.ultrastar import serialize_ultrastar
+    txt = tmp_path / "song.txt"
+    txt.write_text(serialize_ultrastar(to_ultrastar(
+        [note(1.0, 0.5)], [], 300.0,
+        [("TITLE", "Song"), ("VIDEO", "Band - Song.mp4")])), encoding="utf-8")
+    rechart(txt, [{**note(2.0, 0.5, 7, "new"), "kind": "F"}], [])
+    song = parse_ultrastar(txt.read_text(encoding="utf-8"))
+    assert [k for k, _ in song.headers] == ["TITLE", "VIDEO", "BPM", "GAP"]
+    assert get_header(song, "BPM") == "300" and get_header(song, "GAP") == "2000"
+    assert [(n.kind, n.pitch, n.text) for n in song.body] == [("F", 7, "new")]

tests/test_web.py

@@ -1412,13 +1412,16 @@ def test_review_endpoint_groups_notes_by_line(tmp_path):
     wait_status(client, job_id, "done")
     body = client.get(f"/api/jobs/{job_id}/review").json()
     assert body["lines"] == [
-        {"index": 0, "text": "hi", "dropped": False,
-         "notes": [{"start": 1.0, "end": 1.4, "pitch": 0, "low_confidence": False}],
+        {"index": 0, "text": "hi", "dropped": False, "edited": False,
+         "notes": [{"start": 1.0, "end": 1.4, "pitch": 0, "low_confidence": False,
+                    "text": "hi", "kind": ":"}],
          "words": []},
-        {"index": 1, "text": "there", "dropped": False,
-         "notes": [{"start": 5.0, "end": 5.3, "pitch": 2, "low_confidence": True}],
+        {"index": 1, "text": "there", "dropped": False, "edited": False,
+         "notes": [{"start": 5.0, "end": 5.3, "pitch": 2, "low_confidence": True,
+                    "text": "there", "kind": ":"}],
+         "words": []},
+        {"index": 2, "text": "gone", "dropped": True, "edited": False, "notes": [],
          "words": []},
-        {"index": 2, "text": "gone", "dropped": True, "notes": [], "words": []},
     ]
     assert body["missing_spans"] == [{"start": 8.0, "end": 9.5}]
 
