mirror of
https://github.com/superseriousbusiness/gotosocial.git
synced 2024-11-24 12:46:38 +00:00
fd8a724e77
* bump go swagger version * bump swagger version
713 lines
18 KiB
Go
713 lines
18 KiB
Go
package inflect
|
||
|
||
import (
|
||
"fmt"
|
||
"regexp"
|
||
"strconv"
|
||
"strings"
|
||
"unicode"
|
||
"unicode/utf8"
|
||
)
|
||
|
||
// used by rulesets
|
||
type Rule struct {
|
||
suffix string
|
||
replacement string
|
||
exact bool
|
||
}
|
||
|
||
// a Ruleset is the config of pluralization rules
|
||
// you can extend the rules with the Add* methods
|
||
type Ruleset struct {
|
||
uncountables map[string]bool
|
||
plurals []*Rule
|
||
singulars []*Rule
|
||
humans []*Rule
|
||
acronyms []*Rule
|
||
}
|
||
|
||
// create a blank ruleset. Unless you are going to
|
||
// build your own rules from scratch you probably
|
||
// won't need this and can just use the defaultRuleset
|
||
// via the global inflect.* methods
|
||
func NewRuleset() *Ruleset {
|
||
rs := new(Ruleset)
|
||
rs.uncountables = make(map[string]bool)
|
||
rs.plurals = make([]*Rule, 0)
|
||
rs.singulars = make([]*Rule, 0)
|
||
rs.humans = make([]*Rule, 0)
|
||
rs.acronyms = make([]*Rule, 0)
|
||
return rs
|
||
}
|
||
|
||
// create a new ruleset and load it with the default
|
||
// set of common English pluralization rules
|
||
func NewDefaultRuleset() *Ruleset {
|
||
rs := NewRuleset()
|
||
rs.AddPlural("s", "s")
|
||
rs.AddPlural("testis", "testes")
|
||
rs.AddPlural("axis", "axes")
|
||
rs.AddPlural("octopus", "octopi")
|
||
rs.AddPlural("virus", "viri")
|
||
rs.AddPlural("octopi", "octopi")
|
||
rs.AddPlural("viri", "viri")
|
||
rs.AddPlural("alias", "aliases")
|
||
rs.AddPlural("status", "statuses")
|
||
rs.AddPlural("bus", "buses")
|
||
rs.AddPlural("buffalo", "buffaloes")
|
||
rs.AddPlural("tomato", "tomatoes")
|
||
rs.AddPlural("tum", "ta")
|
||
rs.AddPlural("ium", "ia")
|
||
rs.AddPlural("ta", "ta")
|
||
rs.AddPlural("ia", "ia")
|
||
rs.AddPlural("sis", "ses")
|
||
rs.AddPlural("lf", "lves")
|
||
rs.AddPlural("rf", "rves")
|
||
rs.AddPlural("afe", "aves")
|
||
rs.AddPlural("bfe", "bves")
|
||
rs.AddPlural("cfe", "cves")
|
||
rs.AddPlural("dfe", "dves")
|
||
rs.AddPlural("efe", "eves")
|
||
rs.AddPlural("gfe", "gves")
|
||
rs.AddPlural("hfe", "hves")
|
||
rs.AddPlural("ife", "ives")
|
||
rs.AddPlural("jfe", "jves")
|
||
rs.AddPlural("kfe", "kves")
|
||
rs.AddPlural("lfe", "lves")
|
||
rs.AddPlural("mfe", "mves")
|
||
rs.AddPlural("nfe", "nves")
|
||
rs.AddPlural("ofe", "oves")
|
||
rs.AddPlural("pfe", "pves")
|
||
rs.AddPlural("qfe", "qves")
|
||
rs.AddPlural("rfe", "rves")
|
||
rs.AddPlural("sfe", "sves")
|
||
rs.AddPlural("tfe", "tves")
|
||
rs.AddPlural("ufe", "uves")
|
||
rs.AddPlural("vfe", "vves")
|
||
rs.AddPlural("wfe", "wves")
|
||
rs.AddPlural("xfe", "xves")
|
||
rs.AddPlural("yfe", "yves")
|
||
rs.AddPlural("zfe", "zves")
|
||
rs.AddPlural("hive", "hives")
|
||
rs.AddPlural("quy", "quies")
|
||
rs.AddPlural("by", "bies")
|
||
rs.AddPlural("cy", "cies")
|
||
rs.AddPlural("dy", "dies")
|
||
rs.AddPlural("fy", "fies")
