ffmpeg-skill 1.15.0 → 1.16.0

package/scripts/_common/text.py

@@ -0,0 +1,1515 @@
+"""Text people can see: font resolution per script, emoji clusters and their assets, drawtext
+escaping and option building, and the per-character advance table the caption wrap measures with.
+"""
+from __future__ import annotations
+
+import os
+import platform
+import re
+import shutil
+import subprocess
+import unicodedata
+from pathlib import Path
+from typing import Any, Dict, List, Optional, Sequence, Tuple
+from _common.decision import escape_filter_path
+from _common.emit import die, info
+from _common.runner import STATE, _DRAWTEXT_PENDING, _drawtext_tmpdir, ffmpeg_version
+
+
+def drawtext_boxborderw(vertical: int, horizontal: int) -> str:
+    """drawtext's per-side `boxborderw=top|right|bottom|left` (and the two-value `v|h` form)
+    arrived in FFmpeg 6.1; 5.x and 6.0 reject the `|` with "Error setting option boxborderw"
+    (found by the FFmpeg 5.1.1 CI job, #146). Older builds get the larger single value."""
+    if ffmpeg_version() >= (6, 1):
+        return f"{vertical}|{horizontal}"
+    return str(max(vertical, horizontal))
+
+
+def default_font_file(font_name: str) -> Optional[str]:
+    """Resolve `font_name` to a concrete on-disk font file, so a caller can tell drawtext
+    `fontfile=<path>` instead of `font=<name>`, when possible.
+
+    On some real Windows ffmpeg builds (winget's gyan.dev 9.x), drawtext's own fontconfig
+    resolution crashes with an access violation whenever it has to resolve a font by family name
+    -- with or without a valid fonts.conf on FONTCONFIG_FILE. `fontfile=` is the only form
+    confirmed not to crash (#100), since it never touches fontconfig at all. `font_name` itself is
+    ignored on Windows for that reason: a fixed, near-universally-present system font is used
+    instead of trying to resolve the requested family (which would crash the same way).
+
+    On Linux/macOS this is best-effort and uses the real requested family: `fc-match` reports the
+    same file fontconfig would resolve `font_name` to anyway, so a caller gets the identical font,
+    just already resolved to a path -- fontfile= skips a redundant fontconfig lookup and equally
+    sidesteps the same class of crash if it exists on some build there too, but the fallback below
+    (returning None) is exercised routinely there, not just on failure.
+
+    Returns None when nothing could be resolved (fc-match missing/unavailable, or no well-known
+    Windows font file present); the caller falls back to font=<font_name>, the prior behaviour.
+    """
+    if platform.system() == "Windows":
+        windir = os.environ.get("WINDIR", "C:\\Windows")
+        fonts = Path(windir) / "Fonts"
+        # The requested family first: a file whose name starts with the family name with spaces
+        # removed (Noto Sans CJK JP -> NotoSansCJKjp-Regular.otf, Meiryo -> meiryo.ttc), then the
+        # common CJK system fonts when the request looks CJK (so Japanese text does not render as
+        # boxes in Arial), and Arial only as the last resort.
+        wanted = re.sub(r"[^a-z0-9]", "", (font_name or "").lower())
+        try:
+            files = sorted(fonts.iterdir()) if fonts.is_dir() else []
+        except OSError:
+            files = []
+        if wanted:
+            for f in files:
+                stem = re.sub(r"[^a-z0-9]", "", f.stem.lower())
+                if f.suffix.lower() in (".ttf", ".otf", ".ttc") and stem.startswith(wanted):
+                    return str(f)
+        if any(k in wanted for k in ("cjk", "gothic", "mincho", "meiryo", "yugoth", "msgothic", "malgun", "simhei", "simsun", "jp", "kr", "sc", "tc")):
+            for name in ("NotoSansCJKjp-Regular.otf", "NotoSansCJK-Regular.ttc", "meiryo.ttc", "YuGothM.ttc", "msgothic.ttc", "malgun.ttf", "msyh.ttc"):
+                if (fonts / name).exists():
+                    return str(fonts / name)
+        candidate = fonts / "arial.ttf"
+        return str(candidate) if candidate.exists() else None
+    exe = shutil.which("fc-match")
+    if not exe:
+        return None
+    try:
+        proc = subprocess.run([exe, "--format=%{file}\n", font_name], stdout=subprocess.PIPE, stderr=subprocess.PIPE, text=True, timeout=5)
+    except (subprocess.TimeoutExpired, OSError):
+        return None
+    if proc.returncode != 0:
+        return None
+    path = proc.stdout.splitlines()[0].strip() if proc.stdout.strip() else ""
+    return path if path and os.path.exists(path) else None
+
+
+# --------------------------------------------------------------------------- script detection
+# 1.12: non-Latin caption/overlay text used to render as tofu (empty boxes) whenever the default
+# family carried no glyphs for it -- silently, because fontconfig substitutes SOMETHING for every
+# request and ffmpeg exits 0 either way. The tools now detect the script of the text they are about
+# to draw and resolve a font file that actually covers it; nothing found is a failed job, not a
+# warning (a video full of boxes is not a delivery).
+SCRIPTS = ("ja", "zh", "ko", "ar", "he", "hi", "bn", "ta", "th", "lo", "ru", "el", "latin")
+
+
+LANGUAGE_NAMES = {
+    "ja": "Japanese", "zh": "Chinese", "ko": "Korean", "ar": "Arabic", "he": "Hebrew",
+    "hi": "Devanagari (Hindi/Marathi/Nepali)", "th": "Thai", "ru": "Cyrillic (Russian and others)",
+    "el": "Greek", "latin": "Latin", "bn": "Bengali", "ta": "Tamil", "lo": "Lao",
+}
+
+
+# fontconfig's own :lang= codes for each script we detect (zh uses zh-cn, the Simplified subset
+# every CJK font that claims zh carries; the rest are the plain two-letter codes).
+FC_LANG = {"ja": "ja", "zh": "zh-cn", "ko": "ko", "ar": "ar", "he": "he", "hi": "hi", "th": "th", "ru": "ru", "el": "el",
+           "bn": "bn", "ta": "ta", "lo": "lo"}
+
+
+# Families tried in order, best first. The names are matched case-insensitively against the start
+# of any family fontconfig reports for a file, so "Noto Sans CJK JP" also matches
+# "Noto Sans CJK JP Black". Anything not listed still qualifies -- it just sorts after these.
+PREFERRED_FAMILIES = {
+    "ja": ["Noto Sans CJK JP", "Noto Serif CJK JP", "Noto Sans JP", "Source Han Sans", "IPAPGothic", "IPAGothic", "IPA", "VL Gothic", "TakaoGothic", "WenQuanYi Zen Hei"],
+    "zh": ["Noto Sans CJK SC", "Noto Serif CJK SC", "Noto Sans SC", "Source Han Sans", "WenQuanYi Zen Hei", "WenQuanYi Micro Hei", "Droid Sans Fallback"],
+    "ko": ["Noto Sans CJK KR", "Noto Serif CJK KR", "Noto Sans KR", "Source Han Sans K", "NanumGothic", "Nanum Gothic", "Malgun Gothic", "WenQuanYi Zen Hei"],
+    "ar": ["Noto Sans Arabic", "Noto Naskh Arabic", "Amiri", "Scheherazade", "DejaVu Sans", "FreeSans", "FreeSerif"],
+    "he": ["Noto Sans Hebrew", "Noto Serif Hebrew", "DejaVu Sans", "FreeSans", "FreeSerif"],
+    "hi": ["Noto Sans Devanagari", "Noto Serif Devanagari", "Lohit Devanagari", "Mangal", "Nirmala UI", "Samyak Devanagari", "FreeSans", "FreeSerif"],
+    "th": ["Noto Sans Thai", "Noto Serif Thai", "Loma", "Garuda", "Waree", "Umpush", "Norasi", "Sarabun", "Leelawadee UI", "FreeSerif"],
+    "bn": ["Noto Sans Bengali", "Noto Serif Bengali", "Lohit Bengali", "Mukti Narrow", "Vrinda", "Nirmala UI", "FreeSerif"],
+    "ta": ["Noto Sans Tamil", "Noto Serif Tamil", "Lohit Tamil", "Latha", "Nirmala UI", "FreeSerif"],
+    "lo": ["Noto Sans Lao", "Noto Serif Lao", "Phetsarath OT", "Souliyo Unicode", "Saysettha OT", "DokChampa", "Leelawadee UI"],
+    "ru": ["Noto Sans", "DejaVu Sans", "Liberation Sans", "FreeSans", "FreeSerif"],
+    "el": ["Noto Sans", "DejaVu Sans", "Liberation Sans", "FreeSans", "FreeSerif"],
+}
+
+
+# Windows has no fontconfig: the system fonts are looked up by file name instead, best first.
