skilldrop-cli 1.6.0 → 1.7.0

package/packs/experience-design/skills/locale-check/scripts/locale_check.py

@@ -0,0 +1,1105 @@
+#!/usr/bin/env python3
+"""Check translated string files against the source strings before they ship. Stdlib only.
+
+Reads the source file and one or more target files in any of these formats, detected by
+extension or content:
+  JSON (flat or nested; i18next plurals as key_one / key_other ...), ARB, gettext .po/.pot,
+  Android strings.xml (string, plurals, string-array), iOS .strings (UTF-8 or UTF-16) and
+  iOS .stringsdict.
+The locale of each file comes from its path (de.json, fr-CA.json, values-fr/strings.xml,
+fr.lproj/Localizable.strings, fr.po, app_fr.arb, locales/fr/common.json), from an ARB
+@@locale or a .po Language header, or from --locale.
+
+It FAILS a target where
+  - a source key is missing, or its translation is empty
+  - a placeholder differs from the source: {name}, {{name}}, %s, %d, %1$s, %@, %(name)s, and
+    ICU arguments; a positional printf argument whose type changes, or plain %s / %d
+    arguments in a different order (they would swap at runtime)
+  - an ICU message that parses in the source doesn't parse in the target, or a select
+    keyword (male / female / other) was translated or dropped
+  - a plural form the target language needs is missing (CLDR cardinal categories: ar needs
+    zero, one, two, few, many and other; pl one, few, many and other; ja only other); for
+    gettext, the number of msgstr[n] forms differs from the language's nplurals
+  - markup tags (<b>, <a href>, <0>...</0>, <xliff:g>) differ from the source or don't balance
+  - a string is over its character budget (--budget)
+  - a target file doesn't parse at all
+and WARNS where
+  - a key in the target is not in the source
+  - a translation is identical to the source (possibly untranslated), for text over 3 letters
+  - leading or trailing whitespace changed, or the source's ending punctuation was dropped
+  - a word mixes Latin, Cyrillic or Greek letters (a look-alike typed by mistake)
+  - a plural form the language uses only for large numbers (fr, es, it, pt "many") is absent
+  - a gettext entry is marked fuzzy (msgfmt leaves it out, so the source text shows)
+
+--budget FILE is JSON: {"keys": {"checkout.pay": 18}, "ratio": 1.4, "ratio_min_source": 10}
+or a CSV with columns key,max. A per-key limit applies to every locale; the ratio applies to
+keys without one, when the source string is at least ratio_min_source characters long.
+Lengths count visible characters: markup tags and placeholders are not counted.
+Without --budget no length is checked, and the report says so.
+
+Usage:
+  python3 locale_check.py en.json de.json fr.json [--budget budget.json] [--json] [-o OUT]
+  python3 locale_check.py values/strings.xml values-pl/strings.xml values-ar/strings.xml
+  python3 locale_check.py messages.pot fr.po --locale fr
+
+Exit codes: 0 no failures, 1 a failure, 2 bad input (a file missing, an unknown format, a
+source that doesn't parse, or bad arguments).
+"""
+import argparse
+import csv
+import io
+import json
+import math
+import os
+import plistlib
+import re
+import sys
+import unicodedata
+import xml.etree.ElementTree as ET
+
+# --------------------------------------------------------------------------- languages
+
+# CLDR cardinal plural categories (CLDR 44) for the languages most products ship.
+CLDR = {}
+for _langs, _cats in (
+    ("ja zh ko vi th id ms km lo my yo ig jv su", ("other",)),
+    ("en de nl sv da nb nn no fi et el hu tr bg af sw ur hi bn fa ta te ml mr gu kn is fil az ka "
+     "kk ky mn sq eu gl ne pa si zu xh am ps", ("one", "other")),
+    ("fr es it pt ca", ("one", "many", "other")),
+    ("ru uk be", ("one", "few", "many", "other")),
+    ("pl", ("one", "few", "many", "other")),
+    ("cs sk", ("one", "few", "many", "other")),
+    ("lt", ("one", "few", "many", "other")),
+    ("hr sr bs", ("one", "few", "other")),
+    ("ro mo", ("one", "few", "other")),
+    ("lv", ("zero", "one", "other")),
+    ("sl", ("one", "two", "few", "other")),
+    ("he iw", ("one", "two", "other")),
+    ("ar", ("zero", "one", "two", "few", "many", "other")),
+    ("cy", ("zero", "one", "two", "few", "many", "other")),
+    ("ga", ("one", "two", "few", "many", "other")),
+    ("mt", ("one", "two", "few", "many", "other")),
+):
+    for _l in _langs.split():
+        CLDR[_l] = _cats
+# Categories a language has that only very large numbers reach: missing is a warning.
+SOFT = {"fr": {"many"}, "es": {"many"}, "it": {"many"}, "pt": {"many"}, "ca": {"many"}}
+# gettext nplurals as the usual Plural-Forms headers set them.
+NPLURALS = {}
+for _langs, _n in (("ja zh ko vi th id ms km lo my", 1),
+                   ("en de nl sv da nb nn no fi et el hu tr bg af sw ur hi bn fa ta te ml mr gu kn "
+                    "is fil fr es it pt ca az ka kk ky mn sq eu gl ne pa si zu he iw", 2),
+                   ("ru uk be pl cs sk lt hr sr bs ro mo lv", 3), ("sl cy", 4), ("ga", 5), ("ar", 6)):
+    for _l in _langs.split():
+        NPLURALS[_l] = _n
+CATEGORIES = ("zero", "one", "two", "few", "many", "other")
+LANG_RE = re.compile(r"^([a-z]{2,3})(?:[-_](?:[A-Z][a-z]{3}|[A-Za-z]{2}|\d{3}|r[A-Z]{2}))*$")
+
+
+def lang_of(code):
+    """'pt-BR' -> 'pt', 'zh_Hant' -> 'zh', 'values-pt-rBR' handled by the caller."""
+    return re.split(r"[-_]", code.strip())[0].lower() if code else ""
+
+
+def detect_locale(path):
+    """The locale a file's path names, or ''."""
