mirror of
https://git.uploadfilter24.eu/lerentis/terraform-provider-gitea.git
synced 2024-11-05 10:28:12 +00:00
357 lines
9.6 KiB
Go
357 lines
9.6 KiB
Go
/*
|
|
Copyright 2014 Alexander Okoli
|
|
|
|
Licensed under the Apache License, Version 2.0 (the "License");
|
|
you may not use this file except in compliance with the License.
|
|
You may obtain a copy of the License at
|
|
|
|
http://www.apache.org/licenses/LICENSE-2.0
|
|
|
|
Unless required by applicable law or agreed to in writing, software
|
|
distributed under the License is distributed on an "AS IS" BASIS,
|
|
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
See the License for the specific language governing permissions and
|
|
limitations under the License.
|
|
*/
|
|
|
|
/*
|
|
Package goutils provides utility functions to manipulate strings in various ways.
|
|
The code snippets below show examples of how to use goutils. Some functions return
|
|
errors while others do not, so usage would vary as a result.
|
|
|
|
Example:
|
|
|
|
package main
|
|
|
|
import (
|
|
"fmt"
|
|
"github.com/aokoli/goutils"
|
|
)
|
|
|
|
func main() {
|
|
|
|
// EXAMPLE 1: A goutils function which returns no errors
|
|
fmt.Println (goutils.Initials("John Doe Foo")) // Prints out "JDF"
|
|
|
|
|
|
|
|
// EXAMPLE 2: A goutils function which returns an error
|
|
rand1, err1 := goutils.Random (-1, 0, 0, true, true)
|
|
|
|
if err1 != nil {
|
|
fmt.Println(err1) // Prints out error message because -1 was entered as the first parameter in goutils.Random(...)
|
|
} else {
|
|
fmt.Println(rand1)
|
|
}
|
|
}
|
|
*/
|
|
package goutils
|
|
|
|
import (
|
|
"bytes"
|
|
"strings"
|
|
"unicode"
|
|
)
|
|
|
|
// VERSION indicates the current version of goutils
|
|
const VERSION = "1.0.0"
|
|
|
|
/*
|
|
Wrap wraps a single line of text, identifying words by ' '.
|
|
New lines will be separated by '\n'. Very long words, such as URLs will not be wrapped.
|
|
Leading spaces on a new line are stripped. Trailing spaces are not stripped.
|
|
|
|
Parameters:
|
|
str - the string to be word wrapped
|
|
wrapLength - the column (a column can fit only one character) to wrap the words at, less than 1 is treated as 1
|
|
|
|
Returns:
|
|
a line with newlines inserted
|
|
*/
|
|
func Wrap(str string, wrapLength int) string {
|
|
return WrapCustom(str, wrapLength, "", false)
|
|
}
|
|
|
|
/*
|
|
WrapCustom wraps a single line of text, identifying words by ' '.
|
|
Leading spaces on a new line are stripped. Trailing spaces are not stripped.
|
|
|
|
Parameters:
|
|
str - the string to be word wrapped
|
|
wrapLength - the column number (a column can fit only one character) to wrap the words at, less than 1 is treated as 1
|
|
newLineStr - the string to insert for a new line, "" uses '\n'
|
|
wrapLongWords - true if long words (such as URLs) should be wrapped
|
|
|
|
Returns:
|
|
a line with newlines inserted
|
|
*/
|
|
func WrapCustom(str string, wrapLength int, newLineStr string, wrapLongWords bool) string {
|
|
|
|
if str == "" {
|
|
return ""
|
|
}
|
|
if newLineStr == "" {
|
|
newLineStr = "\n" // TODO Assumes "\n" is seperator. Explore SystemUtils.LINE_SEPARATOR from Apache Commons
|
|
}
|
|
if wrapLength < 1 {
|
|
wrapLength = 1
|
|
}
|
|
|
|
inputLineLength := len(str)
|
|
offset := 0
|
|
|
|
var wrappedLine bytes.Buffer
|
|
|
|
for inputLineLength-offset > wrapLength {
|
|
|
|
if rune(str[offset]) == ' ' {
|
|
offset++
|
|
continue
|
|
}
|
|
|
|
end := wrapLength + offset + 1
|
|
spaceToWrapAt := strings.LastIndex(str[offset:end], " ") + offset
|
|
|
|
if spaceToWrapAt >= offset {
|
|
// normal word (not longer than wrapLength)
|
|
wrappedLine.WriteString(str[offset:spaceToWrapAt])