+WINDOWS_FONTS = {
+    "ko": [("malgun.ttf", "Malgun Gothic"), ("gulim.ttc", "Gulim"), ("batang.ttc", "Batang")],
+    "zh": [("msyh.ttc", "Microsoft YaHei"), ("simhei.ttf", "SimHei"), ("simsun.ttc", "SimSun")],
+    "ja": [("meiryo.ttc", "Meiryo"), ("YuGothM.ttc", "Yu Gothic Medium"), ("YuGothR.ttc", "Yu Gothic"), ("msgothic.ttc", "MS Gothic")],
+    "ar": [("tahoma.ttf", "Tahoma"), ("arial.ttf", "Arial")],
+    "he": [("tahoma.ttf", "Tahoma"), ("arial.ttf", "Arial")],
+    "hi": [("mangal.ttf", "Mangal"), ("Nirmala.ttf", "Nirmala UI"), ("NirmalaB.ttf", "Nirmala UI")],
+    "th": [("leelawui.ttf", "Leelawadee UI"), ("leelawad.ttf", "Leelawadee"), ("tahoma.ttf", "Tahoma")],
+    "bn": [("Nirmala.ttf", "Nirmala UI"), ("vrinda.ttf", "Vrinda")],
+    "ta": [("Nirmala.ttf", "Nirmala UI"), ("latha.ttf", "Latha")],
+    "lo": [("leelawui.ttf", "Leelawadee UI"), ("DokChamp.ttf", "DokChampa")],
+    "ru": [("arial.ttf", "Arial"), ("segoeui.ttf", "Segoe UI")],
+    "el": [("arial.ttf", "Arial"), ("segoeui.ttf", "Segoe UI")],
+}
+
+
+_SCRIPT_RANGES = (
+    ("ko", ((0x1100, 0x11FF), (0x3130, 0x318F), (0xA960, 0xA97F), (0xAC00, 0xD7FF))),   # Hangul syllables + Jamo
+    ("kana", ((0x3040, 0x309F), (0x30A0, 0x30FF), (0x31F0, 0x31FF), (0xFF66, 0xFF9F))),  # hiragana/katakana
+    ("han", ((0x3400, 0x4DBF), (0x4E00, 0x9FFF), (0xF900, 0xFAFF), (0x20000, 0x2A6DF))),
+    ("ar", ((0x0600, 0x06FF), (0x0750, 0x077F), (0x08A0, 0x08FF), (0xFB50, 0xFDFF), (0xFE70, 0xFEFF))),
+    ("he", ((0x0590, 0x05FF), (0xFB1D, 0xFB4F))),
+    ("hi", ((0x0900, 0x097F), (0xA8E0, 0xA8FF))),
+    ("bn", ((0x0980, 0x09FF),)),
+    ("ta", ((0x0B80, 0x0BFF),)),
+    ("th", ((0x0E00, 0x0E7F),)),
+    ("lo", ((0x0E80, 0x0EFF),)),
+    ("ru", ((0x0400, 0x04FF), (0x0500, 0x052F), (0x2DE0, 0x2DFF))),
+    ("el", ((0x0370, 0x03FF), (0x1F00, 0x1FFF))),
+)
+
+
+# --------------------------------------------------------------------------- emoji (1.15)
+# Emoji are orthogonal to the writing system: "やった 🎉" is Japanese AND emoji. They are detected
+# separately from detect_script() so a cue's font resolution is still decided by its letters.
+EMOJI_RANGES = (
+    (0x1F300, 0x1FAFF),   # symbols & pictographs, supplemental, extended-A
+    (0x1F000, 0x1F0FF),   # mahjong/domino/playing cards
+    (0x2600, 0x27BF),     # misc symbols + dingbats
+    (0x2B00, 0x2BFF),     # misc symbols and arrows
+    (0xFE0F, 0xFE0F),     # VS16 (emoji presentation selector)
+    (0x1F1E6, 0x1F1FF),   # regional indicators (flags)
+    (0x20E3, 0x20E3),     # combining enclosing keycap
+    (0x1F3FB, 0x1F3FF),   # skin-tone modifiers
+)
+
+
+# U+200D ZWJ is deliberately NOT in EMOJI_RANGES: it is ordinary Indic/Persian orthography
+# (क्‍ष is ka + virama + ZWJ + ssa) and only becomes emoji glue *between two emoji bases*.
+# Characters that never START a cluster: they bind to whatever stands before them.
+_EMOJI_TAIL = frozenset({0x200D, 0xFE0F, 0x20E3} | set(range(0x1F3FB, 0x1F400)))
+
+
+_EMOJI_REGIONAL = range(0x1F1E6, 0x1F200)
+
+
+_ZWJ = 0x200D
+
+
+_VS15 = 0xFE0E   # text-presentation selector: "draw this as a character, not as an emoji"
+
+
+_VS16 = 0xFE0F
+
+
+_KEYCAP = 0x20E3
+
+
+_KEYCAP_BASES = frozenset("0123456789#*")
+
+
+def _is_emoji_char(ch: str) -> bool:
+    cp = ord(ch)
+    return any(lo <= cp <= hi for lo, hi in EMOJI_RANGES)
+
+
+def _is_emoji_base(ch: str) -> bool:
+    """Can this character START an emoji cluster? Pictographs and regional indicators can;
+    the joiners and modifiers (ZWJ, VS16, keycap, skin tone) never can -- they only bind to an
+    emoji base that already stands before them. Without this, a ZWJ or a VS16 sitting after an
+    ordinary letter turned that letter into "an emoji" and the PNG route replaced it with a gap."""
+    return ord(ch) not in _EMOJI_TAIL and _is_emoji_char(ch)
+
+
+def emoji_clusters(text: str) -> "List[Tuple[int, str]]":
+    """(index in `text`, cluster) for every emoji in it, ZWJ sequences, VS16, keycaps, flag pairs
+    and skin-tone modifiers kept together -- 👩‍💻 is one cluster, not three, and 1️⃣ starts at the
+    digit even though the digit is not itself an emoji character.
+
+    A cluster can only START at an emoji base (a pictograph, a regional indicator) or at a keycap
+    base (`0-9 # *`) that is actually followed by U+20E3. A ZWJ is glue *inside* a cluster, never
+    a starter and never a tail on its own: `क्‍ष` (Hindi ka + virama + ZWJ + ssa) and `abc‍def`
+    contain no emoji. A base explicitly marked with U+FE0E (VS15, text presentation) is likewise
+    not an emoji -- the author asked for the character, not the picture.
+    """
+    out: "List[Tuple[int, str]]" = []
+    i = 0
+    n = len(text or "")
+    while i < n:
+        ch = text[i]
+        start = i
+        if _is_emoji_base(ch):
+            j = i + 1
+            if j < n and ord(text[j]) == _VS15:      # text presentation requested: not an emoji
+                i = j + 1
+                continue
+        elif ch in _KEYCAP_BASES:
+            j = i + 1
+            if j < n and ord(text[j]) == _VS16:
+                j += 1
+            if not (j < n and ord(text[j]) == _KEYCAP):
+                i += 1
+                continue
+            j += 1
+        else:
+            i += 1
+            continue
+        # extend: modifiers bind rightwards, a ZWJ only when a real emoji base follows it
+        while j < n:
+            cp = ord(text[j])
+            if cp in (_VS16, _KEYCAP) or 0x1F3FB <= cp <= 0x1F3FF:
+                j += 1
+                continue
+            if cp == _ZWJ and j + 1 < n and _is_emoji_base(text[j + 1]):
+                j += 2
+                continue
+            if (j == start + 1 and ord(ch) in _EMOJI_REGIONAL and cp in _EMOJI_REGIONAL):
+                j += 1
+                continue
+            break
+        out.append((start, text[start:j]))
+        i = j
+    return out
+
+
+def has_emoji(text: str) -> bool:
+    return bool(emoji_clusters(text or ""))
+
+
+def emoji_codepoint_name(cluster: str) -> str:
+    """The asset filename stem for a cluster: lowercase hex code points joined by '-', the
+    Twemoji/Noto convention (1f389, 1f469-200d-1f4bb, 1f1ef-1f1f5)."""
+    return "-".join(f"{ord(c):x}" for c in cluster)
+
+
+def _emoji_name_candidates(cluster: str) -> "List[str]":
+    """Asset stems to try, most specific first: exact, without VS16, without skin tone, the ZWJ
+    sequence reduced to its first code point, the bare base."""
+    cps = [ord(c) for c in cluster]
+    names = [emoji_codepoint_name(cluster)]
+
+    def add(seq):
+        name = "-".join(f"{c:x}" for c in seq)
+        if name and name not in names:
+            names.append(name)
+    add([c for c in cps if c != 0xFE0F])
+    add([c for c in cps if c != 0xFE0F and not (0x1F3FB <= c <= 0x1F3FF)])
+    if 0x200D in cps:
+        add([cps[0]])
+    add([cps[0]])
+    return names
+
+
+def emoji_asset_for(cluster: str, assets_dir: "Optional[str]") -> "Optional[str]":
+    """The PNG for `cluster` under `assets_dir`, or None when nothing matches."""
+    if not assets_dir or not os.path.isdir(assets_dir):
+        return None
+    for name in _emoji_name_candidates(cluster):
+        for ext in (".png", ".PNG"):
+            candidate = os.path.join(assets_dir, name + ext)
+            if os.path.isfile(candidate):
+                return candidate
+    return None
+
+
+EMOJI_ASSET_HINT = (
+    "point --emoji-assets at a directory of PNGs named by code point (1f389.png): "
+    "twemoji/assets/72x72 (Twemoji, CC-BY 4.0) or noto-emoji/png/128 (Noto Emoji, OFL/Apache-2.0) "
+    "are the two people already have. The skill has no network at runtime, so the assets must "
+    "already exist on this machine -- nothing is ever downloaded")
+
+
+_EMOJI_COLOR_FAMILIES = ("Noto Color Emoji", "Apple Color Emoji", "Segoe UI Emoji")
+
+
+_EMOJI_SUPPORT_CACHE: "Dict[Tuple[Optional[str], bool], Dict[str, Any]]" = {}
+
+
+def _emoji_color_font() -> "Tuple[Optional[str], Optional[str], bool]":
+    """(family, file, fontconfig_answered) for the first installed colour emoji family."""
+    exe = shutil.which("fc-list")
+    if not exe:
+        return None, None, False
+    for family in _EMOJI_COLOR_FAMILIES:
+        try:
+            proc = subprocess.run([exe, f":family={family}", "file"], stdout=subprocess.PIPE,
+                                  stderr=subprocess.DEVNULL, text=True, timeout=10)
+        except (subprocess.TimeoutExpired, OSError):
+            return None, None, False
+        if proc.returncode != 0:
+            return None, None, False
+        for line in proc.stdout.splitlines():
+            path = line.split(":", 1)[0].strip()
+            if path and os.path.exists(path):
+                return family, path, True
+    return None, None, True
+
+
+def _libass_color_probe() -> "Optional[bool]":
+    """Does THIS ffmpeg render an emoji in colour through libass? Answered by a render, never by
+    the font listing: Noto Color Emoji installs happily on builds whose freetype/libass has no
+    colour-bitmap path at all, and those render a monochrome outline instead (measured). ~80 ms.
+    None means the probe could not be run (no ffmpeg, a failure) -- unknown, not false."""