+    parts = re.split(r"[\\/]", os.path.abspath(path))
+    name = parts[-1]
+    stem = name.split(".")[0]
+    m = re.match(r"^values-([a-z]{2,3})(?:-r([A-Z]{2}))?", parts[-2] if len(parts) > 1 else "")
+    if m:
+        return m.group(1) + ("-" + m.group(2) if m.group(2) else "")
+    for p in reversed(parts[:-1]):
+        m = re.match(r"^([A-Za-z]{2,3}(?:[-_][A-Za-z]{2,4})?)\.lproj$", p)
+        if m:
+            return m.group(1)
+    if LANG_RE.match(stem) and lang_of(stem) in CLDR:
+        return stem
+    for tok in reversed(re.split(r"[_.]", name)[:-1]):
+        if LANG_RE.match(tok) and lang_of(tok) in CLDR:
+            return tok
+    for p in reversed(parts[-3:-1]):
+        if LANG_RE.match(p) and lang_of(p) in CLDR:
+            return p
+    if len(parts) > 1 and parts[-2] in ("values", "Base.lproj", "en.lproj"):
+        return "en"
+    return ""
+
+
+# --------------------------------------------------------------------------- entries
+
+class Entry(object):
+    """One string, or one plural string as {category: text}."""
+
+    __slots__ = ("key", "value", "line", "plural", "fuzzy", "gettext_n", "icu_ok")
+
+    def __init__(self, key, value, line=None, plural=False, fuzzy=False, gettext_n=None):
+        self.key, self.value, self.line, self.plural = key, value, line, plural
+        self.fuzzy, self.gettext_n, self.icu_ok = fuzzy, gettext_n, None
+
+    def texts(self):
+        return list(self.value.values()) if self.plural else [self.value]
+
+
+class BadInput(Exception):
+    pass
+
+
+class ParseError(Exception):
+    pass
+
+
+def read_text(path):
+    try:
+        raw = open(path, "rb").read()
+    except OSError as e:
+        raise BadInput("can't read %s: %s" % (path, e.strerror or e))
+    for bom, enc in ((b"\xef\xbb\xbf", "utf-8-sig"), (b"\xff\xfe", "utf-16"), (b"\xfe\xff", "utf-16")):
+        if raw.startswith(bom):
+            return raw.decode(enc, errors="replace")
+    if len(raw) > 1 and raw[1:2] == b"\x00" and raw[0:1] != b"\x00":
+        return raw.decode("utf-16-le", errors="replace")
+    return raw.decode("utf-8", errors="replace")
+
+
+def detect_format(path, text):
+    ext = os.path.splitext(path)[1].lower()
+    if ext in (".json", ".arb", ".po", ".pot", ".strings", ".stringsdict", ".xml"):
+        return {"json": "json", "arb": "arb", "po": "po", "pot": "po", "strings": "strings",
+                "stringsdict": "stringsdict", "xml": "android"}[ext[1:]]
+    head = text.lstrip()[:400]
+    if head.startswith("{"):
+        return "arb" if '"@@locale"' in text else "json"
+    if "<resources" in head:
+        return "android"
+    if "<plist" in head or "<!DOCTYPE plist" in head:
+        return "stringsdict"
+    if re.search(r'^\s*msgid\s+"', text, re.M):
+        return "po"
+    if re.search(r'^\s*"(?:[^"\\]|\\.)*"\s*=\s*"', text, re.M):
+        return "strings"
+    raise BadInput("%s: unknown format (use .json, .arb, .po, .xml, .strings or .stringsdict)" % path)
+
+
+def json_key_lines(text):
+    """{dotted key path: line} for every object key in a JSON text, without a full parser."""
+    lines, stack, i, n, line = {}, [], 0, len(text), 1
+    pending, last_str = None, None
+    while i < n:
+        c = text[i]
+        if c == "\n":
+            line += 1
+        elif c == '"':
+            j, buf = i + 1, []
+            while j < n and text[j] != '"':
+                if text[j] == "\\":
+                    j += 1
+                if j < n and text[j] == "\n":
+                    line += 1
+                j += 1
+            last_str = (text[i + 1:j], line)
+            i = j
+        elif c == ":" and last_str is not None:
+            pending = last_str
+            key = ".".join([k for k in stack if k is not None] + [pending[0]])
+            lines.setdefault(key, pending[1])
+            last_str = None
+        elif c in "{[":
+            stack.append(pending[0] if pending else None)
+            pending, last_str = None, None
+        elif c in "}]":
+            if stack:
+                stack.pop()
+            pending, last_str = None, None
+        elif c == ",":
+            pending, last_str = None, None
+        i += 1
+    return lines
+
+
+def flatten(obj, prefix=""):
+    if isinstance(obj, dict):
+        for k, v in obj.items():
+            yield from flatten(v, prefix + ("." if prefix else "") + str(k))
+    elif isinstance(obj, list):
+        for i, v in enumerate(obj):
+            yield from flatten(v, prefix + "." + str(i))
+    elif isinstance(obj, str):
+        yield prefix, obj
+    elif obj is None or isinstance(obj, (int, float, bool)):
+        yield prefix, "" if obj is None else str(obj)
+
+
+PLURAL_SUFFIX = re.compile(r"^(.*)_(zero|one|two|few|many|other)$")
+
+
+def parse_json(text, arb=False):
+    try:
+        doc = json.loads(text)
+    except ValueError as e:
+        raise ParseError("not valid JSON: %s" % e)
+    except RecursionError:
+        raise ParseError("nested too deeply to read")
+    if not isinstance(doc, dict):
+        raise ParseError("the top level is not an object of strings")
+    locale = doc.get("@@locale") if arb and isinstance(doc.get("@@locale"), str) else ""
+    lines = json_key_lines(text)
+    try:
+        flat = [(k, v) for k, v in flatten(doc) if not (arb and k.split(".")[0].startswith("@"))]
+    except RecursionError:
+        raise ParseError("nested too deeply to read")
+    keys = {k for k, _ in flat}
+    entries, grouped = {}, {}
+    for k, v in flat:
+        m = PLURAL_SUFFIX.match(k)
