All Downloads are FREE. Search and download functionalities are using the official Maven repository.

kotlin.text.CharJVM.kt Maven / Gradle / Ivy

There is a newer version: 2024.03.6
Show newest version
/*
 * Copyright 2010-2021 JetBrains s.r.o. and Kotlin Programming Language contributors.
 * Use of this source code is governed by the Apache 2.0 license that can be found in the license/LICENSE.txt file.
 */

@file:kotlin.jvm.JvmMultifileClass
@file:kotlin.jvm.JvmName("CharsKt")

package kotlin.text

import java.util.Locale

/**
 * Returns the Unicode general category of this character.
 */
public actual val Char.category: CharCategory
    get() = CharCategory.valueOf(Character.getType(this))

/**
 * Returns `true` if this character (Unicode code point) is defined in Unicode.
 *
 * A character is considered to be defined in Unicode if its [category] is not [CharCategory.UNASSIGNED].
 */
@kotlin.internal.InlineOnly
public actual inline fun Char.isDefined(): Boolean = Character.isDefined(this)

/**
 * Returns `true` if this character is a letter.
 *
 * A character is considered to be a letter if its [category] is [CharCategory.UPPERCASE_LETTER],
 * [CharCategory.LOWERCASE_LETTER], [CharCategory.TITLECASE_LETTER], [CharCategory.MODIFIER_LETTER], or [CharCategory.OTHER_LETTER].
 *
 * @sample samples.text.Chars.isLetter
 */
@kotlin.internal.InlineOnly
public actual inline fun Char.isLetter(): Boolean = Character.isLetter(this)

/**
 * Returns `true` if this character is a letter or digit.
 *
 * @see isLetter
 * @see isDigit
 *
 * @sample samples.text.Chars.isLetterOrDigit
 */
@kotlin.internal.InlineOnly
public actual inline fun Char.isLetterOrDigit(): Boolean = Character.isLetterOrDigit(this)

/**
 * Returns `true` if this character is a digit.
 *
 * A character is considered to be a digit if its [category] is [CharCategory.DECIMAL_DIGIT_NUMBER].
 *
 * @sample samples.text.Chars.isDigit
 */
@kotlin.internal.InlineOnly
public actual inline fun Char.isDigit(): Boolean = Character.isDigit(this)


/**
 * Returns `true` if this character (Unicode code point) should be regarded as an ignorable
 * character in a Java identifier or a Unicode identifier.
 */
@kotlin.internal.InlineOnly
public inline fun Char.isIdentifierIgnorable(): Boolean = Character.isIdentifierIgnorable(this)

/**
 * Returns `true` if this character is an ISO control character.
 *
 * A character is considered to be an ISO control character if its [category] is [CharCategory.CONTROL],
 * meaning the Char is in the range `'\u0000'..'\u001F'` or in the range `'\u007F'..'\u009F'`.
 *
 * @sample samples.text.Chars.isISOControl
 */
@kotlin.internal.InlineOnly
public actual inline fun Char.isISOControl(): Boolean = Character.isISOControl(this)

/**
 * Returns `true` if this  character (Unicode code point) may be part of a Java identifier as other than the first character.
 * @sample samples.text.Chars.isJavaIdentifierPart
 */
@kotlin.internal.InlineOnly
public inline fun Char.isJavaIdentifierPart(): Boolean = Character.isJavaIdentifierPart(this)

/**
 * Returns `true` if this character is permissible as the first character in a Java identifier.
 * @sample samples.text.Chars.isJavaIdentifierStart
 */
@kotlin.internal.InlineOnly
public inline fun Char.isJavaIdentifierStart(): Boolean = Character.isJavaIdentifierStart(this)

/**
 * Determines whether a character is whitespace according to the Unicode standard.
 * Returns `true` if the character is whitespace.
 *
 * @sample samples.text.Chars.isWhitespace
 */
public actual fun Char.isWhitespace(): Boolean = Character.isWhitespace(this) || Character.isSpaceChar(this)

/**
 * Returns `true` if this character is upper case.
 *
 * A character is considered to be an upper case character if its [category] is [CharCategory.UPPERCASE_LETTER],
 * or it has contributory property `Other_Uppercase` as defined by the Unicode Standard.
 *
 * @sample samples.text.Chars.isUpperCase
 */
@kotlin.internal.InlineOnly
public actual inline fun Char.isUpperCase(): Boolean = Character.isUpperCase(this)