+    exe = shutil.which("ffmpeg")
+    if not exe:
+        return None
+    import tempfile
+    with tempfile.TemporaryDirectory() as td:
+        srt = os.path.join(td, "e.srt")
+        with open(srt, "w", encoding="utf-8") as fh:
+            fh.write("1\n00:00:00,000 --> 00:00:01,000\n\U0001F389\n")
+        try:
+            proc = subprocess.run(
+                [exe, "-hide_banner", "-loglevel", "error", "-f", "lavfi",
+                 "-i", "color=c=black:s=64x64:d=0.04",
+                 "-vf", "subtitles=" + srt.replace("\\", "/"), "-frames:v", "1",
+                 "-f", "rawvideo", "-pix_fmt", "rgb24", "-"],
+                stdout=subprocess.PIPE, stderr=subprocess.DEVNULL, timeout=20)
+        except (subprocess.TimeoutExpired, OSError):
+            return None
+    if proc.returncode != 0 or len(proc.stdout) < 64 * 64 * 3:
+        return None
+    data = proc.stdout
+    for i in range(0, 64 * 64 * 3, 3):
+        r, g, b = data[i], data[i + 1], data[i + 2]
+        if max(r, g, b) - min(r, g, b) > 40:
+            return True
+    return False
+
+
+def emoji_support(assets: "Optional[str]" = None, probe: bool = True) -> "Dict[str, Any]":
+    """What this machine can actually do with emoji, cached per process.
+
+    `mode` is `color` when a render probe proves libass draws colour, else `png` when an assets
+    directory resolves, else `mono` when some installed face has a glyph at all, else `none`.
+    An installed colour emoji font proves nothing on its own -- that is why `libass_color` comes
+    from a render (see references/gotchas.md#emoji). `probe=False` (`contract --json --static`, and every
+    static/JSON-only path) skips the render entirely and leaves `libass_color` unknown.
+    """
+    key = (assets or None, bool(probe))
+    if key in _EMOJI_SUPPORT_CACHE:
+        return dict(_EMOJI_SUPPORT_CACHE[key])
+    family, file, fc_answered = _emoji_color_font()
+    libass_color = _libass_color_probe() if probe else None
+    assets_dir = assets if (assets and os.path.isdir(assets)) else None
+    if libass_color:
+        mode = "color"
+    elif assets_dir:
+        mode = "png"
+    elif family:
+        mode = "mono"
+    elif not fc_answered:
+        # No fontconfig to ask (a static ffmpeg build, a bare container): the PNG path needs none,
+        # so the honest answer is png-or-none, never "none because fc-list is missing".
+        mode = "none"
+    else:
+        mode = "none"
+    if not fc_answered:
+        detail = "no fontconfig on this machine; the PNG overlay path needs none"
+    elif libass_color:
+        detail = f"{family or 'an installed face'} renders in colour through libass on this ffmpeg"
+    elif family and libass_color is False:
+        detail = f"{family} installed but libass renders it monochrome on this build"
+    elif family and libass_color is None:
+        detail = f"{family} installed; the colour render probe was not run"
+    elif assets_dir:
+        detail = "no colour emoji family installed; using the PNG assets directory"
+    else:
+        detail = "no colour emoji family installed and no --emoji-assets directory"
+    result = {"mode": mode, "color_font": family, "color_font_file": file,
+              "libass_color": libass_color, "assets": assets_dir,
+              "detail": detail, "fix": EMOJI_ASSET_HINT}
+    _EMOJI_SUPPORT_CACHE[key] = result
+    return dict(result)
+
+
+def resolve_emoji_assets(flag: "Optional[str]" = None, project: "Optional[str]" = None,
+                         brand: "Optional[dict]" = None) -> "Optional[str]":
+    """--emoji-assets DIR, else the project key, else brand.json, else FFMPEG_SKILL_EMOJI_ASSETS.
+    A directory that was named but does not exist is a failed job, never a silent downgrade."""
+    brand = brand or {}
+    styles = (brand.get("styles") or {}).get("caption") or {}
+    for value, where in ((flag, "--emoji-assets"), (project, "the project's text.emoji_assets"),
+                         (styles.get("emoji_assets"), "brand.json styles.caption.emoji_assets"),
+                         (brand.get("emoji_assets"), "brand.json emoji_assets"),
+                         (os.environ.get("FFMPEG_SKILL_EMOJI_ASSETS"), "FFMPEG_SKILL_EMOJI_ASSETS")):
+        if not value:
+            continue
+        if not os.path.isdir(str(value)):
+            die(f"{where}: {value} is not a readable directory -- {EMOJI_ASSET_HINT}", kind="input")
+        return str(value)
+    return None
+
+
+# --------------------------------------------------------------------------- text measurement (1.12)
+# Moved here in 1.15 so graphics.py's ASS route and the emoji placement share caption.py's table.
+# Average advance width per character, in em (a fraction of the font size). Proportional Latin text
+# averages a bit over half an em; CJK and Thai are drawn on a full-width grid; Arabic/Hebrew and
+# Devanagari sit in between. These are deliberately averages, not per-glyph metrics: measuring the
+# real advance needs a font parser (no stdlib one) and would still be wrong for libass's own
+# shaping, while a cue wrapped from an average is right to within a character on every line.
+# (Latin is measured per character from LATIN_EM below, not from this average.)
+ADVANCE_EM = {"ja": 1.0, "zh": 1.0, "ko": 1.0, "th": 1.0, "hi": 0.7, "ar": 0.6, "he": 0.6,
+              "ru": 0.55, "el": 0.55, "latin": 0.55}
+
+
+# Scripts written without spaces: a line breaks between any two characters.
+NO_SPACE_SCRIPTS = ("ja", "zh", "ko", "th")
+
+
+# Per-character Latin advances in em, read off DejaVu Sans (the default caption family, and close
+# enough to any other proportional sans for a wrap) and rounded UP: a capital runs 0.56-0.99 em
+# against the single 0.55 average that used to stand for all of Latin, so an all-caps caption --
+# the style most burn-ins use -- overflowed the safe area and was silently re-wrapped by libass
+# past --max-lines. Rounding up is the safe direction: libass re-wraps a too-long line, it never
+# un-wraps a short one. Characters outside the table fall back by class (0.7 uppercase/digit,
+# 0.57 lowercase and anything else Latin-ish).
+LATIN_EM = {
+    ' ': 0.32, '!': 0.41, '"': 0.46, '#': 0.84, '$': 0.64, '%': 0.96, '&': 0.78, "'": 0.28,
+    '(': 0.4, ')': 0.4, '*': 0.5, '+': 0.84, ',': 0.32, '-': 0.37, '.': 0.32, '/': 0.34, '0': 0.64,
+    '1': 0.64, '2': 0.64, '3': 0.64, '4': 0.64, '5': 0.64, '6': 0.64, '7': 0.64, '8': 0.64,
+    '9': 0.64, ':': 0.34, ';': 0.34, '<': 0.84, '=': 0.84, '>': 0.84, '?': 0.54, '@': 1.0,
+    'A': 0.69, 'B': 0.69, 'C': 0.7, 'D': 0.78, 'E': 0.64, 'F': 0.58, 'G': 0.78, 'H': 0.76,
+    'I': 0.3, 'J': 0.3, 'K': 0.66, 'L': 0.56, 'M': 0.87, 'N': 0.75, 'O': 0.79, 'P': 0.61,
+    'Q': 0.79, 'R': 0.7, 'S': 0.64, 'T': 0.62, 'U': 0.74, 'V': 0.69, 'W': 0.99, 'X': 0.69,
+    'Y': 0.62, 'Z': 0.69, '[': 0.4, '\\': 0.34, ']': 0.4, '^': 0.84, '_': 0.5, '`': 0.5, 'a': 0.62,
+    'b': 0.64, 'c': 0.55, 'd': 0.64, 'e': 0.62, 'f': 0.36, 'g': 0.64, 'h': 0.64, 'i': 0.28,
+    'j': 0.28, 'k': 0.58, 'l': 0.28, 'm': 0.98, 'n': 0.64, 'o': 0.62, 'p': 0.64, 'q': 0.64,
+    'r': 0.42, 's': 0.53, 't': 0.4, 'u': 0.64, 'v': 0.6, 'w': 0.82, 'x': 0.6, 'y': 0.6, 'z': 0.53,
+    '{': 0.64, '|': 0.34, '}': 0.64, '~': 0.84
+}
+
+
+# Thai and Lao write some vowels BEFORE the consonant they belong to: the break must not land
+# between them and the base that follows.
+LEADING_VOWELS = set(range(0x0E40, 0x0E45)) | set(range(0x0EC0, 0x0EC5))
+
+
+def _is_mark(ch: str) -> bool:
+    """A character that hangs off the one before it: a combining mark (any script) or one of the
+    Thai/Lao vowel signs and tone marks, which are Mn/Mc but carry no combining class."""
+    return unicodedata.combining(ch) != 0 or unicodedata.category(ch) in ("Mn", "Mc")
+
+
+def _char_em(ch: str) -> float:
+    # CJK punctuation and the fullwidth forms (、。，！？　and U+FF01-FF60) are drawn on the same
+    # full-width grid as the ideographs they sit between, even though they are not "Han" to a
+    # script detector -- measuring them as Latin under-counts a wrapped CJK line by a character.
+    cp = ord(ch)
+    # A combining mark is drawn on top of (or under) its base and advances the pen by nothing:
+    # charging it a full em wrapped Thai and Devanagari lines far shorter than they needed to be.
+    if unicodedata.combining(ch) != 0 or unicodedata.category(ch) in ("Mn", "Cf"):
+        # "Cf" catches ZWJ/ZWNJ: an Indic joiner is orthography, and it advances the pen by
+        # nothing -- charging it a full em (it used to count as "emoji") shrank a Hindi line.
+        return 0.0
+    if 0x3000 <= cp <= 0x303F or 0xFF01 <= cp <= 0xFF60 or 0xFFE0 <= cp <= 0xFFE6:
+        return 1.0
+    script = char_script(ch)
+    if script == "emoji":
+        # 1.15: an emoji is drawn (or reserved) at a full em box, not at Latin's 0.57 -- counting
+        # it as Latin overflowed the safe area on an emoji-heavy line.