+        if m and (m.group(1) + "_other") in keys:
+            grouped.setdefault(m.group(1), {})[m.group(2)] = (v, lines.get(k))
+            continue
+        entries[k] = Entry(k, v, lines.get(k))
+    for base, forms in grouped.items():
+        line = min((ln for _, ln in forms.values() if ln), default=None)
+        if base in entries:  # a bare key beside its plural forms is the fallback form
+            forms.setdefault("other", (entries.pop(base).value, line))
+        entries[base] = Entry(base, {c: t for c, (t, _) in forms.items()}, line, plural=True)
+    return entries, locale
+
+
+def po_unquote(chunks):
+    out = []
+    for c in chunks:
+        try:
+            out.append(json.loads(c))
+        except ValueError:
+            out.append(c.strip('"'))
+    return "".join(out)
+
+
+def parse_po(text, source=False):
+    entries, header_lang, nplurals = {}, "", None
+    items, cur, field, fuzzy = [], None, None, False
+
+    def close():
+        if cur and "msgid" in cur["f"]:
+            items.append(cur)
+
+    for no, raw in enumerate(text.replace("\r\n", "\n").split("\n"), 1):
+        s = raw.strip()
+        if not s:
+            close()
+            cur, field = None, None
+            continue
+        if s.startswith("#"):
+            if s.startswith("#,") and "fuzzy" in s:
+                fuzzy = True
+            continue
+        m = re.match(r'^(msgctxt|msgid_plural|msgid|msgstr(?:\[\d+\])?)\s+(".*")\s*$', s)
+        if m:
+            kw = m.group(1)
+            if cur is None or (kw in ("msgctxt", "msgid") and any(k.startswith("msgstr") for k in cur["f"])):
+                close()
+                cur = {"f": {}, "line": no, "fuzzy": fuzzy}
+                fuzzy = False
+            if kw == "msgid":
+                cur["line"] = no
+            field = kw
+            cur["f"].setdefault(kw, []).append(m.group(2))
+        elif s.startswith('"') and cur is not None and field:
+            cur["f"][field].append(s)
+        else:
+            raise ParseError("line %d: can't read %r" % (no, s[:40]))
+    close()
+    for it in items:
+        fields = it["f"]
+        msgid = po_unquote(fields["msgid"])
+        if msgid == "" and "msgctxt" not in fields:
+            hdr = po_unquote(fields.get("msgstr", []))
+            m = re.search(r"^Language:\s*(\S+)", hdr, re.M)
+            header_lang = m.group(1) if m else ""
+            m = re.search(r"nplurals\s*=\s*(\d+)", hdr)
+            nplurals = int(m.group(1)) if m else None
+            continue
+        key = (po_unquote(fields["msgctxt"]) + "\x04" if "msgctxt" in fields else "") + msgid
+        if "msgid_plural" in fields:
+            forms = {}
+            for f, v in fields.items():
+                m = re.match(r"msgstr\[(\d+)\]", f)
+                if m:
+                    forms[int(m.group(1))] = po_unquote(v)
+            plural_src = po_unquote(fields["msgid_plural"])
+            if source:
+                value = {"one": forms.get(0) or msgid, "other": forms.get(1) or plural_src}
+            else:
+                value = {"[%d]" % i: forms[i] for i in sorted(forms)}
+            e = Entry(key, value, it["line"], plural=True, fuzzy=it["fuzzy"], gettext_n=len(forms))
+        else:
+            msgstr = po_unquote(fields.get("msgstr", []))
+            e = Entry(key, (msgstr or msgid) if source else msgstr, it["line"], fuzzy=it["fuzzy"])
+        entries[key] = e
+    if not entries and not header_lang:
+        raise ParseError("no msgid entries")
+    return entries, header_lang, nplurals
+
+
+def inner_xml(el):
+    """An element's content with its markup, as Android keeps it."""
+    parts = [el.text or ""]
+    for child in el:
+        s = ET.tostring(child, encoding="unicode")
+        parts.append(re.sub(r'\sxmlns:\w+="[^"]*"', "", s))
+    return "".join(parts)
+
+
+def android_raw_problem(raw):
+    """aapt rejects an apostrophe without a backslash outside a double-quoted string."""
+    body = raw.strip()
+    if len(body) >= 2 and body[0] == '"' and body[-1] == '"':
+        return None
+    text = re.sub(r"<[^>]*>", "", body)
+    if re.search(r"(?<!\\)'", text):
+        return "an apostrophe without a backslash: the Android build fails (write \\' )"
+    return None
+
+
+def android_unescape(s):
+    s = s.strip()
+    if len(s) >= 2 and s[0] == '"' and s[-1] == '"':
+        s = s[1:-1]
+    return re.sub(r"\\([nt'\"@?\\])", lambda m: {"n": "\n", "t": "\t"}.get(m.group(1), m.group(1)), s)
+
+
+def parse_android(text):
+    try:
+        root = ET.fromstring(text.encode("utf-8"))
+    except ET.ParseError as e:
+        raise ParseError("not valid XML: %s" % e)
+    if root.tag != "resources":
+        raise ParseError("the root element is <%s>, not <resources>" % root.tag)
+
+    def line_of(name, tag):
+        m = re.search(r'<%s\b[^>]*\bname="%s"' % (tag, re.escape(name)), text)
+        return text.count("\n", 0, m.start()) + 1 if m else None
+
+    entries = {}
+    for el in root:
+        name = el.get("name")
+        if not name or el.get("translatable") == "false":
+            continue
+        if el.tag == "string":
+            raw = inner_xml(el)
+            entries[name] = Entry(name, android_unescape(raw), line_of(name, "string"))
+            entries[name].icu_ok = android_raw_problem(raw)
+        elif el.tag == "plurals":
+            raws = {it.get("quantity"): inner_xml(it) for it in el if it.get("quantity")}
+            forms = {q: android_unescape(r) for q, r in raws.items()}
+            entries[name] = Entry(name, forms, line_of(name, "plurals"), plural=True)