|
|
wrappedLine.WriteString(newLineStr)
|
|
offset = spaceToWrapAt + 1
|
|
|
|
} else {
|
|
// long word or URL
|
|
if wrapLongWords {
|
|
end := wrapLength + offset
|
|
// long words are wrapped one line at a time
|
|
wrappedLine.WriteString(str[offset:end])
|
|
wrappedLine.WriteString(newLineStr)
|
|
offset += wrapLength
|
|
} else {
|
|
// long words aren't wrapped, just extended beyond limit
|
|
end := wrapLength + offset
|
|
index := strings.IndexRune(str[end:len(str)], ' ')
|
|
if index == -1 {
|
|
wrappedLine.WriteString(str[offset:len(str)])
|
|
offset = inputLineLength
|
|
} else {
|
|
spaceToWrapAt = index + end
|
|
wrappedLine.WriteString(str[offset:spaceToWrapAt])
|
|
wrappedLine.WriteString(newLineStr)
|
|
offset = spaceToWrapAt + 1
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
wrappedLine.WriteString(str[offset:len(str)])
|
|
|
|
return wrappedLine.String()
|
|
|
|
}
|
|
|
|
/*
|
|
Capitalize capitalizes all the delimiter separated words in a string. Only the first letter of each word is changed.
|
|
To convert the rest of each word to lowercase at the same time, use CapitalizeFully(str string, delimiters ...rune).
|
|
The delimiters represent a set of characters understood to separate words. The first string character
|
|
and the first non-delimiter character after a delimiter will be capitalized. A "" input string returns "".
|
|
Capitalization uses the Unicode title case, normally equivalent to upper case.
|
|
|
|
Parameters:
|
|
str - the string to capitalize
|
|
delimiters - set of characters to determine capitalization, exclusion of this parameter means whitespace would be delimeter
|
|
|
|
Returns:
|
|
capitalized string
|
|
*/
|
|
func Capitalize(str string, delimiters ...rune) string {
|
|
|
|
var delimLen int
|
|
|
|
if delimiters == nil {
|
|
delimLen = -1
|
|
} else {
|
|
delimLen = len(delimiters)
|
|
}
|
|
|
|
if str == "" || delimLen == 0 {
|
|
return str
|
|
}
|
|
|
|
buffer := []rune(str)
|
|
capitalizeNext := true
|
|
for i := 0; i < len(buffer); i++ {
|
|
ch := buffer[i]
|
|
if isDelimiter(ch, delimiters...) {
|
|
capitalizeNext = true
|
|
} else if capitalizeNext {
|
|
buffer[i] = unicode.ToTitle(ch)
|
|
capitalizeNext = false
|
|
}
|
|
}
|
|
return string(buffer)
|
|
|
|
}
|
|
|
|
/*
|
|
CapitalizeFully converts all the delimiter separated words in a string into capitalized words, that is each word is made up of a
|
|
titlecase character and then a series of lowercase characters. The delimiters represent a set of characters understood
|
|
to separate words. The first string character and the first non-delimiter character after a delimiter will be capitalized.
|
|
Capitalization uses the Unicode title case, normally equivalent to upper case.
|
|
|
|
Parameters:
|
|
str - the string to capitalize fully
|
|
delimiters - set of characters to determine capitalization, exclusion of this parameter means whitespace would be delimeter
|
|
|
|
Returns:
|
|
capitalized string
|
|
*/
|
|
func CapitalizeFully(str string, delimiters ...rune) string {
|
|
|
|
var delimLen int
|
|
|
|
if delimiters == nil {
|
|
delimLen = -1
|
|
} else {
|
|
delimLen = len(delimiters)
|
|
}
|
|
|
|
if str == "" || delimLen == 0 {
|
|
return str
|
|
}
|
|
str = strings.ToLower(str)
|
|
return Capitalize(str, delimiters...)
|
|
}
|
|
|
|
/*
|
|
Uncapitalize uncapitalizes all the whitespace separated words in a string. Only the first letter of each word is changed.
|
|
The delimiters represent a set of characters understood to separate words. The first string character and the first non-delimiter
|
|
character after a delimiter will be uncapitalized. Whitespace is defined by unicode.IsSpace(char).