+        return 1.0
+    if script == "latin":
+        if ch in LATIN_EM:
+            return LATIN_EM[ch]
+        if ch.isupper() or ch.isdigit():
+            return 0.7
+        return 0.57
+    return ADVANCE_EM.get(script, 0.55)
+
+
+def text_width_em(text: str, emoji_em: float = 1.0) -> float:
+    """Width of `text` in em, from the per-script average advance table. `emoji_em` is what one
+    emoji cluster costs (--emoji-scale), so a wrap counts the box that will actually be drawn."""
+    total = 0.0
+    spans = {i: len(c) for i, c in emoji_clusters(text)}
+    i = 0
+    while i < len(text):
+        if i in spans:
+            total += emoji_em
+            i += spans[i]
+            continue
+        total += _char_em(text[i])
+        i += 1
+    return total
+
+
+def emoji_filter_chain(plan, base_label, out_label, first_input=1):
+    """(chains, inputs) that composite the planned PNGs on top of `base_label`.
+
+    `inputs` is a list of argv fragments, each ending in the asset path, to be appended to the
+    ffmpeg command in order (an overlay that fades needs `-loop 1` on its input so the still has
+    a timeline the fade filter can move along; one that does not is a plain `-i`).
+    """
+    overlays = plan.get("overlays") or []
+    if not overlays:
+        return [], []
+    # Group by everything that makes two uses of the same PNG a different STREAM: the fade is
+    # expressed in the cue's own timeline, so two cues cannot share one faded input.
+    def _key(o):
+        fades = (round(float(o.get("fade_in") or 0.0), 3), round(float(o.get("fade_out") or 0.0), 3))
+        window = (round(float(o["start"]), 3), round(float(o["end"]), 3)) if any(fades) else (None, None)
+        return (o["asset"], o["box"]) + fades + window
+
+    groups: "List[Tuple]" = []
+    for o in overlays:
+        if _key(o) not in groups:
+            groups.append(_key(o))
+    chains: List[str] = []
+    inputs: "List[List[str]]" = []
+    pads: "Dict[Tuple, List[str]]" = {}
+    for k, key in enumerate(groups):
+        asset, box, fin, fout, gstart, gend = key
+        uses = [o for o in overlays if _key(o) == key]
+        idx = first_input + k
+        labels = [f"e{k}_{j}" for j in range(len(uses))]
+        chain = f"[{idx}:v]format=rgba,scale={box}:{box}"
+        if fin or fout:
+            # -loop 1 gives the still an advancing timeline on the SAME clock as the main video,
+            # so the fade times below are the cue's own seconds. The emoji then appears and
+            # leaves with the text instead of popping in against a fading line.
+            # -t bounds the loop at the cue's end: an unbounded looped still never EOFs and the
+            # whole encode hangs (overlay keeps pulling from it after the main video is done).
+            inputs.append(["-loop", "1", "-t", f"{gend:.3f}", "-i", asset])
+            if fin:
+                chain += f",fade=t=in:st={gstart:.3f}:d={fin:.3f}:alpha=1"
+            if fout:
+                chain += f",fade=t=out:st={max(gstart, gend - fout):.3f}:d={fout:.3f}:alpha=1"
+        else:
+            inputs.append(["-i", asset])
+        if len(labels) > 1:
+            chain += f",split={len(labels)}"
+        chains.append(chain + "".join(f"[{l}]" for l in labels))
+        pads[key] = labels
+    cur = base_label
+    remaining = {key: list(v) for key, v in pads.items()}
+    for j, o in enumerate(overlays):
+        label = remaining[_key(o)].pop(0)
+        nxt = out_label if j == len(overlays) - 1 else f"eov{j}"
+        x = o["x"]
+        x = f"'{x}'" if isinstance(x, str) else x
+        # No eof_action=pass here: a PNG input is a SINGLE frame at pts 0, and eof_action=pass
+        # switches off overlay's default "hold the last frame of the secondary input", so the
+        # asset would be composited on frame 0 only and vanish for the rest of the cue (that is
+        # exactly what shipped first). eof_action=repeat (the default) holds the still for the
+        # whole timeline; enable= is what confines it to the cue's window.
+        chains.append(f"[{cur}][{label}]overlay=x={x}:y={o['y']}:"
+                      f"enable='between(t,{o['start']:.3f},{o['end']:.3f})'[{nxt}]")
+        cur = nxt
+    return chains, inputs
+
+
+# --------------------------------------------------------------------------- shaping (1.15)
+# Scripts whose correct rendering needs harfbuzz-class reordering and re-clustering (Indic matras,
+# Thai/Lao mark stacking). drawtext does NOT use harfbuzz even in an --enable-libharfbuzz build, so
+# these come out wrong through drawtext on every build and must go through libass. Arabic and
+# Hebrew are NOT here: drawtext's text_shaping uses fribidi, which does bidi and Arabic joining
+# correctly -- they only join this set on a build compiled without fribidi.
+SHAPING_SCRIPTS = frozenset({"hi", "bn", "ta", "te", "kn", "ml", "gu", "pa", "si", "th", "lo", "km", "my"})
+
+
+BIDI_SCRIPTS = frozenset({"ar", "he"})
+
+
+_SHAPING_BUILD_CACHE: "Dict[str, bool]" = {}
+
+
+def drawtext_shaping() -> "Dict[str, bool]":
+    """Which shaping libraries THIS ffmpeg was built with, from -buildconf (falling back to the
+    `configuration:` line of -version). Cached per process."""
+    if _SHAPING_BUILD_CACHE:
+        return dict(_SHAPING_BUILD_CACHE)
+    text = ""
+    exe = shutil.which("ffmpeg")
+    if exe:
+        for flag in ("-buildconf", "-version"):
+            try:
+                proc = subprocess.run([exe, "-hide_banner", flag], stdout=subprocess.PIPE,
+                                      stderr=subprocess.STDOUT, text=True, timeout=10)
+            except (subprocess.TimeoutExpired, OSError):
+                break
+            if proc.returncode == 0 and proc.stdout.strip():
+                text = proc.stdout
+                break
+    _SHAPING_BUILD_CACHE.update({"fribidi": "--enable-libfribidi" in text,
+                                 "harfbuzz": "--enable-libharfbuzz" in text})
+    return dict(_SHAPING_BUILD_CACHE)
+
+
+def needs_shaping(script: str) -> bool:
+    """Whether drawtext would render `script` wrongly on this build."""
+    if script in SHAPING_SCRIPTS:
+        return True
+    return script in BIDI_SCRIPTS and not drawtext_shaping()["fribidi"]
+
+
+def font_family_of_file(path: str) -> "Optional[str]":
+    """The family name of a font FILE -- what libass wants, given a --font-file. `fc-scan` reads
+    the file directly; without fontconfig the file stem is the honest best guess."""
+    if not path or not os.path.isfile(path):
+        return None
+    exe = shutil.which("fc-scan")
+    if exe:
+        try:
+            proc = subprocess.run([exe, "--format", "%{family[0]}", path], stdout=subprocess.PIPE,
+                                  stderr=subprocess.DEVNULL, text=True, timeout=10)
+            if proc.returncode == 0 and proc.stdout.strip():
+                return proc.stdout.strip().splitlines()[0].strip()
+        except (subprocess.TimeoutExpired, OSError):
+            pass
+    return Path(path).stem
+
+
+def char_script(ch: str) -> str:
+    """The script of one character: one of SCRIPTS, or "latin" for anything else (including
+    digits, punctuation and spaces -- they are measured and wrapped like Latin)."""
+    cp = ord(ch)
+    # 1.15: an emoji cluster is not Latin. detect_script() skips "emoji" the way it skips "latin",
+    # so font resolution still follows the letters around it.
+    if _is_emoji_char(ch):
+        return "emoji"
+    for name, ranges in _SCRIPT_RANGES:
+        for lo, hi in ranges:
+            if lo <= cp <= hi:
+                return "ja" if name == "kana" else ("zh" if name == "han" else name)
+    return "latin"
+
+
+def detect_script(text: str, lang: "Optional[str]" = None) -> str:
+    """Which script `text` is written in, as one of SCRIPTS.
+
+    Hangul wins for Korean, any kana makes the whole string Japanese (Japanese mixes kana and
+    Han), Han alone is Chinese. Mixed text is decided by character count: the non-Latin script
+    with the most characters wins, ties going to whichever appeared first, and text with no
+    non-Latin characters at all is "latin". `lang` (a --lang/--language hint, or brand.json's
+    `lang`) only resolves the one ambiguity the characters genuinely cannot: Han with no kana
+    is Chinese by default but Japanese (or Korean hanja) when the caller says so.
+    """
+    counts: "Dict[str, int]" = {}
+    order: "List[str]" = []
+    kana = 0
+    for ch in text or "":
+        s = char_script(ch)
+        if s in ("latin", "emoji"):
+            continue
+        if ord(ch) in range(0x3040, 0x3100) or ord(ch) in range(0x31F0, 0x3200) or ord(ch) in range(0xFF66, 0xFFA0):
+            kana += 1
+        if s not in counts:
+            order.append(s)
+        counts[s] = counts.get(s, 0) + 1
+    if kana:  # Japanese: the Han characters in the same string are Japanese too
+        counts["ja"] = counts.pop("ja", 0) + counts.pop("zh", 0)
+        order = [s for s in order if s != "zh"]
+    if not counts:
+        return "latin"
+    best = max(counts, key=lambda s: (counts[s], -order.index(s)))
+    hint = (lang or "").strip().lower().replace("_", "-").split("-")[0]
+    if not kana and best == "zh" and hint in ("ja", "zh", "ko"):
+        return hint  # Han-only text: only the caller knows whether it is Chinese, Japanese or hanja
+    return best
+
+
+_SCRIPT_FONT_CACHE: "Dict[Tuple[str, Optional[str]], Optional[Tuple[str, str]]]" = {}
+
+
+def _fc_list_fonts(fc_lang: str) -> "Optional[List[Tuple[str, List[str]]]]":
+    """(file, families) for every font fontconfig says covers `fc_lang`.