+            entries[name].icu_ok = next((android_raw_problem(r) for r in raws.values() if android_raw_problem(r)), None)
+        elif el.tag == "string-array":
+            ln = line_of(name, "string-array")
+            for i, it in enumerate(el):
+                k = "%s[%d]" % (name, i)
+                entries[k] = Entry(k, android_unescape(inner_xml(it)), ln)
+    return entries
+
+
+STRINGS_RE = re.compile(r'"((?:[^"\\]|\\.)*)"\s*=\s*"((?:[^"\\]|\\.)*)"\s*;|([A-Za-z_][\w.]*)\s*=\s*"((?:[^"\\]|\\.)*)"\s*;')
+
+
+def strings_unescape(s):
+    def rep(m):
+        c = m.group(1)
+        if c.startswith("U") or c.startswith("u"):
+            try:
+                return chr(int(c[1:], 16))
+            except ValueError:
+                return c
+        return {"n": "\n", "t": "\t", "r": "\r", "0": "\0"}.get(c, c)
+    return re.sub(r"\\([Uu][0-9A-Fa-f]{4}|.)", rep, s)
+
+
+def parse_strings(text):
+    body = re.sub(r"/\*.*?\*/", lambda m: re.sub(r"[^\n]", " ", m.group(0)), text, flags=re.S)
+    body = re.sub(r"(?m)^\s*//.*$", "", body)
+    entries = {}
+    for m in STRINGS_RE.finditer(body):
+        key = strings_unescape(m.group(1) if m.group(1) is not None else m.group(3))
+        val = strings_unescape(m.group(2) if m.group(2) is not None else m.group(4))
+        entries[key] = Entry(key, val, body.count("\n", 0, m.start()) + 1)
+    rest = STRINGS_RE.sub("", body)
+    stray = re.search(r'"[^"\n]*"\s*=|=\s*"', rest)
+    if stray:
+        raise ParseError("line %d: an entry that isn't \"key\" = \"value\";" % (body.count("\n", 0, stray.start()) + 1))
+    if not entries and rest.strip():
+        raise ParseError('no "key" = "value"; entries')
+    return entries
+
+
+def parse_stringsdict(text):
+    try:
+        doc = plistlib.loads(text.encode("utf-8"))
+    except Exception as e:  # plistlib raises several types for bad input
+        raise ParseError("not a valid plist: %s" % e)
+    if not isinstance(doc, dict):
+        raise ParseError("the plist's top level is not a dictionary")
+    entries = {}
+    for key, spec in doc.items():
+        if not isinstance(spec, dict):
+            continue
+        m = re.search(r"^\s*<key>%s</key>" % re.escape(key), text, re.M)
+        line = text.count("\n", 0, m.start()) + 1 if m else None
+        fmt = spec.get("NSStringLocalizedFormatKey")
+        if isinstance(fmt, str):
+            entries[key] = Entry(key, fmt, line)
+        for var, rule in spec.items():
+            if isinstance(rule, dict) and rule.get("NSStringFormatSpecTypeKey") == "NSStringPluralRuleType":
+                forms = {c: rule[c] for c in CATEGORIES if isinstance(rule.get(c), str)}
+                k = "%s#%s" % (key, var)
+                entries[k] = Entry(k, forms, line, plural=True)
+    return entries
+
+
+def load(path, source, locale_override=""):
+    text = read_text(path)
+    fmt = detect_format(path, text)
+    locale, nplurals = "", None
+    if fmt in ("json", "arb"):
+        entries, locale = parse_json(text, arb=fmt == "arb")
+    elif fmt == "po":
+        entries, locale, nplurals = parse_po(text, source=source)
+    elif fmt == "android":
+        entries = parse_android(text)
+    elif fmt == "strings":
+        entries = parse_strings(text)
+    else:
+        entries = parse_stringsdict(text)
+    locale = locale_override or locale or detect_locale(path)
+    return {"path": path, "format": fmt, "locale": locale, "entries": entries, "nplurals": nplurals}
+
+
+# --------------------------------------------------------------------------- placeholders
+
+PRINTF = re.compile(r"%(?:%|(?:(\d+)\$|\((\w+)\))?[-+ 0#']*(?:\d+|\*)?(?:\.(?:\d+|\*))?"
+                    r"(?:hh|h|ll|l|L|q|j|z|t)?([diouxXeEfFgGaAcCsSpn@]))")
+DOUBLE = re.compile(r"\{\{\s*([^{}\s,]+)[^{}]*\}\}")
+SINGLE = re.compile(r"(?<!\{)\{\s*([A-Za-z_][\w.]*|\d+)\s*\}(?!\})")
+TAG = re.compile(r"<(/?)([A-Za-z][\w:.-]*|\d+)(\s[^<>]*?)?(/?)>")
+TYPE = {"i": "d", "u": "d", "o": "d", "x": "d", "X": "d", "F": "f", "e": "f", "E": "f",
+        "g": "f", "G": "f", "a": "f", "A": "f", "S": "s", "C": "c"}
+
+
+def printf_args(s):
+    """{position or name: type}, plus the error when plain and positional are mixed."""
+    out, plain, i = {}, 0, 0
+    for m in PRINTF.finditer(s):
+        if m.group(0) == "%%":
+            continue
+        t = TYPE.get(m.group(3), m.group(3))
+        if m.group(1):
+            out[int(m.group(1))] = t
+        elif m.group(2):
+            out[m.group(2)] = t
+        else:
+            plain += 1
+            out[plain] = t
+    return out
+
+
+def printf_tokens(s):
+    """{position or name: the specifier as written}, numbered as printf_args numbers them."""
+    out, plain = {}, 0
+    for m in PRINTF.finditer(s):
+        if m.group(0) == "%%":
+            continue
+        if m.group(1):
+            out[int(m.group(1))] = m.group(0)
+        elif m.group(2):
+            out[m.group(2)] = m.group(0)
+        else:
+            plain += 1
+            out[plain] = m.group(0)
+    return out
+
+
+def brace_args(s):
+    names = [m.group(1) for m in DOUBLE.finditer(s)]
+    names += ["{%s}" % m.group(1) for m in SINGLE.finditer(DOUBLE.sub("", s))]
+    return names
+
+
+# --------------------------------------------------------------------------- ICU
+
+class IcuError(Exception):
+    pass
+
+
+def icu_parse(s):
+    """Parse an ICU MessageFormat string. Returns [(name, type, {keyword: ...})] for every
+    argument, nested ones included. Raises IcuError when the message doesn't parse."""
