Blame for rules/units.py

import random
import re
from dataclasses import dataclass
from decimal import Decimal

from num2fawords import words
from num2words import num2words
from persiantools import digits

from rules.base import BaseProcessor
from rules.fraction import FRACTION_BODY, VULGAR_FRACTIONS, FractionProcessor
from rules.wrappers import CLOSERS, OPENERS
from utils import NO_LETTERS, detect_digit_language, detect_language


@dataclass(frozen=True)
class Unit:
    fa: str
    en: str
    en_plural: str
    # Next smaller unit and how many of it make one of this unit, used to read
    # `1000.5 m` as "1000 metres and 50 centimetres".
    minor: "Unit" = None
    factor: int = 1


MILLIMETRE = Unit("میلیمتر", "millimetre", "millimetres")
CENTIMETRE = Unit("سانتیمتر", "centimetre", "centimetres", MILLIMETRE, 10)
METRE = Unit("متر", "metre", "metres", CENTIMETRE, 100)
KILOMETRE = Unit("کیلومتر", "kilometre", "kilometres", METRE, 1000)
MILE = Unit("مایل", "mile", "miles")
FOOT = Unit("فوت", "foot", "feet")

MILLIGRAM = Unit("میلی‌گرم", "milligram", "milligrams")
GRAM = Unit("گرم", "gram", "grams", MILLIGRAM, 1000)
KILOGRAM = Unit("کیلوگرم", "kilogram", "kilograms", GRAM, 1000)
POUND = Unit("پوند", "pound", "pounds")
OUNCE = Unit("اونس", "ounce", "ounces")

MILLILITRE = Unit("میلی‌لیتر", "millilitre", "millilitres")
LITRE = Unit("لیتر", "litre", "litres", MILLILITRE, 1000)

MILLISECOND = Unit("میلی‌ثانیه", "millisecond", "milliseconds")
SECOND = Unit("ثانیه", "second", "seconds", MILLISECOND, 1000)
MINUTE = Unit("دقیقه", "minute", "minutes", SECOND, 60)
HOUR = Unit("ساعت", "hour", "hours", MINUTE, 60)
DAY = Unit("روز", "day", "days", HOUR, 24)
WEEK = Unit("هفته", "week", "weeks", DAY, 7)
MONTH = Unit("ماه", "month", "months")
YEAR = Unit("سال", "year", "years")

PERCENT = Unit("درصد", "percent", "percent")

KILOMETRE_PER_HOUR = Unit("کیلومتر بر ساعت", "kilometre per hour", "kilometres per hour")
METRE_PER_SECOND = Unit("متر بر ثانیه", "metre per second", "metres per second")

CELSIUS = Unit("درجه سانتیگراد", "degree Celsius", "degrees Celsius")
FAHRENHEIT = Unit("درجه فارنهایت", "degree Fahrenheit", "degrees Fahrenheit")

# Symbols are case-sensitive on purpose (e.g. `5G` is a network, not grams).
# Single `s` and `d` are left out on purpose: `1990s` and `3d` are not units.
UNITS = {
    "km": KILOMETRE,
    "m": METRE,
    "cm": CENTIMETRE,
    "mm": MILLIMETRE,
    "mi": MILE,
    "ft": FOOT,
    "kg": KILOGRAM,
    "KG": KILOGRAM,
    "Kg": KILOGRAM,
    "g": GRAM,
    "mg": MILLIGRAM,
    "lb": POUND,
    "lbs": POUND,
    "oz": OUNCE,
    "l": LITRE,
    "L": LITRE,
    "ml": MILLILITRE,
    "mL": MILLILITRE,
    "ms": MILLISECOND,
    "sec": SECOND,
    "secs": SECOND,
    "second": SECOND,
    "seconds": SECOND,
    "min": MINUTE,
    "mins": MINUTE,
    "minute": MINUTE,
    "minutes": MINUTE,
    "h": HOUR,
    "hr": HOUR,
    "hrs": HOUR,
    "hour": HOUR,
    "hours": HOUR,
    "day": DAY,
    "days": DAY,
    "week": WEEK,
    "weeks": WEEK,
    "month": MONTH,
    "months": MONTH,
    "year": YEAR,
    "years": YEAR,
    "%": PERCENT,
    "٪": PERCENT,
    "km/h": KILOMETRE_PER_HOUR,
    "m/s": METRE_PER_SECOND,
    "°C": CELSIUS,
    "°F": FAHRENHEIT,
}