+
+    `[]` means fontconfig answered and nothing covers the language; `None` means it could not be
+    asked at all (no `fc-list` on PATH, or it failed/timed out) -- the difference between
+    "missing" and "unknown", which the caller must not collapse: unknown is not a refusal.
+    """
+    exe = shutil.which("fc-list")
+    if not exe:
+        return None
+    try:
+        proc = subprocess.run([exe, f":lang={fc_lang}", "file", "family"],
+                              stdout=subprocess.PIPE, stderr=subprocess.DEVNULL, text=True, timeout=10)
+    except (subprocess.TimeoutExpired, OSError):
+        return None
+    if proc.returncode != 0:
+        return None
+    out = []
+    for line in proc.stdout.splitlines():
+        if ": " not in line:
+            continue
+        path, _, families = line.partition(": ")
+        path = path.strip()
+        if not path or not os.path.exists(path):
+            continue
+        names = [f.replace("\\-", "-").strip() for f in families.split(",") if f.strip()]
+        out.append((path, names or [Path(path).stem]))
+    return out
+
+
+def _family_rank(families: "Sequence[str]", preferred: "Sequence[str]") -> int:
+    for i, want in enumerate(preferred):
+        w = want.lower()
+        if any(f.lower().startswith(w) for f in families):
+            return i
+    return len(preferred)
+
+
+def _script_font_entry(script: str, family_hint: "Optional[str]" = None) -> "Optional[Tuple[str, str]]":
+    key = (script, family_hint)
+    if key in _SCRIPT_FONT_CACHE:
+        return _SCRIPT_FONT_CACHE[key]
+    _SCRIPT_FONT_CACHE[key] = result = _script_font_uncached(script, family_hint)
+    return result
+
+
+FC_UNKNOWN = "unknown"  # sentinel: fontconfig could not be asked (absent or failing), not "no font"
+
+
+def _script_font_uncached(script: str, family_hint: "Optional[str]" = None):
+    """(file, family), None when nothing covers `script`, or FC_UNKNOWN when it cannot be asked."""
+    if script not in FC_LANG:
+        return None
+    if platform.system() == "Windows":
+        fonts = Path(os.environ.get("WINDIR", "C:\\Windows")) / "Fonts"
+        for name, family in WINDOWS_FONTS.get(script, []):
+            if (fonts / name).exists():
+                return str(fonts / name), family
+        return None
+    preferred = list(PREFERRED_FAMILIES.get(script, []))
+    if family_hint:
+        preferred.insert(0, family_hint)
+    candidates = _fc_list_fonts(FC_LANG[script])
+    if candidates is None:
+        return FC_UNKNOWN
+    if not candidates:
+        return None
+    scored = []
+    for path, families in candidates:
+        joined = " ".join(families).lower()
+        stem = Path(path).stem.lower()
+        # "Unifont Sample" is fontconfig's tofu-with-hex-digits fallback: it "covers" every script
+        # by drawing the code point, which is exactly the unreadable result this feature exists to
+        # avoid -- it is only ever chosen when nothing else covers the script at all. A *Mono* face
+        # is legible but wrong for a caption band, so it sorts after every proportional one.
+        last_resort = 1 if "unifont" in joined else 0
+        mono = 1 if "mono" in joined else 0
+        # regular weights before Bold/Italic/Oblique cuts, so a default caption is not bold by accident
+        styled = 1 if any(k in stem for k in ("bold", "italic", "oblique", "light", "thin", "black")) else 0
+        scored.append((last_resort, _family_rank(families, preferred), mono, styled, path, families[0]))
+    scored.sort(key=lambda row: (row[0], row[1], row[2], row[3], row[4]))
+    best = scored[0]
+    return best[4], best[5]
+
+
+def font_for_script(script: str, family_hint: "Optional[str]" = None) -> "Optional[str]":
+    """A font FILE path that covers `script`, or None when this machine has none.
+
+    Linux/macOS ask fontconfig (`fc-list :lang=xx file family`) and rank what it reports by the
+    PREFERRED_FAMILIES table; Windows has no fontconfig, so the known system files are probed by
+    name. Cached per process: a caption job resolves the same script for every cue.
+    """
+    entry = _script_font_entry(script, family_hint)
+    return entry[0] if entry and entry is not FC_UNKNOWN else None
+
+
+def font_family_for_script(script: str, family_hint: "Optional[str]" = None) -> "Optional[str]":
+    """The family NAME of font_for_script()'s file -- what libass wants in an ASS Fontname."""
+    entry = _script_font_entry(script, family_hint)
+    return entry[1] if entry and entry is not FC_UNKNOWN else None
+
+
+def script_font_status(script: str) -> str:
+    """"available" (a font file covers `script`), "missing" (fontconfig answered, none does) or
+    "unknown" (there is no working fontconfig to ask). Only "missing" is a refusal."""
+    entry = _script_font_entry(script)
+    if entry is FC_UNKNOWN:
+        return "unknown"
+    return "available" if entry else "missing"
+
+
+def font_covers_script(font_name: str, script: str) -> bool:
+    """Whether the installed family `font_name` actually carries glyphs for `script`.
+
+    `fc-match` cannot answer this: given a family that IS installed it returns that family
+    whatever `:lang=` asks for (verified -- `fc-match "DejaVu Sans:lang=zh-cn"` answers
+    "DejaVu Sans", which has no Han glyphs at all). `fc-list :lang=xx:family=<name>` does: it
+    lists only files that satisfy BOTH, so an empty listing is the proof of no coverage. Unknown
+    (no fontconfig at all) counts as covering: a warning nobody can verify is worse than none.
+    """
+    if script not in FC_LANG or not font_name:
+        return True
+    exe = shutil.which("fc-list")
+    if not exe:
+        return True
+    try:
+        proc = subprocess.run([exe, f":lang={FC_LANG[script]}:family={font_name}", "file"],
+                              stdout=subprocess.PIPE, stderr=subprocess.DEVNULL, text=True, timeout=10)
+    except (subprocess.TimeoutExpired, OSError):
+        return True
+    if proc.returncode != 0:
+        return True
+    return bool(proc.stdout.strip())
+
+
+# Named flags differ per tool: overlay.py and graphics.py take a font FILE, caption.py takes a
+# directory of faces plus the family name -- naming a flag the tool does not have is worse than
+# naming none, so the hint says both (review 10).
+FONT_FLAG_HINT = "pass a font file (--font-file on overlay.py/graphics.py, --fonts-dir with --font on caption.py)"
+
+
+FONT_INSTALL_HINT = ("install fonts-noto-cjk / fonts-noto-core (apt), "
+                     "brew install --cask font-noto-sans-cjk / font-noto-sans-arabic (mac), or "
+                     + FONT_FLAG_HINT)
+
+
+def fonts_dir_covers_script(fonts_dir: str, script: str) -> "Optional[bool]":
+    """Does any font under `fonts_dir` cover `script`? None when it cannot be checked.
+
+    `--fonts-dir` says "also look here", not "this exact face", so it must not switch the
+    coverage guarantee off. fontconfig's `fc-scan` reads the files directly (no cache, no
+    installed-font database), which is exactly the question: `%{lang}` lists the languages each
+    face claims.
+    """
+    if script not in FC_LANG or not fonts_dir or not os.path.isdir(fonts_dir):
+        return None
+    exe = shutil.which("fc-scan")
+    if not exe:
+        return None
+    try:
+        proc = subprocess.run([exe, "--format", "%{lang}\n", fonts_dir],
+                              stdout=subprocess.PIPE, stderr=subprocess.DEVNULL, text=True, timeout=10)
+    except (subprocess.TimeoutExpired, OSError):
+        return None
+    if proc.returncode != 0:
+        return None
+    want = FC_LANG[script].lower()
+    for line in proc.stdout.splitlines():
+        if want in [tag.strip().lower() for tag in line.split("|")]:
+            return True
+    return False
+
+
+def script_font_for_text(text: str, *, lang: "Optional[str]" = None, font: "Optional[str]" = None,
+                         font_explicit: bool = False, font_file: "Optional[str]" = None,
+                         fonts_dir: "Optional[str]" = None
+                         ) -> "Tuple[str, Optional[str], Optional[str]]":
+    """(script, font file, family) to draw `text` with, resolving by script when nothing explicit
+    was asked for.
+
+    Returns (script, None, None) when the caller's own choice stands: Latin text, an explicit
+    --font-file, an explicit --font (which is kept even when fontconfig says it does not cover
+    the script -- with one info line saying so, because overriding a user's stated font silently
+    is worse than a warning), or a --fonts-dir that does carry the script. Otherwise the resolved
+    file is returned with ONE info line naming it.
+
+    A script fontconfig says nothing covers is a failed job (tofu is not a delivery). A machine
+    with no working fontconfig at all answers "unknown", not "missing": the job continues with
+    the caller's font -- libass and drawtext still have their own font backends -- and one info
+    line says the coverage could not be verified.
+    """
+    script = detect_script(text or "", lang)
+    if script == "latin":
+        return script, None, None
+    if font_file:
+        return script, None, None
+    if font_explicit and font:
+        if not font_covers_script(font, script):
+            info(f"font: '{font}' does not cover {LANGUAGE_NAMES[script]} text on this machine; keeping it as asked "
+                 f"(drop --font, or {FONT_FLAG_HINT}, to pick one by script automatically)")
+        return script, None, None
+    if fonts_dir:
+        covered = fonts_dir_covers_script(fonts_dir, script)
+        if covered:
+            return script, None, None
+        if covered is None:
+            info(f"font: could not verify that {fonts_dir} covers {LANGUAGE_NAMES[script]} text "
+                 "(no fc-scan on this machine); using it as given")
+            return script, None, None
+        info(f"font: no face in {fonts_dir} covers {LANGUAGE_NAMES[script]} text; "
+             "picking one by script instead (the directory is still searched first)")
+    entry = _script_font_entry(script)
+    if entry is FC_UNKNOWN:
+        info(f"font: could not verify that this machine can render {LANGUAGE_NAMES[script]} text "
+             "(no working fontconfig); rendering with the font as given -- "
+             "doctor --json .fonts.scripts reports what is known")
+        return script, None, None
+    if not entry:
+        die(f"no installed font covers {LANGUAGE_NAMES[script]} text on this machine — {FONT_INSTALL_HINT}", kind="input")
+    info(f"font: {entry[0]} (covers {script})")
+    return script, entry[0], entry[1]
+
+
+def escape_drawtext(text: str) -> str:
+    """Escape a FONT NAME for a single-quoted drawtext option value (`font='<this>'`).