+    args, pos = [], [0]
+    n = len(s)
+
+    def message(depth, in_plural):
+        if depth > 50:
+            raise IcuError("nested more than 50 levels deep")
+        while pos[0] < n:
+            c = s[pos[0]]
+            if c == "'":
+                if pos[0] + 1 < n and s[pos[0] + 1] == "'":
+                    pos[0] += 2
+                elif pos[0] + 1 < n and s[pos[0] + 1] in "{}#|":
+                    end = s.find("'", pos[0] + 1)
+                    pos[0] = n if end < 0 else end + 1
+                else:
+                    pos[0] += 1
+            elif c == "{":
+                pos[0] += 1
+                argument(depth)
+            elif c == "}":
+                if depth == 0:
+                    raise IcuError("an unmatched '}' at %d" % pos[0])
+                return
+            else:
+                pos[0] += 1
+        if depth:
+            raise IcuError("a '{' that is never closed")
+
+    def ws():
+        while pos[0] < n and s[pos[0]].isspace():
+            pos[0] += 1
+
+    def ident():
+        ws()
+        m = re.compile(r"[^\s{},'#]+").match(s, pos[0])
+        if not m:
+            raise IcuError("expected a name at %d" % pos[0])
+        pos[0] = m.end()
+        return m.group(0)
+
+    def argument(depth):
+        name = ident()
+        ws()
+        if pos[0] < n and s[pos[0]] == "}":
+            pos[0] += 1
+            args.append((name, "", {}))
+            return
+        if pos[0] >= n or s[pos[0]] != ",":
+            raise IcuError("expected ',' or '}' after %r" % name)
+        pos[0] += 1
+        kind = ident().lower()
+        ws()
+        if kind in ("plural", "selectordinal", "select"):
+            if pos[0] >= n or s[pos[0]] != ",":
+                raise IcuError("expected ',' after %s" % kind)
+            pos[0] += 1
+            branches = {}
+            while True:
+                ws()
+                if pos[0] < n and s[pos[0]] == "}":
+                    pos[0] += 1
+                    break
+                if pos[0] >= n:
+                    raise IcuError("a %s that is never closed" % kind)
+                key = ident()
+                if key.startswith("offset:"):
+                    continue
+                ws()
+                if pos[0] >= n or s[pos[0]] != "{":
+                    raise IcuError("expected '{' after %r in %s" % (key, name))
+                pos[0] += 1
+                message(depth + 1, kind != "select")
+                if pos[0] >= n or s[pos[0]] != "}":
+                    raise IcuError("a branch of %s that is never closed" % name)
+                pos[0] += 1
+                branches[key] = True
+            if "other" not in branches:
+                raise IcuError("%s %r has no 'other' branch" % (kind, name))
+            args.append((name, kind, branches))
+        else:
+            if pos[0] < n and s[pos[0]] == ",":
+                end = s.find("}", pos[0])
+                if end < 0:
+                    raise IcuError("an argument that is never closed")
+                pos[0] = end
+            ws()
+            if pos[0] >= n or s[pos[0]] != "}":
+                raise IcuError("expected '}' to close %r" % name)
+            pos[0] += 1
+            args.append((name, kind, {}))
+
+    message(0, False)
+    return args
+
+
+ICU_COMPLEX = re.compile(r"\{\s*[\w.]+\s*,\s*(?:plural|select|selectordinal|number|date|time|spellout|ordinal)\b")
+
+
+def looks_icu(s):
+    """An ICU message with a formatted argument. A string with only {name} placeholders is
+    checked as plain text, so a dropped placeholder gets its own finding and fix."""
+    return "{{" not in s and bool(ICU_COMPLEX.search(s))
+
+
+# --------------------------------------------------------------------------- checks
+
+def visible(s):
+    """Characters a reader sees: markup and placeholders out, combining marks not counted."""
+    s = TAG.sub("", s)
+    s = PRINTF.sub(lambda m: "%" if m.group(0) == "%%" else "", s)
+    s = DOUBLE.sub("", s)
+    s = SINGLE.sub("", s)
+    return sum(1 for ch in unicodedata.normalize("NFC", s) if not unicodedata.combining(ch))
+
+
+def tags_of(s):
+    return [(m.group(1) == "/", m.group(2).lower(), m.group(4) == "/") for m in TAG.finditer(s)]
+
+
+def tag_problem(src, tgt):
+    a, b = tags_of(src), tags_of(tgt)
+    if sorted(a) != sorted(b):
+        names = lambda ts: ", ".join("<%s%s%s>" % ("/" if c else "", t, "/" if sc else "") for c, t, sc in ts) or "none"
+        return "markup differs from the source: source has %s, target has %s" % (names(a), names(b))
+    stack = []
+    for close, name, self_close in b:
+        if self_close:
+            continue
+        if not close:
+            stack.append(name)
+        elif not stack or stack.pop() != name:
+            return "markup tags don't nest: </%s> closes the wrong tag" % name
+    if stack:
+        return "markup tag <%s> is never closed" % stack[-1]
+    return None
+
+
+SCRIPT_OF = (("LATIN", "Latin"), ("CYRILLIC", "Cyrillic"), ("GREEK", "Greek"))
+
+
+def mixed_script(s):
+    for word in re.findall(r"\w+", s):
+        seen = set()
+        for ch in word:
+            if ch.isalpha():
+                name = unicodedata.name(ch, "")
+                for k, v in SCRIPT_OF:
+                    if name.startswith(k):
+                        seen.add(v)
+        if len(seen) > 1:
+            return word, sorted(seen)
+    return None
+
+
+END_PUNCT = set(".!?:…。！？：؟;·।")
+
+
+def letters(s):
+    return sum(1 for ch in TAG.sub("", PRINTF.sub("", SINGLE.sub("", DOUBLE.sub("", s)))) if ch.isalpha())
+
+
+class Report(object):
+    def __init__(self):
+        self.failures, self.warnings = [], []
+
+    def add(self, level, locale, key, where, problem, source="", target="", fix=""):
+        row = {"locale": locale, "key": key.replace("\x04", " | "), "where": where, "problem": problem,
+               "source": source, "target": target}
+        if fix:
+            row["fix"] = fix
+        (self.failures if level == "fail" else self.warnings).append(row)
+
+
+def where(t, e):
+    return "%s:%s" % (os.path.basename(t["path"]), e.line) if e and e.line else os.path.basename(t["path"])
+
+
+def short(v):
+    if isinstance(v, dict):
+        return "; ".join("%s: %s" % (k, x) for k, x in v.items())
+    return v
+
+
+def compare_placeholders(src_texts, tgt_text):
+    """A problem with one target string's placeholders against the source's, or None."""