/**
 * Returns `true` if this character is lower case.
 *
 * A character is considered to be a lower case character if its [category] is [CharCategory.LOWERCASE_LETTER],
 * or it has contributory property `Other_Lowercase` as defined by the Unicode Standard.
 *
 * @sample samples.text.Chars.isLowerCase
 */
@kotlin.internal.InlineOnly
public actual inline fun Char.isLowerCase(): Boolean = Character.isLowerCase(this)

/**
 * Converts this character to upper case using Unicode mapping rules of the invariant locale.
 */
@Deprecated("Use uppercaseChar() instead.", ReplaceWith("uppercaseChar()"))
@DeprecatedSinceKotlin(warningSince = "1.5")
@kotlin.internal.InlineOnly
public actual inline fun Char.toUpperCase(): Char = uppercaseChar()

/**
 * Converts this character to upper case using Unicode mapping rules of the invariant locale.
 *
 * This function performs one-to-one character mapping.
 * To support one-to-many character mapping use the [uppercase] function.
 * If this character has no mapping equivalent, the character itself is returned.
 *
 * @sample samples.text.Chars.uppercase
 */
@SinceKotlin("1.5")
@WasExperimental(ExperimentalStdlibApi::class)
@kotlin.internal.InlineOnly
public actual inline fun Char.uppercaseChar(): Char = Character.toUpperCase(this)

/**
 * Converts this character to upper case using Unicode mapping rules of the invariant locale.
 *
 * This function supports one-to-many character mapping, thus the length of the returned string can be greater than one.
 * For example, `'\uFB00'.uppercase()` returns `"\u0046\u0046"`,
 * where `'\uFB00'` is the LATIN SMALL LIGATURE FF character (`ff`).
 * If this character has no upper case mapping, the result of `toString()` of this char is returned.
 *
 * @sample samples.text.Chars.uppercase
 */
@SinceKotlin("1.5")
@WasExperimental(ExperimentalStdlibApi::class)
@kotlin.internal.InlineOnly
public actual inline fun Char.uppercase(): String = toString().uppercase()

/**
 * Converts this character to upper case using Unicode mapping rules of the specified [locale].
 *
 * This function supports one-to-many character mapping, thus the length of the returned string can be greater than one.
 * For example, `'\uFB00'.uppercase(Locale.US)` returns `"\u0046\u0046"`,
 * where `'\uFB00'` is the LATIN SMALL LIGATURE FF character (`ff`).
 * If this character has no upper case mapping, the result of `toString()` of this char is returned.
 *
 * @sample samples.text.Chars.uppercaseLocale
 */
@SinceKotlin("1.5")
@WasExperimental(ExperimentalStdlibApi::class)
public fun Char.uppercase(locale: Locale): String = toString().uppercase(locale)

/**
 * Converts this character to lower case using Unicode mapping rules of the invariant locale.
 */
@Deprecated("Use lowercaseChar() instead.", ReplaceWith("lowercaseChar()"))
@DeprecatedSinceKotlin(warningSince = "1.5")
@kotlin.internal.InlineOnly
public actual inline fun Char.toLowerCase(): Char = lowercaseChar()

/**
 * Converts this character to lower case using Unicode mapping rules of the invariant locale.
 *
 * This function performs one-to-one character mapping.
 * To support one-to-many character mapping use the [lowercase] function.
 * If this character has no mapping equivalent, the character itself is returned.
 *
 * @sample samples.text.Chars.lowercase
 */
@SinceKotlin("1.5")
@WasExperimental(ExperimentalStdlibApi::class)
@kotlin.internal.InlineOnly
public actual inline fun Char.lowercaseChar(): Char = Character.toLowerCase(this)

/**
 * Converts this character to lower case using Unicode mapping rules of the invariant locale.
 *
 * This function supports one-to-many character mapping, thus the length of the returned string can be greater than one.
 * For example, `'\u0130'.lowercase()` returns `"\u0069\u0307"`,
 * where `'\u0130'` is the LATIN CAPITAL LETTER I WITH DOT ABOVE character (`İ`).
 * If this character has no lower case mapping, the result of `toString()` of this char is returned.
 *
 * @sample samples.text.Chars.lowercase
 */
@SinceKotlin("1.5")
@WasExperimental(ExperimentalStdlibApi::class)
@kotlin.internal.InlineOnly
public actual inline fun Char.lowercase(): String = toString().lowercase()