+
+    Since 1.15 this is no longer the route for drawn TEXT -- use drawtext_text_opts(), which puts
+    the text in a file and keeps `\'` and `%` verbatim. It remains the escape for the font-name
+    fallback, where the value is a family name that never legitimately contains a quote or a
+    percent sign.
+
+    Every ffmpeg filter-graph special character (`\\ : % , [ ] ;`) needs a backslash escape
+    regardless of the surrounding quotes -- the graph parser still splits on an unescaped `,`/`;`
+    or ends an option list on an unescaped `:`/`[`/`]` even while "inside" a quoted value. The
+    quote character itself has no reliable backslash escape at all: `\\'` and the POSIX shell
+    close-insert-reopen trick both parse fine in a simple `-vf` chain but silently corrupt a
+    `-filter_complex` chain that uses explicit `[label]` pads (confirmed by rendering the result:
+    trailing option names leak into the picture as literal text). `%` has the same problem as far
+    as drawtext's own expansion scanner is concerned. Both are therefore dropped here rather than
+    escaped -- which is exactly why drawn text no longer comes through this function.
+    """
+    text = re.sub(r"[\x00-\x1f\x7f]", "", text)
+    return (
+        text.replace("'", "")
+        .replace("%", "")
+        .replace("\\", "\\\\")
+        .replace(":", "\\:")
+        .replace(",", "\\,")
+        .replace("[", "\\[")
+        .replace("]", "\\]")
+        .replace(";", "\\;")
+    )
+
+
+def drawtext_text_opts(text: str, tmpdir: "Optional[str]" = None) -> str:
+    """`textfile=<path>:expansion=none` for drawtext -- the one route that is provably safe for
+    every character on every build shape this repo uses.
+
+    The filter-graph parser never sees the text at all: only the PATH is parsed, and
+    escape_filter_path() already handles that. `expansion=none` switches off drawtext's own
+    `%{...}` scanner, which is the reason `%` was unsafe (a bare `\%` logs "Stray %" on one build
+    and fails the whole filter chain on another). With the scanner off, `'`, `%`, `:`, `,`, `[`,
+    `]`, `;` and `\` all reach the picture verbatim -- 1.15 fixes `overlay.py --text "it's 100%
+    done"` losing both characters. Control characters are still stripped: a one-line burnt-in
+    label has no use for them.
+
+    The file is UTF-8, mode 0600, in a private per-run directory (see _drawtext_tmpdir) that is
+    removed when the process ends. It is *registered* here and written by run() only if the
+    command about to run actually names it, so --dry-run and the ASS route write nothing; a
+    printed plan therefore names a path that no longer exists once the run is over, which is the
+    same promise every other temp file in this skill makes.
+    """
+    cleaned = re.sub(r"[\x00-\x1f\x7f]", "", text or "")
+    import hashlib
+    name = "t_" + hashlib.sha256(cleaned.encode("utf-8")).hexdigest()[:16] + ".txt"
+    if tmpdir is None:
+        tmpdir = _drawtext_tmpdir(create=not STATE.dry_run)
+    path = os.path.join(tmpdir, name)
+    _DRAWTEXT_PENDING[path] = cleaned
+    return f"textfile={escape_filter_path(path)}:expansion=none"
+
+
+# --------------------------------------------------------------- caption line breaking (1.16)
+# Lifted out of caption.py in 1.16.0 so graphics.py can wrap the same way (caption.py keeps the
+# names it exported, re-imported from here). The whole breaker is pure: a string in, a list of
+# lines out, no subprocess and no probe, which is what makes the eval regression corpus cheap
+# to lock down in unit tests.
+
+# How much of the frame width a caption line may use. libass's own default SRT margins are 10 of a
+# 384-wide script (2.6 % a side); 5 % a side is the safe area every platform check in this repo uses.
+SAFE_WIDTH_FRACTION = 0.9
+# ORPHAN_MIN_EM: one full-width CJK/Thai character plus a hair. A last line narrower than this is a
+# single stranded character -- eval 14's th1 (a lone 'ล') and dl3 (a lone '行').
+ORPHAN_MIN_EM = 1.1
+
+WRAP_MODES = ("phrase", "measured")
+
+# R3's Japanese preference table. These are *preferences* applied only among positions that
+# already fit the line, so the table can never make a line too wide or change the line count.
+#
+# JA_PARTICLES is a "do not strand at the start of a line" table, which is the direction kinsoku
+# practice actually goes: a particle is enclitic -- it attaches to the word BEFORE it and marks
+# that word's role -- so a line beginning with は or が reads as a fragment torn off its phrase.
+# A break AFTER a particle is therefore preferred (the particle stays with what it marks) and a
+# break BEFORE one is forbidden. The list is the eight case/topic particles named in the 1.16.0
+# task brief (は が を に で と の へ) plus も や から まで より, which a reader of Japanese would
+# add for the same reason. It is a judgement call with no upstream source; treat it as tunable
+# data, not as grammar.
+JA_PARTICLES = "はがをにでとのへもや"              # a break AFTER one of these is preferred, BEFORE one forbidden
+# The multi-character members of the same table. They are matched as whole strings against the
+# text on each side of a candidate break -- putting them in the character string above turned
+# か, ら, ま, で, よ and り into one-character particles of their own, which none of them is.
+JA_PARTICLE_WORDS = ("から", "まで", "より")
+JA_SENTENCE_END = "。、！？」』）"                  # a break AFTER one of these is preferred
+# Characters that may never start a line: small kana, the prolonged sound mark, closing brackets
+# and the Japanese punctuation that hangs on the end of the line before it.
+JA_NO_LINE_START = "ぁぃぅぇぉっゃゅょァィゥェォッャュョーヽヾゝゞ、。！？）」』】〕》’”％"
+JA_NO_LINE_END = "（「『【〔《‘“"                   # ... and the ones that may never end a line
+
+# R4. Function words belong to the phrase that FOLLOWS them: an article or preposition begins the
+# noun phrase it governs, so a break before one is the good break (the word opens the next line
+# with its phrase) and a break after one is the bad break (it is stranded at the end of a line,
+# away from what it governs). Both directions are scored, which is what makes the rule decide
+# rather than merely veto. Frozen data, matched case-folded on the atom with its punctuation
+# stripped; six languages because those are the Latin-script languages the eval corpus covers. A
+# word in several sets means the same thing structurally in each, so the union is used when no
+# --lang was given.
+FUNCTION_WORDS = {
+    "en": {"a", "an", "the", "of", "to", "in", "on", "at", "for", "with", "by", "from", "and",
+           "or", "as", "is", "it", "its", "this", "that", "into", "than", "but", "so"},
+    "es": {"el", "la", "los", "las", "un", "una", "unos", "unas", "de", "del", "al", "en", "con",
+           "por", "para", "y", "o", "que", "su", "sus", "lo", "se", "es"},
+    "pt": {"o", "a", "os", "as", "um", "uma", "de", "do", "da", "dos", "das", "em", "no", "na",
+           "nos", "nas", "com", "por", "para", "e", "que", "se", "ao", "aos"},
+    "fr": {"le", "la", "les", "un", "une", "de", "du", "des", "à", "au", "aux", "en", "dans",
+           "et", "ou", "que", "qui", "ce", "ces", "son", "sa", "ses", "par", "pour", "avec", "sur"},
+    "de": {"der", "die", "das", "ein", "eine", "einen", "einem", "einer", "den", "dem", "des",
+           "zu", "in", "im", "auf", "mit", "und", "oder", "von", "vom", "für", "aus", "an"},
+    "it": {"il", "lo", "la", "i", "gli", "le", "un", "una", "uno", "di", "del", "della", "da",
+           "in", "nel", "con", "per", "e", "che", "su", "al", "ai"},
+}
+_FUNCTION_WORDS_ANY = frozenset().union(*FUNCTION_WORDS.values())
+
+# Penalty scores. Only the ordering matters; 1.0 means "never choose this if anything else fits".
+PENALTY_FORBIDDEN = 1.0
+PENALTY_OKURIGANA = 0.9       # between a kanji stem and the hiragana that inflects it
+PENALTY_FUNCTION_WORD = 0.8   # R4: the line before the break ends in an article/preposition
+PENALTY_IDEOGRAPHS = 0.6      # between two kanji: no evidence either way, mildly discouraged
+PENALTY_NEUTRAL = 0.5         # between two content words, or two characters with nothing to say
+PENALTY_FUNCTION_WORD_START = 0.2  # R4: the next line opens with the article/preposition it governs
+PENALTY_PARTICLE = 0.2        # R3: after a particle, so the particle stays with the word it marks
+PENALTY_SENTENCE_END = 0.0    # R3: after 。、！？ -- the one break a reader expects
+
+_HYPHENS = ("-", "‐")    # ‑ (non-breaking hyphen) is deliberately NOT here
+
+
+def _atoms(line: str) -> "List[Tuple[str, bool]]":
+    """Break a line into the smallest pieces a wrap may separate -- one atom per CJK/Thai
+    character, one per emoji cluster, one per whitespace-delimited word otherwise -- each with
+    whether a space stood before it in the original. The flag is what puts the text back together
+    exactly as written: "Hello 世界" keeps its space, "世界です" gains none."""
+    out: "List[Tuple[str, bool]]" = []
+    word = ""
+    spaced = False        # a space stands before the atom being built
+    pending = False       # a space stands before the NEXT atom
+    attach_next = False   # a leading Thai/Lao vowel is waiting for its base consonant
+    # An emoji cluster is one atom: a wrap must never land inside a ZWJ sequence, a flag pair or
+    # between a base and its skin-tone modifier (the same rule combining marks already follow).