|
||
rs.AddPlural("gy", "gies")
|
||
rs.AddPlural("hy", "hies")
|
||
rs.AddPlural("jy", "jies")
|
||
rs.AddPlural("ky", "kies")
|
||
rs.AddPlural("ly", "lies")
|
||
rs.AddPlural("my", "mies")
|
||
rs.AddPlural("ny", "nies")
|
||
rs.AddPlural("py", "pies")
|
||
rs.AddPlural("qy", "qies")
|
||
rs.AddPlural("ry", "ries")
|
||
rs.AddPlural("sy", "sies")
|
||
rs.AddPlural("ty", "ties")
|
||
rs.AddPlural("vy", "vies")
|
||
rs.AddPlural("wy", "wies")
|
||
rs.AddPlural("xy", "xies")
|
||
rs.AddPlural("zy", "zies")
|
||
rs.AddPlural("x", "xes")
|
||
rs.AddPlural("ch", "ches")
|
||
rs.AddPlural("ss", "sses")
|
||
rs.AddPlural("sh", "shes")
|
||
rs.AddPlural("matrix", "matrices")
|
||
rs.AddPlural("vertix", "vertices")
|
||
rs.AddPlural("indix", "indices")
|
||
rs.AddPlural("matrex", "matrices")
|
||
rs.AddPlural("vertex", "vertices")
|
||
rs.AddPlural("index", "indices")
|
||
rs.AddPlural("mouse", "mice")
|
||
rs.AddPlural("louse", "lice")
|
||
rs.AddPlural("mice", "mice")
|
||
rs.AddPlural("lice", "lice")
|
||
rs.AddPluralExact("ox", "oxen", true)
|
||
rs.AddPluralExact("oxen", "oxen", true)
|
||
rs.AddPluralExact("quiz", "quizzes", true)
|
||
rs.AddSingular("s", "")
|
||
rs.AddSingular("news", "news")
|
||
rs.AddSingular("ta", "tum")
|
||
rs.AddSingular("ia", "ium")
|
||
rs.AddSingular("analyses", "analysis")
|
||
rs.AddSingular("bases", "basis")
|
||
rs.AddSingular("diagnoses", "diagnosis")
|
||
rs.AddSingular("parentheses", "parenthesis")
|
||
rs.AddSingular("prognoses", "prognosis")
|
||
rs.AddSingular("synopses", "synopsis")
|
||
rs.AddSingular("theses", "thesis")
|
||
rs.AddSingular("analyses", "analysis")
|
||
rs.AddSingular("aves", "afe")
|
||
rs.AddSingular("bves", "bfe")
|
||
rs.AddSingular("cves", "cfe")
|
||
rs.AddSingular("dves", "dfe")
|
||
rs.AddSingular("eves", "efe")
|
||
rs.AddSingular("gves", "gfe")
|
||
rs.AddSingular("hves", "hfe")
|
||
rs.AddSingular("ives", "ife")
|
||
rs.AddSingular("jves", "jfe")
|
||
rs.AddSingular("kves", "kfe")
|
||
rs.AddSingular("lves", "lfe")
|
||
rs.AddSingular("mves", "mfe")
|
||
rs.AddSingular("nves", "nfe")
|
||
rs.AddSingular("oves", "ofe")
|
||
rs.AddSingular("pves", "pfe")
|
||
rs.AddSingular("qves", "qfe")
|
||
rs.AddSingular("rves", "rfe")
|
||
rs.AddSingular("sves", "sfe")
|
||
rs.AddSingular("tves", "tfe")
|
||
rs.AddSingular("uves", "ufe")
|
||
rs.AddSingular("vves", "vfe")
|
||
rs.AddSingular("wves", "wfe")
|
||
rs.AddSingular("xves", "xfe")
|
||
rs.AddSingular("yves", "yfe")
|
||
rs.AddSingular("zves", "zfe")
|
||
rs.AddSingular("hives", "hive")
|
||
rs.AddSingular("tives", "tive")
|
||
rs.AddSingular("lves", "lf")
|
||
rs.AddSingular("rves", "rf")
|
||
rs.AddSingular("quies", "quy")
|
||
rs.AddSingular("bies", "by")
|
||
rs.AddSingular("cies", "cy")
|
||
rs.AddSingular("dies", "dy")
|
||
rs.AddSingular("fies", "fy")
|
||
rs.AddSingular("gies", "gy")
|
||
rs.AddSingular("hies", "hy")
|
||
rs.AddSingular("jies", "jy")
|
||
rs.AddSingular("kies", "ky")
|
||
rs.AddSingular("lies", "ly")
|
||
rs.AddSingular("mies", "my")
|
||
rs.AddSingular("nies", "ny")