AMOUNT = r"\d+(?:[.,٫٬]\d+)*"

# Longest symbols first so `km/h` wins over `km`, and `mg` over `m`.
SYMBOLS = "|".join(re.escape(symbol) for symbol in sorted(UNITS, key=len, reverse=True))

PERCENT_SIGNS = ("%", "٪")
OTHER_SYMBOLS = "|".join(
    re.escape(symbol)
    for symbol in sorted(UNITS, key=len, reverse=True)
    if symbol not in PERCENT_SIGNS
)

# A number or fraction, optional spaces, then a unit symbol: `5km`, `۵ km`,
# `2.5 kg`, `1/2 kg`, `2 1/2 kg`. The symbol must stand alone, so `5 km²`
# or `5G` don't match unless they are in the table.
# The amount and/or the symbol may be wrapped: `2 (kg)`, `3 {m}`, `(5) kg`.
# After a percent sign a unit symbol may follow (`5% kg`): that is `per_unit`.
UNIT_PATTERN = re.compile(
    rf"""
    (?<![\w./,٫٬])
    (?:(?P<aopen>{OPENERS})[^\S\n]*)?
    (?:{FRACTION_BODY}|(?P<amount>{AMOUNT}))
    (?(aopen)[^\S\n]*(?P<aclose>{CLOSERS}))
    [^\S\n]*
    (?:(?P<sopen>{OPENERS})[^\S\n]*)?
    (?P<symbol>{SYMBOLS})
    (?(sopen)[^\S\n]*(?P<sclose>{CLOSERS}))
    (?![\w/])
    (?P<tail>[^\S\n]+(?P<per_unit>{OTHER_SYMBOLS})(?![\w/]))?
    """,
    re.VERBOSE,
)


# A percent sign typed before the number (`%20`, `% 20`, `٪ ۲۰`). It is moved
# after the number so the normal pattern can read it. `%d`, `%s`, `% 5d` and
# `a%20b` don't match.
PREFIX_PERCENT_PATTERN = re.compile(
    rf"(?<![\w%٪])(?P<sign>[%٪])[^\S\n]*(?P<amount>{AMOUNT})(?![\w%٪/])"
)


class UnitProcessor(BaseProcessor):
    """Converts a number followed by a unit symbol into spoken text.

    Persian: `5 km`  -> `پنج کیلومتر`
             `۷۵kg`  -> `هفتاد و پنج کیلوگرم`
    English: `5 km`  -> `five kilometres`
             `1 kg`  -> `one kilogram`
             `25°C`  -> `twenty-five degrees Celsius`
             `2 %`   -> `two percent` (Persian: `۲ ٪` -> `دو درصد`)
             `10 sec` -> `ten seconds`

    Decimals are split into smaller units when the unit has one:
             `1000.5 m` -> `یک هزار متر و پنجاه سانتیمتر`
             `2.5 kg`   -> `two kilograms and five hundred grams`
             `1.5 h`    -> `one hour and thirty minutes`
             `1/2 %`    -> `one half percent`
    Other decimals are read as a plain number: `2.5 mi` -> `two point five miles`.

    A percent sign may come before the number (`% 5`, `٪ ۵`), and a unit symbol
    after a percent sign is read on its own: `5% kg` -> `five percent kilogram`.

    With nothing but units in the text, Persian digits give Persian
    (`۱۰۰۰m` -> `یک هزار متر`) and ASCII digits give English.

    Works on the whole text (not word by word), so both `5km` and `5 km`
    are handled. Add a unit by adding one line to `UNITS`.
    """

    def __init__(self, epsilon: float = 1.0):
        super(UnitProcessor, self).__init__(epsilon=epsilon)

    def process(self, text: str, language: str = None, *args, **kwargs) -> str:
        if random.random() > self.epsilon:
            return text

        language = self._resolve_language(text, language)
        if language is None:
            return text

        def convert(match: re.Match) -> str:
            return self._convert(match, language)

        text = PREFIX_PERCENT_PATTERN.sub(self._move_prefix_percent, text)
        return UNIT_PATTERN.sub(convert, text)

    @staticmethod
    def _move_prefix_percent(match: re.Match) -> str:
        """`% 20 cm` -> `20% cm`, so it reads as `twenty percent centimetre`."""
        before = match.string[: match.start()].rstrip(" \t")
        # In `5 % 2` the sign belongs to the 5 before it, not to the 2 after it.
        if before and (before[-1].isdigit() or before[-1] in VULGAR_FRACTIONS):
            return match.group(0)
        return f"{match.group('amount')}{match.group('sign')}"

    @staticmethod
    def _resolve_language(text: str, language: str):
        """Detects the language ignoring the unit symbols themselves.