+    clusters = {i: len(cl) for i, cl in emoji_clusters(line)}
+    i = 0
+    while i < len(line):
+        ch = line[i]
+        if i in clusters:
+            cluster = line[i:i + clusters[i]]
+            if word:
+                out.append((word, spaced))
+                word = ""
+            out.append((cluster, pending))
+            pending = False
+            attach_next = False
+            i += clusters[i]
+            continue
+        i += 1
+        if char_script(ch) in NO_SPACE_SCRIPTS:
+            if word:
+                out.append((word, spaced))
+                word = ""
+            if out and (attach_next or _is_mark(ch)):
+                # never break between a base and the mark (or the leading vowel) that belongs to
+                # it: the line would start with an orphaned tone mark or vowel sign
+                out[-1] = (out[-1][0] + ch, out[-1][1])
+            else:
+                out.append((ch, pending))
+                pending = False
+            attach_next = ord(ch) in LEADING_VOWELS
+        elif ch.isspace():
+            if word:
+                out.append((word, spaced))
+                word = ""
+            pending = True
+        else:
+            if not word:
+                spaced, pending = pending, False
+            word += ch
+    if word:
+        out.append((word, spaced))
+    return out
+
+
+def _split_hyphens(atoms: "List[Tuple[str, bool]]") -> "List[Tuple[str, bool]]":
+    """R1's one addition to the atom list: a hyphenated token may break *after* its hyphen.
+
+    "end-to-end" becomes `end-` / `to-` / `end`, each piece carrying the space flag of the token
+    it came from for the first piece and False for the rest, so _join() puts it back with no space
+    at all. A hyphen that is the first or last character of the token (`-5`, `well-`) is never a
+    break point: the guard is that both sides must be non-empty."""
+    out: "List[Tuple[str, bool]]" = []
+    for atom, spaced in atoms:
+        if len(atom) < 3 or not any(h in atom[1:-1] for h in _HYPHENS):
+            out.append((atom, spaced))
+            continue
+        piece = ""
+        first = True
+        for i, ch in enumerate(atom):
+            piece += ch
+            if ch in _HYPHENS and 0 < i < len(atom) - 1:
+                out.append((piece, spaced if first else False))
+                piece = ""
+                first = False
+        if piece:
+            out.append((piece, spaced if first else False))
+    return out
+
+
+def _join(left: str, atom: str, spaced: bool) -> str:
+    """Put an atom back on a line, restoring the space that stood before it."""
+    if not left:
+        return atom
+    return left + (" " if spaced else "") + atom
+
+
+def _break_spaced(first: str, second: str) -> bool:
+    """Did a space stand at the break between these two wrapped lines? Only spaced scripts put one
+    there -- a CJK/Thai break sits between two characters that were written with nothing between
+    them, and re-joining them with a space would insert a character the cue never had."""
+    if not first or not second:
+        return False
+    return char_script(first[-1]) not in NO_SPACE_SCRIPTS and char_script(second[0]) not in NO_SPACE_SCRIPTS \
+        and char_script(first[-1]) != "emoji" and char_script(second[0]) != "emoji"
+
+
+def _is_kana(ch: str) -> bool:
+    return 0x3040 <= ord(ch) <= 0x30FF
+
+
+def _is_hiragana(ch: str) -> bool:
+    return 0x3040 <= ord(ch) <= 0x309F
+
+
+def _is_ideograph(ch: str) -> bool:
+    cp = ord(ch)
+    return 0x3400 <= cp <= 0x4DBF or 0x4E00 <= cp <= 0x9FFF or 0xF900 <= cp <= 0xFAFF
+
+
+def _is_weak_line(line: str) -> "bool":
+    """A line no reader should be given on its own (R2).
+
+    1.15 asked only "is the last line one atom narrower than ORPHAN_MIN_EM", which a full-width
+    character passes: dl3 still showed a lone `2` and a stranded `行`. Three cases instead, any of
+    which makes a line too thin to stand alone:
+      - a single character narrower than ORPHAN_MIN_EM (1.15's rule, kept);
+      - nothing but digits, punctuation and symbols, at most two characters ("2", "--");
+      - a single kana, whatever its width -- a kana is a full em and passes the width test, but a
+        line holding one is a syllable, not a word.
+    """
+    stripped = (line or "").strip()
+    if not stripped:
+        return True
+    if len(stripped) == 1 and text_width_em(stripped) < ORPHAN_MIN_EM:
+        return True
+    if len(stripped) <= 2 and all(unicodedata.category(c)[0] in "NPS" for c in stripped):
+        return True
+    if len(stripped) == 1 and char_script(stripped) == "ja" and _is_kana(stripped):
+        return True
+    return False
+
+
+def _function_words(lang: "Optional[str]") -> "frozenset":
+    """R4's table for this language. An unknown or absent language gets the union of the six sets:
+    a token that appears in several of them is the same kind of word in each, which is why the
+    rule is a penalty and not a refusal."""
+    key = (lang or "").strip().lower().split("-")[0]
+    if key in FUNCTION_WORDS:
+        return frozenset(FUNCTION_WORDS[key])
+    return _FUNCTION_WORDS_ANY
+
+
+def _bare_word(atom: str) -> str:
+    return "".join(c for c in (atom or "") if c.isalpha() or c == "'").strip("'").lower()
+
+
+def _particle_starts(text: str) -> bool:
+    """Does `text` begin with a particle -- one character, or one of the two-character ones?"""
+    if not text:
+        return False
+    return text[0] in JA_PARTICLES or text.startswith(JA_PARTICLE_WORDS)
+
+
+def _particle_ends(text: str) -> bool:
+    """Does `text` end with a particle? `から` counts, a bare `ら` does not."""
+    if not text:
+        return False
+    return text[-1] in JA_PARTICLES or text.endswith(JA_PARTICLE_WORDS)
+
+
+def break_penalty(prev_char: str, next_char: str, lang: "Optional[str]" = None,
+                  before: str = "", after: str = "") -> float:
+    """How bad a break between these two characters is, 0.0 (preferred) to 1.0 (forbidden).
+
+    Only consulted among break positions that already fit `max_em`, so a preference can never
+    widen a line or change the line count. Japanese gets the particle half of the table -- a break
+    AFTER a particle is preferred and a break BEFORE one forbidden, because a particle attaches to
+    the word before it; Chinese gets only the sentence-end and forbidden halves, because particles
+    are Japanese grammar.
+
+    `before`/`after` are the text on each side of the break when the caller has it, which is what
+    lets the two-character particles (から/まで/より) be matched as words. Without them only the
+    single-character table applies."""
+    if not prev_char or not next_char:
+        return PENALTY_NEUTRAL
+    script = (lang or "").strip().lower().split("-")[0]
+    if script not in ("ja", "zh"):
+        # A kana on either side settles it: only Japanese has them, and char_script() reads a bare
+        # Han character as Chinese, which used to switch the particle rules off for exactly the
+        # break they exist to judge (`...が|決まる` -- kana before, kanji after).
+        if _is_kana(prev_char) or _is_kana(next_char):
+            script = "ja"
+        else:
+            script = char_script(next_char)
+            if script not in ("ja", "zh"):
+                script = char_script(prev_char)
+    if next_char in JA_NO_LINE_START or prev_char in JA_NO_LINE_END or _is_mark(next_char):
+        return PENALTY_FORBIDDEN
+    if script not in ("ja", "zh"):
+        return PENALTY_NEUTRAL
+    if prev_char in JA_SENTENCE_END:
+        return PENALTY_SENTENCE_END
+    if script == "ja" and _particle_starts(after or next_char):
+        # a particle may not open a line: it belongs to the word before it (kinsoku)
+        return PENALTY_FORBIDDEN
+    if script == "ja" and _particle_ends(before or prev_char):
+        return PENALTY_PARTICLE
+    if script == "ja" and _is_ideograph(prev_char) and _is_hiragana(next_char):
+        # okurigana: 決|まる is inside a word even though neither half is a "word" on its own
+        return PENALTY_OKURIGANA
+    if _is_ideograph(prev_char) and _is_ideograph(next_char):
+        return PENALTY_IDEOGRAPHS
+    return PENALTY_NEUTRAL
+
+
+def _cut_penalty(atoms: "Sequence[Tuple[str, bool]]", cut: int, lang: "Optional[str]") -> float:
+    """The penalty of breaking `atoms` before index `cut`."""
+    prev_atom = atoms[cut - 1][0]
+    next_atom = atoms[cut][0]
+    if not prev_atom or not next_atom:
+        return PENALTY_NEUTRAL
+    if atoms[cut][1]:
+        # a space stood here: a spaced script, so R4 is the rule that applies, in both directions
+        if all(not ch.isalnum() for ch in next_atom):
+            return PENALTY_FORBIDDEN   # never strand punctuation at the start of a line
+        words = _function_words(lang)
+        if _bare_word(prev_atom) in words:
+            return PENALTY_FUNCTION_WORD        # stranded at the end of a line, away from its noun
+        if _bare_word(next_atom) in words:
+            return PENALTY_FUNCTION_WORD_START  # opens the next line with the phrase it governs
+        return PENALTY_NEUTRAL
+    if prev_atom.endswith(_HYPHENS):
+        return PENALTY_NEUTRAL         # R1: a hyphen is a legitimate break point
+    # the text on each side, so a two-character particle (から/まで/より) is seen as one
+    before = "".join(a for a, _sp in atoms[:cut])
+    after = "".join(a for a, _sp in atoms[cut:])
+    return break_penalty(prev_atom[-1], next_atom[0], lang, before=before, after=after)
+
+
+def best_break(atoms: "Sequence[Tuple[str, bool]]", max_em: float,
+               lang: "Optional[str]" = None) -> "Optional[int]":
+    """The index to break `atoms` at so they become two lines, or None when none fits.
+
+    Among every position whose two halves both fit `max_em`, the one minimising
+    (penalty, widest line, |width difference|) wins: R1-R4 choose first, and 1.15's
+    minimise-the-widest-line rule breaks the ties it used to decide alone."""