|
||
rs.AddSingular("pies", "py")
|
||
rs.AddSingular("qies", "qy")
|
||
rs.AddSingular("ries", "ry")
|
||
rs.AddSingular("sies", "sy")
|
||
rs.AddSingular("ties", "ty")
|
||
rs.AddSingular("vies", "vy")
|
||
rs.AddSingular("wies", "wy")
|
||
rs.AddSingular("xies", "xy")
|
||
rs.AddSingular("zies", "zy")
|
||
rs.AddSingular("series", "series")
|
||
rs.AddSingular("movies", "movie")
|
||
rs.AddSingular("xes", "x")
|
||
rs.AddSingular("ches", "ch")
|
||
rs.AddSingular("sses", "ss")
|
||
rs.AddSingular("shes", "sh")
|
||
rs.AddSingular("mice", "mouse")
|
||
rs.AddSingular("lice", "louse")
|
||
rs.AddSingular("buses", "bus")
|
||
rs.AddSingular("oes", "o")
|
||
rs.AddSingular("shoes", "shoe")
|
||
rs.AddSingular("crises", "crisis")
|
||
rs.AddSingular("axes", "axis")
|
||
rs.AddSingular("testes", "testis")
|
||
rs.AddSingular("octopi", "octopus")
|
||
rs.AddSingular("viri", "virus")
|
||
rs.AddSingular("statuses", "status")
|
||
rs.AddSingular("aliases", "alias")
|
||
rs.AddSingularExact("oxen", "ox", true)
|
||
rs.AddSingular("vertices", "vertex")
|
||
rs.AddSingular("indices", "index")
|
||
rs.AddSingular("matrices", "matrix")
|
||
rs.AddSingularExact("quizzes", "quiz", true)
|
||
rs.AddSingular("databases", "database")
|
||
rs.AddIrregular("person", "people")
|
||
rs.AddIrregular("man", "men")
|
||
rs.AddIrregular("child", "children")
|
||
rs.AddIrregular("sex", "sexes")
|
||
rs.AddIrregular("move", "moves")
|
||
rs.AddIrregular("zombie", "zombies")
|
||
rs.AddUncountable("equipment")
|
||
rs.AddUncountable("information")
|
||
rs.AddUncountable("rice")
|
||
rs.AddUncountable("money")
|
||
rs.AddUncountable("species")
|
||
rs.AddUncountable("series")
|
||
rs.AddUncountable("fish")
|
||
rs.AddUncountable("sheep")
|
||
rs.AddUncountable("jeans")
|
||
rs.AddUncountable("police")
|
||
return rs
|
||
}
|
||
|
||
func (rs *Ruleset) Uncountables() map[string]bool {
|
||
return rs.uncountables
|
||
}
|
||
|
||
// add a pluralization rule
|
||
func (rs *Ruleset) AddPlural(suffix, replacement string) {
|
||
rs.AddPluralExact(suffix, replacement, false)
|
||
}
|
||
|
||
// add a pluralization rule with full string match
|
||
func (rs *Ruleset) AddPluralExact(suffix, replacement string, exact bool) {
|
||
// remove uncountable
|
||
delete(rs.uncountables, suffix)
|
||
// create rule
|
||
r := new(Rule)
|
||
r.suffix = suffix
|
||
r.replacement = replacement
|
||
r.exact = exact
|
||
// prepend
|
||
rs.plurals = append([]*Rule{r}, rs.plurals...)
|
||
}
|
||
|
||
// add a singular rule
|
||
func (rs *Ruleset) AddSingular(suffix, replacement string) {
|
||
rs.AddSingularExact(suffix, replacement, false)
|
||
}
|
||
|
||
// same as AddSingular but you can set `exact` to force
|
||
// a full string match
|
||
func (rs *Ruleset) AddSingularExact(suffix, replacement string, exact bool) {
|
||
// remove from uncountable
|
||
delete(rs.uncountables, suffix)
|
||
// create rule
|
||
r := new(Rule)
|
||
r.suffix = suffix
|
||
r.replacement = replacement
|
||
r.exact = exact
|
||
rs.singulars = append([]*Rule{r}, rs.singulars...)