/**
 * Converts this character to lower case using Unicode mapping rules of the specified [locale].
 *
 * This function supports one-to-many character mapping, thus the length of the returned string can be greater than one.
 * For example, `'\u0130'.lowercase(Locale.US)` returns `"\u0069\u0307"`,
 * where `'\u0130'` is the LATIN CAPITAL LETTER I WITH DOT ABOVE character (`İ`).
 * If this character has no lower case mapping, the result of `toString()` of this char is returned.
 *
 * @sample samples.text.Chars.lowercaseLocale
 */
@SinceKotlin("1.5")
@WasExperimental(ExperimentalStdlibApi::class)
public fun Char.lowercase(locale: Locale): String = toString().lowercase(locale)

/**
 * Returns `true` if this character is a title case letter.
 *
 * A character is considered to be a title case letter if its [category] is [CharCategory.TITLECASE_LETTER].
 *
 * @sample samples.text.Chars.isTitleCase
 */
@kotlin.internal.InlineOnly
public actual inline fun Char.isTitleCase(): Boolean = Character.isTitleCase(this)

/**
 * Converts this character to title case using Unicode mapping rules of the invariant locale.
 *
 * @see Character.toTitleCase
 */
@Deprecated("Use titlecaseChar() instead.", ReplaceWith("titlecaseChar()"))
@DeprecatedSinceKotlin(warningSince = "1.5")
@kotlin.internal.InlineOnly
public inline fun Char.toTitleCase(): Char = titlecaseChar()

/**
 * Converts this character to title case using Unicode mapping rules of the invariant locale.
 *
 * This function performs one-to-one character mapping.
 * To support one-to-many character mapping use the [titlecase] function.
 * If this character has no mapping equivalent, the result of calling [uppercaseChar] is returned.
 *
 * @sample samples.text.Chars.titlecase
 */
@SinceKotlin("1.5")
@WasExperimental(ExperimentalStdlibApi::class)
@kotlin.internal.InlineOnly
public actual inline fun Char.titlecaseChar(): Char = Character.toTitleCase(this)

/**
 * Converts this character to title case using Unicode mapping rules of the specified [locale].
 *
 * This function supports one-to-many character mapping, thus the length of the returned string can be greater than one.
 * For example, `'\uFB00'.titlecase(Locale.US)` returns `"\u0046\u0066"`,
 * where `'\uFB00'` is the LATIN SMALL LIGATURE FF character (`ff`).
 * If this character has no title case mapping, the result of `uppercase(locale)` is returned instead.
 *
 * @sample samples.text.Chars.titlecaseLocale
 */
@SinceKotlin("1.5")
@WasExperimental(ExperimentalStdlibApi::class)
public fun Char.titlecase(locale: Locale): String {
    val localizedUppercase = uppercase(locale)
    if (localizedUppercase.length > 1) {
        return if (this == '\u0149') localizedUppercase else localizedUppercase[0] + localizedUppercase.substring(1).lowercase()
    }
    if (localizedUppercase != uppercase()) {
        return localizedUppercase
    }
    return titlecaseChar().toString()
}

/**
 * Returns the Unicode directionality property for the given character.
 */
public val Char.directionality: CharDirectionality get() = CharDirectionality.valueOf(Character.getDirectionality(this).toInt())

/**
 * Returns `true` if this character is a Unicode high-surrogate code unit (also known as leading-surrogate code unit).
 */
@kotlin.internal.InlineOnly
public actual inline fun Char.isHighSurrogate(): Boolean = Character.isHighSurrogate(this)

/**
 * Returns `true` if this character is a Unicode low-surrogate code unit (also known as trailing-surrogate code unit).
 */
@kotlin.internal.InlineOnly
public actual inline fun Char.isLowSurrogate(): Boolean = Character.isLowSurrogate(this)

// TODO Provide name for JVM7+
///**
// * Returns the Unicode name of this character, or `null` if the code point of this character is unassigned.
// */
//public fun Char.name(): String? = Character.getName(this.toInt())



internal actual fun digitOf(char: Char, radix: Int): Int = Character.digit(char.code, radix)

/**
 * Checks whether the given [radix] is valid radix for string to number and number to string conversion.
 */
@PublishedApi
internal actual fun checkRadix(radix: Int): Int {
    if (radix !in Character.MIN_RADIX..Character.MAX_RADIX) {
        throw IllegalArgumentException("radix $radix was not in valid range ${Character.MIN_RADIX..Character.MAX_RADIX}")
    }
    return radix
}




© 2015 - 2024 Weber Informatics LLC | Privacy Policy