+    src_join = " ".join(src_texts)
+    sp, tp = {}, printf_args(tgt_text)
+    shown = {}
+    for s in src_texts:
+        for k, v in printf_args(s).items():
+            sp.setdefault(k, v)
+        for k, v in printf_tokens(s).items():
+            shown.setdefault(("s", k), v)
+    for k, v in printf_tokens(tgt_text).items():
+        shown.setdefault(("t", k), v)
+
+    def fmt_printf(k, t, side="s"):
+        return shown.get((side, k)) or ("%%%s$%s" % (k, t) if isinstance(k, int) else "%%(%s)%s" % (k, t))
+    if len(src_texts) == 1:
+        if sp != tp:
+            missing = [k for k in sp if k not in tp]
+            extra = [k for k in tp if k not in sp]
+            changed = [k for k in sp if k in tp and sp[k] != tp[k]]
+            bits = []
+            if missing:
+                bits.append("missing %s" % ", ".join(fmt_printf(k, sp[k]) for k in missing))
+            if extra:
+                bits.append("extra %s" % ", ".join(fmt_printf(k, tp[k], "t") for k in extra))
+            if changed:
+                named = all(not isinstance(k, int) for k in changed)
+                bits.append("the arguments are %s in the source but %s here, %s" % (
+                    " ".join(fmt_printf(k, sp[k]) for k in sorted(sp, key=str)),
+                    " ".join(fmt_printf(k, tp[k], "t") for k in sorted(tp, key=str)),
+                    "a changed type that msgfmt --check and translation builds reject" if named else
+                    "so the values swap at runtime or the app crashes formatting them"))
+            if not bits:
+                bits.append("arguments reordered")
+            return "printf placeholders differ: " + "; ".join(bits)
+    else:
+        extra = [k for k in tp if k not in sp or sp[k] != tp[k]]
+        if extra:
+            return "printf placeholder not in the source: %s" % ", ".join(fmt_printf(k, tp[k], "t") for k in extra)
+    sb, tb = brace_args(src_join) if len(src_texts) > 1 else brace_args(src_texts[0]), brace_args(tgt_text)
+    if len(src_texts) == 1:
+        if sorted(sb) != sorted(tb):
+            missing = [x for x in set(sb) if sb.count(x) > tb.count(x)]
+            extra = [x for x in set(tb) if tb.count(x) > sb.count(x)]
+            bits = (["missing " + ", ".join(sorted(fmt_brace(x) for x in missing))] if missing else []) + \
+                   (["extra " + ", ".join(sorted(fmt_brace(x) for x in extra))] if extra else [])
+            return "placeholders differ: " + "; ".join(bits)
+    else:
+        extra = sorted(set(tb) - set(sb))
+        if extra:
+            return "placeholder not in the source: " + ", ".join(fmt_brace(x) for x in extra)
+    return None
+
+
+def fmt_printf(k, t):
+    return "%%%s$%s" % (k, t) if isinstance(k, int) else "%%(%s)%s" % (k, t)
+
+
+def fmt_brace(x):
+    return x if x.startswith("{") else "{{%s}}" % x
+
+
+def required_plurals(lang):
+    cats = CLDR.get(lang)
+    return set(cats) if cats else None
+
+
+def check_target(src, tgt, budget, rep):
+    loc = tgt["locale"] or "?"
+    lang = lang_of(tgt["locale"])
+    S, T = src["entries"], tgt["entries"]
+    gettext_n = NPLURALS.get(lang)
+    if tgt["format"] == "po" and tgt["nplurals"] and gettext_n and tgt["nplurals"] != gettext_n:
+        rep.add("fail", loc, "(Plural-Forms header)", os.path.basename(tgt["path"]),
+                "the header sets nplurals=%d; %s needs %d" % (tgt["nplurals"], lang, gettext_n))
+    for key, se in S.items():
+        te = T.get(key)
+        w = where(tgt, te)
+        if te is None:
+            rep.add("fail", loc, key, os.path.basename(tgt["path"]), "missing", short(se.value), "")
+            continue
+        if tgt["format"] == "android" and te.icu_ok:
+            rep.add("fail", loc, key, w, te.icu_ok, short(se.value), short(te.value))
+        if te.fuzzy:
+            rep.add("warn", loc, key, w, "marked fuzzy: msgfmt leaves it out, so the source text shows",
+                    short(se.value), short(te.value))
+        if se.plural and not te.plural:
+            te = Entry(te.key, {"other": te.value}, te.line, plural=True)
+        if not se.plural and te.plural:
+            rep.add("fail", loc, key, w, "a plural set where the source has one string", se.value, short(te.value))
+            continue
+        if se.plural:
+            check_plural(se, te, tgt, lang, loc, w, rep)
+            src_texts = [x for x in se.value.values() if x]
+        else:
+            src_texts = [se.value]
+            if te.value.strip() == "" and se.value.strip():
+                rep.add("fail", loc, key, w, "empty translation", se.value, te.value)
+                continue
+        for label, tv in (te.value.items() if te.plural else [("", te.value)]):
+            if te.plural and tv == "":
+                continue
+            tag = ("[%s] " % label) if label else ""
+            sv = src_texts[0] if len(src_texts) == 1 else (se.value.get(label) or se.value.get("other") or src_texts[0])
+            icu_src = [x for x in src_texts if looks_icu(x)]
+            icu_handled = False
+            if icu_src:
+                try:
+                    sargs = [a for x in icu_src for a in icu_parse(x)]
+                    src_ok = True
+                except IcuError:
+                    src_ok = False
+                if src_ok:
+                    icu_handled = True
+                    try:
+                        targs = icu_parse(tv)
+                    except IcuError as e:
+                        rep.add("fail", loc, key, w, tag + "ICU message doesn't parse: %s" % e, sv, tv)
+                        continue
+                    check_icu(sargs, targs, lang, loc, key, w, tag, sv, tv, rep)
+            if not icu_handled:
+                last = label == "other" or (tgt["format"] == "po" and label == "[%d]" % (len(te.value) - 1))
+                strict = (not se.plural) or (last and len(te.value) > 1)
+                p = compare_placeholders([se.value.get("other", sv)] if se.plural and strict else
+                                         (src_texts if se.plural else [sv]), tv)
+                if p:
+                    rep.add("fail", loc, key, w, tag + p, sv, tv, fix_for(sv, tv))
+            elif PRINTF.search(sv.replace("%%", "")) or PRINTF.search(tv.replace("%%", "")):
+                p = compare_placeholders([sv], tv)
+                if p and "printf" in p:
+                    rep.add("fail", loc, key, w, tag + p, sv, tv)
+            t = tag_problem(sv, tv)
+            if t:
+                rep.add("fail", loc, key, w, tag + t, sv, tv)
+            limit = budget_limit(budget, key, sv)
+            if limit is not None and visible(tv) > limit:
+                rep.add("fail", loc, key, w, tag + "%d characters, over the budget of %d" % (visible(tv), limit), sv, tv)
+            style_warnings(sv, tv, loc, key, w, tag, rep, tgt)
+    for key, te in T.items():
+        if key not in S and tgt["format"] == "android" and te.icu_ok:
+            rep.add("fail", loc, key, where(tgt, te), te.icu_ok, "", short(te.value))
+        if key not in S:
+            rep.add("warn", loc, key, where(tgt, te), "not in the source (stale or misspelled key)", "", short(te.value))
+
+
+def check_plural(se, te, tgt, lang, loc, w, rep):
+    if tgt["format"] == "po":
+        need = NPLURALS.get(lang)
+        have = te.gettext_n if te.gettext_n is not None else len(te.value)
+        if need and have != need:
+            rep.add("fail", loc, se.key, w, "has %d plural form%s; %s needs %d (msgstr[0] to msgstr[%d])" % (
+                have, "" if have == 1 else "s", lang, need, need - 1),
+                    short(se.value), short(te.value))
+        for k, v in te.value.items():
+            if not v.strip():
+                rep.add("fail", loc, se.key, w, "plural form %s is empty" % k, short(se.value), short(te.value))
+        return
+    need = required_plurals(lang)
+    have = {k for k, v in te.value.items() if v != ""}
+    if need is None:
+        if "other" not in have:
+            rep.add("fail", loc, se.key, w, "no 'other' plural form", short(se.value), short(te.value))
+        return
+    missing = need - have
+    soft = missing & SOFT.get(lang, set())
+    hard = missing - soft
+    if hard:
+        rep.add("fail", loc, se.key, w, "plural forms %s missing; %s needs %s" % (
+            ", ".join(c for c in CATEGORIES if c in hard), lang, ", ".join(c for c in CATEGORIES if c in need)),
+            short(se.value), short(te.value))
+    if soft:
+        rep.add("warn", loc, se.key, w, "plural form 'many' missing; %s uses it for large numbers such as 1 000 000" % lang,
+                short(se.value), short(te.value))
+    extra = {k for k in have if k in CATEGORIES} - need
+    if extra:
+        rep.add("warn", loc, se.key, w, "plural forms %s are never used in %s" % (
+            ", ".join(c for c in CATEGORIES if c in extra), lang), short(se.value), short(te.value))
+
+
+def check_icu(sargs, targs, lang, loc, key, w, tag, sv, tv, rep):
+    sn = {(a, k if k in ("plural", "select", "selectordinal") else "") for a, k, _ in sargs}
+    tn = {(a, k if k in ("plural", "select", "selectordinal") else "") for a, k, _ in targs}
+    if sn != tn:
+        missing = sorted("{%s}" % a for a, _ in sn - tn)
+        extra = sorted("{%s}" % a for a, _ in tn - sn)
+        bits = (["missing " + ", ".join(missing)] if missing else []) + (["extra " + ", ".join(extra)] if extra else [])
+        rep.add("fail", loc, key, w, tag + "ICU arguments differ: " + "; ".join(bits or ["an argument changed type"]), sv, tv)
+    sel_src = {a: set(b) for a, k, b in sargs if k == "select"}
+    for a, k, b in targs:
+        if k == "select" and a in sel_src:
+            lost = sel_src[a] - set(b)
+            if lost:
+                rep.add("fail", loc, key, w, tag + "select %r lost keyword %s (keywords are code, not text)" % (
+                    a, ", ".join(sorted(lost))), sv, tv)
+        if k == "plural":
+            need = required_plurals(lang)
+            if need:
+                have = {x for x in b if x in CATEGORIES}
+                missing = need - have - SOFT.get(lang, set())
+                if missing:
+                    rep.add("fail", loc, key, w, tag + "ICU plural %r is missing %s; %s needs %s" % (
+                        a, ", ".join(c for c in CATEGORIES if c in missing), lang,
+                        ", ".join(c for c in CATEGORIES if c in need)), sv, tv)
+            bad = [x for x in b if x not in CATEGORIES and not re.match(r"^=\d+$", x)]
+            if bad:
+                rep.add("fail", loc, key, w, tag + "ICU plural %r has keyword %s, which isn't a plural category" % (
+                    a, ", ".join(sorted(bad))), sv, tv)
+
+
+def fix_for(sv, tv):
+    """A mechanical fix when exactly one placeholder was dropped and the source has it once:
+    say what to restore, never where in the sentence, which takes the language."""
+    sb, tb = brace_args(sv), brace_args(tv)
+    sp, tp = printf_args(sv), printf_args(tv)
+    gone = [x for x in set(sb) if sb.count(x) > tb.count(x)]
+    if len(gone) == 1 and sp == tp and not [x for x in tb if x not in sb]:
+        return "restore %s" % fmt_brace(gone[0])
+    if sb == tb and len(sp) > 1 and sorted(sp.values()) == sorted(tp.values()) and set(sp) == set(tp) and sp != tp:
+        # Number each of the target's plain specifiers with the source argument of its type, in order.