@@ -1872,3 +1875,105 @@ def test_a_job_is_only_reported_finished_once_it_has_finished(tmp_path, monkeypa
     job = wait_status(client, job_id, "done")
     assert job["finished_at"] is not None
     assert not (cfg.work_dir / "uploads" / job_id / "v.mp4").exists()
+
+
+def chart_runner(inp, cfg, log=print, force_from=None, stages=None):
+    """Leaves what a real build does for the editor: generated notes and a
+    chart in the work dir, and the packaged chart."""
+    from songtrace.grid import to_ultrastar
+    from songtrace.ultrastar import serialize_ultrastar
+    work = cfg.work_dir / song_slug(inp)
+    work.mkdir(parents=True, exist_ok=True)
+    notes = [{"start": 1.0, "duration": 0.5, "pitch": 0, "text": "one",
+              "low_confidence": False, "line_index": 0},
+             {"start": 3.0, "duration": 0.5, "pitch": 2, "text": "three",
+              "low_confidence": True, "line_index": 1}]
+    warnings = ["1 low-confidence note near 0:03–0:03"]
+    (work / "notes_sec.json").write_text(json.dumps(
+        {"notes": notes, "breaks": [2.0], "warnings": warnings}))
+    chart = serialize_ultrastar(to_ultrastar(notes, [2.0], 300.0, [("TITLE", inp.title)]))
+    (work / "song.txt").write_text(chart, encoding="utf-8")
+    out = cfg.output_dir / song_slug(inp)
+    out.mkdir(parents=True, exist_ok=True)
+    (out / f"{song_slug(inp)}.txt").write_text(chart, encoding="utf-8")
+    return PipelineResult(output_dir=out, warnings=warnings)
+
+
+def make_charted_job(client, lyrics="one\nthree"):
+    job_id = client.post("/api/jobs", data={
+        "title": "T", "artist": "A", "lyrics": lyrics, "youtube_url": "http://y"}).json()["id"]
+    wait_status(client, job_id, "done")
+    return job_id
+
+
+EDIT = {"notes": [{"start": 3.1, "duration": 0.5, "pitch": 9, "text": "three",
+                   "kind": "*"}]}
+
+
+def test_saving_a_line_rewrites_both_charts(tmp_path):
+    from songtrace.ultrastar import Note, parse_ultrastar
+    client, cfg = make_client(tmp_path, chart_runner)
+    job_id = make_charted_job(client)
+    resp = client.put(f"/api/jobs/{job_id}/lines/1", json=EDIT)
+    assert resp.status_code == 200
+    line = resp.json()["lines"][1]
+    assert line["edited"] and line["notes"][0]["kind"] == "*" and line["notes"][0]["pitch"] == 9
+    assert resp.json()["grid"] == {"bpm": 300.0, "gap_ms": 1000.0}
+    for txt in (cfg.work_dir / "A - T" / "song.txt", cfg.output_dir / "A - T" / "A - T.txt"):
+        notes = [n for n in parse_ultrastar(txt.read_text(encoding="utf-8")).body
+                 if isinstance(n, Note)]
+        assert [(n.kind, n.pitch) for n in notes] == [(":", 0), ("*", 9)], txt
+    # the fixed line was the one being warned about
+    assert client.get(f"/api/jobs/{job_id}").json()["warnings"] == []
+
+
+def test_reverting_a_line_returns_it_to_the_generated_notes(tmp_path):
+    client, _ = make_client(tmp_path, chart_runner)
+    job_id = make_charted_job(client)
+    client.put(f"/api/jobs/{job_id}/lines/1", json=EDIT)
+    resp = client.delete(f"/api/jobs/{job_id}/lines/1")
+    assert resp.status_code == 200
+    line = resp.json()["lines"][1]
+    assert not line["edited"] and line["notes"][0]["pitch"] == 2
+    assert client.delete(f"/api/jobs/{job_id}/lines/1").status_code == 404
+
+
+def test_line_edits_are_checked(tmp_path):
+    client, _ = make_client(tmp_path, chart_runner)
+    job_id = make_charted_job(client)
+    assert client.put(f"/api/jobs/{job_id}/lines/5", json=EDIT).status_code == 404
+    bad = {"notes": [{**EDIT["notes"][0], "kind": "X"}]}
+    assert client.put(f"/api/jobs/{job_id}/lines/1", json=bad).status_code == 400
+    assert client.put(f"/api/jobs/{job_id}/lines/1", json={"x": 1}).status_code == 400
+
+
+def test_a_song_cannot_be_edited_while_it_is_being_rebuilt(tmp_path):
+    gate = threading.Event()
+
+    def runner(inp, cfg, log=print, force_from=None, stages=None):
+        if force_from:
+            gate.wait(timeout=5)
+        return chart_runner(inp, cfg)
+
+    client, _ = make_client(tmp_path, runner)
+    try:
+        job_id = make_charted_job(client)
+        client.post(f"/api/jobs/{job_id}/rebuild-full")
+        assert client.put(f"/api/jobs/{job_id}/lines/1", json=EDIT).status_code == 409
+    finally:
+        gate.set()
+
+
+def test_a_stale_edit_can_be_discarded(tmp_path):
+    from songtrace import edits
+    client, cfg = make_client(tmp_path, chart_runner)
+    first = make_charted_job(client)
+    client.put(f"/api/jobs/{first}/lines/1", json=EDIT)
+    # the same song again with that line's words changed
+    second = make_charted_job(client, lyrics="one\nthree more")
+    work = cfg.work_dir / "A - T"
+    assert len(edits.load(work, [["one"], ["three", "more"]])["stale"]) == 1
+    resp = client.delete(f"/api/jobs/{second}/stale/0")
+    assert resp.status_code == 200 and resp.json()["stale"] == []
+    assert not (work / edits.EDITS_FILE).exists()
+    assert client.delete(f"/api/jobs/{second}/stale/0").status_code == 404