|
|
|
|
Parameters:
|
|
str - the string to uncapitalize fully
|
|
delimiters - set of characters to determine capitalization, exclusion of this parameter means whitespace would be delimeter
|
|
|
|
Returns:
|
|
uncapitalized string
|
|
*/
|
|
func Uncapitalize(str string, delimiters ...rune) string {
|
|
|
|
var delimLen int
|
|
|
|
if delimiters == nil {
|
|
delimLen = -1
|
|
} else {
|
|
delimLen = len(delimiters)
|
|
}
|
|
|
|
if str == "" || delimLen == 0 {
|
|
return str
|
|
}
|
|
|
|
buffer := []rune(str)
|
|
uncapitalizeNext := true // TODO Always makes capitalize/un apply to first char.
|
|
for i := 0; i < len(buffer); i++ {
|
|
ch := buffer[i]
|
|
if isDelimiter(ch, delimiters...) {
|
|
uncapitalizeNext = true
|
|
} else if uncapitalizeNext {
|
|
buffer[i] = unicode.ToLower(ch)
|
|
uncapitalizeNext = false
|
|
}
|
|
}
|
|
return string(buffer)
|
|
}
|
|
|
|
/*
|
|
SwapCase swaps the case of a string using a word based algorithm.
|
|
|
|
Conversion algorithm:
|
|
|
|
Upper case character converts to Lower case
|
|
Title case character converts to Lower case
|
|
Lower case character after Whitespace or at start converts to Title case
|
|
Other Lower case character converts to Upper case
|
|
Whitespace is defined by unicode.IsSpace(char).
|
|
|
|
Parameters:
|
|
str - the string to swap case
|
|
|
|
Returns:
|
|
the changed string
|
|
*/
|
|
func SwapCase(str string) string {
|
|
if str == "" {
|
|
return str
|
|
}
|
|
buffer := []rune(str)
|
|
|
|
whitespace := true
|
|
|
|
for i := 0; i < len(buffer); i++ {
|
|
ch := buffer[i]
|
|
if unicode.IsUpper(ch) {
|
|
buffer[i] = unicode.ToLower(ch)
|
|
whitespace = false
|
|
} else if unicode.IsTitle(ch) {
|
|
buffer[i] = unicode.ToLower(ch)
|
|
whitespace = false
|
|
} else if unicode.IsLower(ch) {
|
|
if whitespace {
|
|
buffer[i] = unicode.ToTitle(ch)
|
|
whitespace = false
|
|
} else {
|
|
buffer[i] = unicode.ToUpper(ch)
|
|
}
|
|
} else {
|
|
whitespace = unicode.IsSpace(ch)
|
|
}
|
|
}
|
|
return string(buffer)
|
|
}
|
|
|
|
/*
|
|
Initials extracts the initial letters from each word in the string. The first letter of the string and all first
|
|
letters after the defined delimiters are returned as a new string. Their case is not changed. If the delimiters
|
|
parameter is excluded, then Whitespace is used. Whitespace is defined by unicode.IsSpacea(char). An empty delimiter array returns an empty string.
|
|
|
|
Parameters:
|
|
str - the string to get initials from
|
|
delimiters - set of characters to determine words, exclusion of this parameter means whitespace would be delimeter
|
|
Returns:
|
|
string of initial letters
|
|
*/
|
|
func Initials(str string, delimiters ...rune) string {
|
|
if str == "" {
|
|
return str
|
|
}
|
|
if delimiters != nil && len(delimiters) == 0 {
|
|
return ""
|
|
}
|
|
strLen := len(str)
|
|
var buf bytes.Buffer
|
|
lastWasGap := true
|
|
for i := 0; i < strLen; i++ {
|
|
ch := rune(str[i])
|
|
|
|
if isDelimiter(ch, delimiters...) {
|
|
lastWasGap = true
|
|
} else if lastWasGap {
|
|
buf.WriteRune(ch)
|
|
lastWasGap = false
|
|
}
|
|
}
|
|
return buf.String()
|
|
}
|
|
|
|
// private function (lower case func name)
|
|
func isDelimiter(ch rune, delimiters ...rune) bool {
|
|
if delimiters == nil {
|
|
return unicode.IsSpace(ch)
|
|
}
|
|
for _, delimiter := range delimiters {
|
|
if ch == delimiter {
|
|
return true
|
|
}
|
|
}
|
|
return false
|
|
}
|