|
||
}
|
||
|
||
// Human rules are applied by humanize to show more friendly
|
||
// versions of words
|
||
func (rs *Ruleset) AddHuman(suffix, replacement string) {
|
||
r := new(Rule)
|
||
r.suffix = suffix
|
||
r.replacement = replacement
|
||
rs.humans = append([]*Rule{r}, rs.humans...)
|
||
}
|
||
|
||
// Add any inconsistent pluralizing/sinularizing rules
|
||
// to the set here.
|
||
func (rs *Ruleset) AddIrregular(singular, plural string) {
|
||
delete(rs.uncountables, singular)
|
||
delete(rs.uncountables, plural)
|
||
rs.AddPlural(singular, plural)
|
||
rs.AddPlural(plural, plural)
|
||
rs.AddSingular(plural, singular)
|
||
}
|
||
|
||
// if you use acronym you may need to add them to the ruleset
|
||
// to prevent Underscored words of things like "HTML" coming out
|
||
// as "h_t_m_l"
|
||
func (rs *Ruleset) AddAcronym(word string) {
|
||
r := new(Rule)
|
||
r.suffix = word
|
||
r.replacement = rs.Titleize(strings.ToLower(word))
|
||
rs.acronyms = append(rs.acronyms, r)
|
||
}
|
||
|
||
// add a word to this ruleset that has the same singular and plural form
|
||
// for example: "rice"
|
||
func (rs *Ruleset) AddUncountable(word string) {
|
||
rs.uncountables[strings.ToLower(word)] = true
|
||
}
|
||
|
||
func (rs *Ruleset) isUncountable(word string) bool {
|
||
// handle multiple words by using the last one
|
||
words := strings.Split(word, " ")
|
||
if _, exists := rs.uncountables[strings.ToLower(words[len(words)-1])]; exists {
|
||
return true
|
||
}
|
||
return false
|
||
}
|
||
|
||
// returns the plural form of a singular word
|
||
func (rs *Ruleset) Pluralize(word string) string {
|
||
if len(word) == 0 {
|
||
return word
|
||
}
|
||
if rs.isUncountable(word) {
|
||
return word
|
||
}
|
||
for _, rule := range rs.plurals {
|
||
if rule.exact {
|
||
if word == rule.suffix {
|
||
return rule.replacement
|
||
}
|
||
} else {
|
||
if strings.HasSuffix(word, rule.suffix) {
|
||
return replaceLast(word, rule.suffix, rule.replacement)
|
||
}
|
||
}
|
||
}
|
||
return word + "s"
|
||
}
|
||
|
||
// returns the singular form of a plural word
|
||
func (rs *Ruleset) Singularize(word string) string {
|
||
if len(word) == 0 {
|
||
return word
|
||
}
|
||
if rs.isUncountable(word) {
|
||
return word
|
||
}
|
||
for _, rule := range rs.singulars {
|
||
if rule.exact {
|
||
if word == rule.suffix {
|
||
return rule.replacement
|
||
}
|
||
} else {
|
||
if strings.HasSuffix(word, rule.suffix) {
|
||
return replaceLast(word, rule.suffix, rule.replacement)
|
||
}
|
||
}
|
||
}
|
||
return word
|
||
}
|
||
|
||
// uppercase first character
|
||
func (rs *Ruleset) Capitalize(word string) string {
|
||
return strings.ToUpper(word[:1]) + word[1:]
|
||
}
|
||
|
||
// "dino_party" -> "DinoParty"
|
||
func (rs *Ruleset) Camelize(word string) string {
|
||
words := splitAtCaseChangeWithTitlecase(word)
|
||
return strings.Join(words, "")
|
||
}
|
||
|
||
// same as Camelcase but with first letter downcased
|
||
func (rs *Ruleset) CamelizeDownFirst(word string) string {
|
||
word = Camelize(word)
|
||
return strings.ToLower(word[:1]) + word[1:]
|
||
}
|
||
|
||
// Captitilize every word in sentance "hello there" -> "Hello There"
|
||
func (rs *Ruleset) Titleize(word string) string {
|
||
words := splitAtCaseChangeWithTitlecase(word)
|
||