+    best = None
+    for cut in range(1, len(atoms)):
+        a = b = ""
+        for atom, sp in atoms[:cut]:
+            a = _join(a, atom, sp)
+        for atom, sp in atoms[cut:]:
+            b = _join(b, atom, sp)
+        wa, wb = text_width_em(a), text_width_em(b)
+        if max(wa, wb) > max_em:
+            continue
+        if _is_weak_line(a) or _is_weak_line(b):
+            continue
+        key = (_cut_penalty(atoms, cut, lang), max(wa, wb), abs(wa - wb))
+        if best is None or key < best[0]:
+            best = (key, cut)
+    return None if best is None else best[1]
+
+
+def _fix_orphans(lines: "List[str]", max_em: float) -> "List[str]":
+    """No last line that is a single stranded atom.
+
+    Greedy wrapping leaves one character alone whenever the line before it filled exactly: eval 14
+    produced a Thai cue ending in a lone `ล` and a Japanese one ending in a lone `行`. While the
+    last line is one atom narrower than ORPHAN_MIN_EM, the last atom of the line above moves down
+    onto it -- but only while the result still fits and the line above does not become an orphan
+    itself, so a two-word cue is never made worse."""
+    lines = list(lines)
+    for _ in range(len(lines)):
+        if len(lines) < 2:
+            break
+        tail = _atoms(lines[-1])
+        if len(tail) != 1 or text_width_em(lines[-1]) >= ORPHAN_MIN_EM:
+            break
+        prev = _atoms(lines[-2])
+        if len(prev) < 2:
+            break
+        moved, spaced = prev[-1]
+        new_prev = ""
+        for atom, sp in prev[:-1]:
+            new_prev = _join(new_prev, atom, sp)
+        new_last = _join(moved, tail[0][0], _break_spaced(lines[-2], lines[-1]))
+        if text_width_em(new_last) > max_em or text_width_em(new_prev) < ORPHAN_MIN_EM:
+            break
+        lines[-2], lines[-1] = new_prev, new_last
+    return lines
+
+
+def _fix_weak_lines(lines: "List[str]", max_em: float) -> "List[str]":
+    """R2, generalised: _fix_orphans run at *every* boundary, against _is_weak_line.
+
+    1.15 only ever looked at the last line, so a stranded digit or kana in the middle of a
+    three-line cue survived. Walking upward from the last line, while a line is weak the last atom
+    of the line above moves down onto it -- with 1.15's two guards intact (the result must still
+    fit, and the line above must not itself become weak), so the line count never changes."""
+    lines = list(lines)
+    for i in range(len(lines) - 1, 0, -1):
+        for _ in range(len(lines)):
+            if not _is_weak_line(lines[i]):
+                break
+            prev = _atoms(lines[i - 1])
+            if len(prev) < 2:
+                break
+            moved, _spaced = prev[-1]
+            new_prev = ""
+            for atom, sp in prev[:-1]:
+                new_prev = _join(new_prev, atom, sp)
+            new_last = _join(moved, lines[i], _break_spaced(lines[i - 1], lines[i]))
+            if text_width_em(new_last) > max_em or _is_weak_line(new_prev) \
+                    or text_width_em(new_prev) < ORPHAN_MIN_EM:
+                break
+            lines[i - 1], lines[i] = new_prev, new_last
+    return lines
+
+
+def _rebalance(lines: "List[str]", max_em: float) -> "List[str]":
+    """Move each break to the one that minimises the widest line of the pair, without changing the
+    line count.
+
+    Greedy wrapping fills line 1 to the brim and leaves line 2 short, which is what split eval 14's
+    `"A third line the tool times for me"` mid-phrase. Only spaced scripts are rebalanced: a
+    non-spaced script has no phrase structure in its atom list, so moving the break there only
+    moves the ragged edge. A break is never placed before a punctuation-only atom."""
+    if len(lines) < 2:
+        return lines
+    out = list(lines)
+    for i in range(len(out) - 1):
+        first, second = out[i], out[i + 1]
+        tail_atoms = _atoms(second)
+        if tail_atoms:
+            tail_atoms[0] = (tail_atoms[0][0], _break_spaced(first, second))
+        atoms = _atoms(first) + tail_atoms
+        if not atoms or any(char_script(ch) in NO_SPACE_SCRIPTS for ch in first + second):
+            continue
+        best = None
+        for cut in range(1, len(atoms)):
+            if not atoms[cut][1]:
+                continue  # only break where a space stood
+            if all(not ch.isalnum() for ch in atoms[cut][0]):
+                continue  # never strand punctuation at the start of a line
+            a = b = ""
+            for atom, sp in atoms[:cut]:
+                a = _join(a, atom, sp)
+            for atom, sp in atoms[cut:]:
+                b = _join(b, atom, sp)
+            wa, wb = text_width_em(a), text_width_em(b)
+            if max(wa, wb) > max_em:
+                continue
+            key = (max(wa, wb), abs(wa - wb))
+            if best is None or key < best[0]:
+                best = (key, a, b)
+        if best is not None:
+            out[i], out[i + 1] = best[1], best[2]
+    return out
+
+
+def _rebalance_phrase(lines: "List[str]", max_em: float, lang: "Optional[str]") -> "Tuple[List[str], int]":
+    """_rebalance with R1-R4 deciding, for every script rather than spaced ones only.
+
+    Returns the new lines and how many breaks a phrase rule moved away from the position 1.15's
+    widest-line rule alone would have chosen -- the `phrase_breaks` count in the result."""
+    if len(lines) < 2:
+        return list(lines), 0
+    out = list(lines)
+    moved = 0
+    for i in range(len(out) - 1):
+        first, second = out[i], out[i + 1]
+        tail_atoms = _atoms(second)
+        if tail_atoms:
+            tail_atoms[0] = (tail_atoms[0][0], _break_spaced(first, second))
+        atoms = _split_hyphens(_atoms(first) + tail_atoms)
+        if len(atoms) < 2:
+            continue
+        cut = best_break(atoms, max_em, lang)
+        if cut is None:
+            continue
+        a = b = ""
+        for atom, sp in atoms[:cut]:
+            a = _join(a, atom, sp)
+        for atom, sp in atoms[cut:]:
+            b = _join(b, atom, sp)
+        if (a, b) != (first, second):
+            moved += 1
+        out[i], out[i + 1] = a, b
+    return out, moved
+
+
+def _greedy_chunks(raw: str, max_em: float) -> "List[str]":
+    """The greedy fill on its own: the line count every mode must keep."""
+    current = ""
+    chunk: "List[str]" = []
+    for atom, spaced in _atoms(raw):
+        candidate = _join(current, atom, spaced)
+        if current and text_width_em(candidate) > max_em:
+            chunk.append(current)
+            current = atom
+        else:
+            current = candidate
+    if current:
+        chunk.append(current)
+    return chunk
+
+
+def _balance(chunk: "List[str]", max_em: float, mode: str, lang: "Optional[str]") -> "List[str]":
+    """The post-passes for one greedy chunk, in the mode's own order. Never changes the count:
+    a pass that would is discarded, exactly as 1.15 did."""
+    if len(chunk) < 2:
+        return chunk
+    if mode == "measured":
+        fixed = _fix_orphans(chunk, max_em)
+        rebalanced = _rebalance(fixed, max_em)
+    else:
+        fixed = _fix_weak_lines(_fix_orphans(chunk, max_em), max_em)
+        rebalanced, _moved = _rebalance_phrase(fixed, max_em, lang)
+        rebalanced = _fix_weak_lines(rebalanced, max_em)
+    if len(rebalanced) == len(chunk):
+        return rebalanced
+    return fixed if len(fixed) == len(chunk) else chunk
+
+
+def wrap_text(text: str, max_em: float, *, balance: bool = True, mode: str = "phrase",
+              lang: "Optional[str]" = None) -> "List[str]":
+    """Wrap `text` to lines no wider than `max_em` em, keeping the manual breaks it already has.
+
+    An atom wider than the whole line (one very long word) is left alone on its line rather than
+    cut mid-word: an over-long line is readable, a chopped word is not.
+
+    `mode="phrase"` (the default since 1.16) then applies the four phrase rules -- never inside a
+    word or across a hyphen's wrong side (R1), no line that is a lone digit, punctuation or kana
+    (R2), Japanese/Chinese breaks preferred at sentence ends and after particles, never before one
+    and never inside a word (R3), and an article or preposition kept with the phrase it governs by
+    preferring the break before it and avoiding the break after it (R4). `mode="measured"` is
+    1.15's behaviour exactly: no one-character orphan line, and a break chosen only to minimise the
+    widest line. Neither mode ever changes the number of lines the greedy fill produced.
+    """
+    lines: "List[str]" = []
+    for raw in text.split("\n"):
+        if not raw.strip():
+            continue
+        chunk = _greedy_chunks(raw, max_em)
+        lines.extend(_balance(chunk, max_em, mode, lang) if balance else chunk)
+    return lines or [text]
+def wrap_variants(text: str, max_em: float, *, mode: str = "phrase",
+                  lang: "Optional[str]" = None) -> "Tuple[List[str], List[str], List[str]]":
+    """`(wrapped, greedy, measured)` for one cue from a single greedy fill.
+
+    layout_cues needs all three -- `wrapped` is what is burnt in, `greedy` is what `rebalanced`
+    counts against and `measured` what `phrase_breaks` counts against -- and used to call
+    wrap_text() three times, re-running the atomiser and the greedy fill each time. The fill is
+    the same for every mode, so it is done once here and only the post-passes are repeated.
+    `measured` is the same list object as `wrapped` when that is already the mode.
+    """
+    wrapped: "List[str]" = []
+    greedy: "List[str]" = []
+    measured: "List[str]" = []
+    for raw in text.split("\n"):
+        if not raw.strip():
+            continue
+        chunk = _greedy_chunks(raw, max_em)
+        greedy.extend(chunk)
+        wrapped.extend(_balance(list(chunk), max_em, mode, lang))
+        measured.extend(chunk if mode == "measured" else _balance(list(chunk), max_em, "measured", None))
+    if not greedy:
+        greedy = [text]
+    return (wrapped or [text], greedy, measured or [text])