+        free = {t: [k for k in sorted(k for k in sp if isinstance(k, int)) if sp[k] == t] for t in set(sp.values())}
+        out, ok = [], True
+
+        def num(m):
+            nonlocal ok
+            if m.group(0) == "%%" or m.group(1) or m.group(2):
+                return m.group(0)
+            t = TYPE.get(m.group(3), m.group(3))
+            if not free.get(t):
+                ok = False
+                return m.group(0)
+            k = free[t].pop(0)
+            return m.group(0).replace("%", "%%%d$" % k, 1)
+        fixed = PRINTF.sub(num, tv)
+        return ("use %s" % fixed) if ok and len({v for v in sp.values()}) == len(sp) else \
+            "number the arguments in source order (%s) so a translation can move them" % " ".join(
+                "%%%d$%s" % (k, v) for k, v in sorted(sp.items()))
+    return ""
+
+
+def style_warnings(sv, tv, loc, key, w, tag, rep, tgt):
+    if sv == tv and letters(sv) > 3 and not re.match(r"^[\w.+-]+@|^https?://", sv) and lang_of(tgt["locale"]) != "en":
+        rep.add("warn", loc, key, w, tag + "identical to the source: untranslated?", sv, tv)
+    if (sv[:1].isspace()) != (tv[:1].isspace()) or (sv[-1:].isspace()) != (tv[-1:].isspace()):
+        rep.add("warn", loc, key, w, tag + "leading or trailing space changed", repr(sv), repr(tv))
+    s_end = sv.rstrip()[-1:] if sv.rstrip() else ""
+    t_end = tv.rstrip()[-1:] if tv.rstrip() else ""
+    if s_end in END_PUNCT and t_end not in END_PUNCT and tv.strip():
+        rep.add("warn", loc, key, w, tag + "ending punctuation %r dropped" % s_end, sv, tv)
+    mix = mixed_script(tv)
+    if mix:
+        rep.add("warn", loc, key, w, tag + "%r mixes %s letters (a look-alike typed by mistake?)" % (
+            mix[0], " and ".join(mix[1])), sv, tv)
+
+
+# --------------------------------------------------------------------------- budget
+
+def load_budget(path):
+    text = read_text(path)
+    if path.lower().endswith(".csv"):
+        keys = {}
+        rows = csv.DictReader(io.StringIO(text))
+        cols = {c.lower().strip(): c for c in (rows.fieldnames or [])}
+        if "key" not in cols or not ({"max", "max_chars", "limit"} & set(cols)):
+            raise BadInput("%s: a budget CSV needs columns key and max" % path)
+        mc = cols.get("max") or cols.get("max_chars") or cols.get("limit")
+        for r in rows:
+            try:
+                keys[r[cols["key"]].strip()] = int(r[mc])
+            except (TypeError, ValueError):
+                raise BadInput("%s: %r is not a number of characters" % (path, r.get(mc)))
+        return {"keys": keys, "ratio": None, "ratio_min_source": 10}
+    try:
+        doc = json.loads(text)
+    except ValueError as e:
+        raise BadInput("%s: not valid JSON: %s" % (path, e))
+    if not isinstance(doc, dict):
+        raise BadInput("%s: the budget must be an object" % path)
+    keys = doc.get("keys", {})
+    if not isinstance(keys, dict) or not all(isinstance(v, int) and not isinstance(v, bool) for v in keys.values()):
+        raise BadInput("%s: \"keys\" must map each key to a whole number of characters" % path)
+    ratio = doc.get("ratio")
+    if ratio is not None and (not isinstance(ratio, (int, float)) or isinstance(ratio, bool) or ratio <= 0):
+        raise BadInput("%s: \"ratio\" must be a positive number" % path)
+    rmin = doc.get("ratio_min_source", 10)
+    if not isinstance(rmin, int) or isinstance(rmin, bool):
+        raise BadInput("%s: \"ratio_min_source\" must be a whole number" % path)
+    return {"keys": keys, "ratio": ratio, "ratio_min_source": rmin}
+
+
+def budget_limit(budget, key, sv):
+    if not budget:
+        return None
+    k = key.replace("\x04", " | ")
+    if k in budget["keys"]:
+        return budget["keys"][k]
+    if budget["ratio"] and visible(sv) >= budget["ratio_min_source"]:
+        return int(math.ceil(visible(sv) * budget["ratio"]))
+    return None
+
+
+# --------------------------------------------------------------------------- output
+
+def esc(s, width=90):
+    s = "" if s is None else str(s)
+    s = s.replace("\n", "\\n").replace("|", "\\|")
+    return s if len(s) <= width else s[:width - 3] + "..."
+
+
+def render(src, targets, rep, budget):
+    out = ["# Translation check", ""]
+    out.append("Source: `%s` (%s), %d strings." % (os.path.basename(src["path"]), src["locale"] or "locale unknown",
+                                                    len(src["entries"])))
+    for t in targets:
+        out.append("Target: `%s` (%s), %d strings." % (os.path.basename(t["path"]), t["locale"] or "locale unknown",
+                                                       len(t["entries"])))
+    out.append("")
+    nf, nw = len(rep.failures), len(rep.warnings)
+    out.append("**%s.** %d warning(s)." % ("No failures" if not nf else "%d failure(s)" % nf, nw))
+    if not budget:
+        out.append("No character budget given, so string lengths were not checked.")
+    out.append("")
+    for title, rows in (("Failures", rep.failures), ("Warnings", rep.warnings)):
+        if not rows:
+            continue
+        out += ["## " + title, "", "| Locale | Key | Problem | Where | Source | Target | Fix |", "|---|---|---|---|---|---|---|"]
+        for r in sorted(rows, key=lambda r: (r["locale"], r["key"])):
+            out.append("| %s | `%s` | %s | %s | %s | %s | %s |" % (
+                esc(r["locale"]), esc(r["key"]), esc(r["problem"], 240), esc(r["where"]), esc(r["source"]),
+                esc(r["target"]), esc(r.get("fix", ""), 160)))
+        out.append("")
+    return "\n".join(out) + "\n"
+
+
+def main(argv=None):
+    ap = argparse.ArgumentParser(description=__doc__.split("\n")[0])
+    ap.add_argument("source")
+    ap.add_argument("targets", nargs="+")
+    ap.add_argument("--locale", action="append", default=[],
+                    help="the locale of each target, in order (overrides the path)")
+    ap.add_argument("--source-locale", default="")
+    ap.add_argument("--budget", help="JSON or CSV of character limits")
+    ap.add_argument("--json", action="store_true")
+    ap.add_argument("-o", "--out")
+    a = ap.parse_args(argv)
+    try:
+        try:
+            src = load(a.source, True, a.source_locale)
+        except ParseError as e:
+            raise BadInput("%s: %s" % (a.source, e))
+        budget = load_budget(a.budget) if a.budget else None
+        rep, targets = Report(), []
+        for i, path in enumerate(a.targets):
+            override = a.locale[i] if i < len(a.locale) else ""
+            try:
+                t = load(path, False, override)
+            except ParseError as e:
+                loc = override or detect_locale(path) or "?"
+                rep.add("fail", loc, "(file)", os.path.basename(path), "the file doesn't parse: %s" % e)
+                continue
+            targets.append(t)
+            check_target(src, t, budget, rep)
+    except BadInput as e:
+        print("error: %s" % e, file=sys.stderr)
+        return 2
+    if a.json:
+        text = json.dumps({"source": {"file": src["path"], "locale": src["locale"], "strings": len(src["entries"])},
+                           "targets": [{"file": t["path"], "locale": t["locale"], "format": t["format"],
+                                        "strings": len(t["entries"])} for t in targets],
+                           "lengths_checked": bool(budget), "failures": rep.failures, "warnings": rep.warnings},
+                          ensure_ascii=False, indent=2) + "\n"
+    else:
+        text = render(src, targets, rep, budget)
+    if a.out:
+        try:
+            with open(a.out, "w", encoding="utf-8") as fh:
+                fh.write(text)
+        except OSError as e:
+            print("error: can't write %s: %s" % (a.out, e.strerror or e), file=sys.stderr)
+            return 2
+    else:
+        sys.stdout.write(text)
+    return 1 if rep.failures else 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())