        Symbols like `km` are Latin letters and would otherwise tip a Persian
        sentence towards English. If nothing but units is left (e.g. `۱۰۰۰m`),
        Persian digits mean Persian; otherwise the given `language` is used,
        and with no language given ASCII digits mean English.
        """
        detected = detect_language(UNIT_PATTERN.sub(" ", text))
        if detected in ("fa", "en"):
            return detected
        if detected != NO_LETTERS:
            return None

        digit_language = detect_digit_language(text)
        if digit_language == "fa":
            return "fa"
        if language in ("fa", "en"):
            return language
        if language is None or language == NO_LETTERS:
            return digit_language
        return None

    @staticmethod
    def _convert(match: re.Match, language: str) -> str:
        unit = UNITS[match.group("symbol")]
        pieces = UnitProcessor._amount_pieces(match, unit, language)
        if pieces is None:
            return match.group(0)

        # Wrappers stay where they were: `2 (kg)` -> `two (kilograms)`.
        rendered = []
        for index, (amount_text, unit_text) in enumerate(pieces):
            if index == 0 and match.group("aopen"):
                amount_text = f"{match.group('aopen')}{amount_text}{match.group('aclose')}"
            if match.group("sopen"):
                unit_text = f"{match.group('sopen')}{unit_text}{match.group('sclose')}"
            rendered.append(f"{amount_text} {unit_text}")

        if language == "fa" or len(rendered) == 1:
            text = " و ".join(rendered) if language == "fa" else rendered[0]
        else:
            text = f"{', '.join(rendered[:-1])} and {rendered[-1]}"

        per_unit = match.group("per_unit")
        if per_unit is None:
            return text

        if unit is PERCENT:
            # `5% kg` -> `five percent kilogram`: the unit has no number of its own.
            other = UNITS[per_unit]
            return f"{text} {other.fa if language == 'fa' else other.en}"

        # A symbol after any other unit is not ours to convert: put it back.
        return text + match.group("tail")

    @staticmethod
    def _amount_pieces(match: re.Match, unit: Unit, language: str):
        """Reads the amount as [(amount words, unit words), ...]; None if it
        should stay unchanged. Several pieces mean a split decimal (`2.5 kg`)."""
        if match.group("amount") is None:
            return UnitProcessor._fraction_pieces(match, unit, language)

        amount = digits.fa_to_en(match.group("amount"))
        clean = amount.replace(",", "").replace("٬", "").replace("٫", ".")

        parts = UnitProcessor._split_into_units(Decimal(clean), unit)

        if language == "fa":
            if parts is None:
                return [(words(float(clean)), unit.fa)]
            return [(words(number), part.fa) for number, part in parts]

        if parts is None:
            return [(num2words(Decimal(clean)), unit.en_plural)]
        return [
            (num2words(number), part.en if number == 1 else part.en_plural)
            for number, part in parts
        ]

    @staticmethod
    def _split_into_units(value: Decimal, unit: Unit):
        """Splits a decimal amount into whole numbers of smaller units.

        `1000.5 m` -> [(1000, metre), (50, centimetre)]. Returns None when the
        fraction does not end exactly on the smallest known unit (or the unit
        has no smaller unit), so the caller reads it as a plain decimal.
        """
        parts = []
        while True:
            whole = int(value)
            fraction = value - whole
            if whole:
                parts.append((whole, unit))
            if fraction == 0:
                return parts or [(0, unit)]
            if unit.minor is None:
                return None
            value = fraction * unit.factor
            unit = unit.minor

    @staticmethod
    def _fraction_pieces(match: re.Match, unit: Unit, language: str):
        """`1/2 kg` -> `one half` + `kilogram`, `1,000/2 kg` -> `one thousand halves` + `kilograms`."""
        parsed = FractionProcessor.parse(match)
        if parsed is None:
            return None

        whole, numerator, denominator = parsed
        text = FractionProcessor.to_text(whole, numerator, denominator, language)

        if language == "fa":
            return [(text, unit.fa)]

        # A part of one unit stays singular ("one half kilogram").
        is_singular = whole is None and numerator <= denominator
        return [(text, unit.en if is_singular else unit.en_plural)]