return strings.Join(words, " ")
|
||
}
|
||
|
||
func (rs *Ruleset) safeCaseAcronyms(word string) string {
|
||
// convert an acroymn like HTML into Html
|
||
for _, rule := range rs.acronyms {
|
||
word = strings.ReplaceAll(word, rule.suffix, rule.replacement)
|
||
}
|
||
return word
|
||
}
|
||
|
||
func (rs *Ruleset) seperatedWords(word, sep string) string {
|
||
word = rs.safeCaseAcronyms(word)
|
||
words := splitAtCaseChange(word)
|
||
return strings.Join(words, sep)
|
||
}
|
||
|
||
// lowercase underscore version "BigBen" -> "big_ben"
|
||
func (rs *Ruleset) Underscore(word string) string {
|
||
return rs.seperatedWords(word, "_")
|
||
}
|
||
|
||
// First letter of sentance captitilized
|
||
// Uses custom friendly replacements via AddHuman()
|
||
func (rs *Ruleset) Humanize(word string) string {
|
||
word = replaceLast(word, "_id", "") // strip foreign key kinds
|
||
// replace and strings in humans list
|
||
for _, rule := range rs.humans {
|
||
word = strings.ReplaceAll(word, rule.suffix, rule.replacement)
|
||
}
|
||
sentance := rs.seperatedWords(word, " ")
|
||
return strings.ToUpper(sentance[:1]) + sentance[1:]
|
||
}
|
||
|
||
// an underscored foreign key name "Person" -> "person_id"
|
||
func (rs *Ruleset) ForeignKey(word string) string {
|
||
return rs.Underscore(rs.Singularize(word)) + "_id"
|
||
}
|
||
|
||
// a foreign key (with an underscore) "Person" -> "personid"
|
||
func (rs *Ruleset) ForeignKeyCondensed(word string) string {
|
||
return rs.Underscore(word) + "id"
|
||
}
|
||
|
||
// Rails style pluralized table names: "SuperPerson" -> "super_people"
|
||
func (rs *Ruleset) Tableize(word string) string {
|
||
return rs.Pluralize(rs.Underscore(rs.Typeify(word)))
|
||
}
|
||
|
||
var notURLSafe = regexp.MustCompile(`[^\w\d\-_ ]`)
|
||
|
||
// param safe dasherized names like "my-param"
|
||
func (rs *Ruleset) Parameterize(word string) string {
|
||
return ParameterizeJoin(word, "-")
|
||
}
|
||
|
||
// param safe dasherized names with custom separator
|
||
func (rs *Ruleset) ParameterizeJoin(word, sep string) string {
|
||
word = strings.ToLower(word)
|
||
word = rs.Asciify(word)
|
||
word = notURLSafe.ReplaceAllString(word, "")
|
||
word = strings.ReplaceAll(word, " ", sep)
|
||
if len(sep) > 0 {
|
||
squash, err := regexp.Compile(sep + "+")
|
||
if err == nil {
|
||
word = squash.ReplaceAllString(word, sep)
|
||
}
|
||
}
|
||
word = strings.Trim(word, sep+" ")
|
||
return word
|
||
}
|
||
|
||
var lookalikes = map[string]*regexp.Regexp{
|
||
"A": regexp.MustCompile(`À|Á|Â|Ã|Ä|Å`),
|
||
"AE": regexp.MustCompile(`Æ`),
|
||
"C": regexp.MustCompile(`Ç`),
|
||
"E": regexp.MustCompile(`È|É|Ê|Ë`),
|
||
"G": regexp.MustCompile(`Ğ`),
|
||
"I": regexp.MustCompile(`Ì|Í|Î|Ï|İ`),
|
||
"N": regexp.MustCompile(`Ñ`),
|
||
"O": regexp.MustCompile(`Ò|Ó|Ô|Õ|Ö|Ø`),
|
||
"S": regexp.MustCompile(`Ş`),
|
||
"U": regexp.MustCompile(`Ù|Ú|Û|Ü`),
|
||
"Y": regexp.MustCompile(`Ý`),
|
||
"ss": regexp.MustCompile(`ß`),
|
||
"a": regexp.MustCompile(`à|á|â|ã|ä|å`),
|
||
"ae": regexp.MustCompile(`æ`),
|
||
"c": regexp.MustCompile(`ç`),
|
||
"e": regexp.MustCompile(`è|é|ê|ë`),
|
||
"g": regexp.MustCompile(`ğ`),
|
||
"i": regexp.MustCompile(`ì|í|î|ï|ı`),
|
||
"n": regexp.MustCompile(`ñ`),
|
||
"o": regexp.MustCompile(`ò|ó|ô|õ|ö|ø`),
|
||
"s": regexp.MustCompile(`ş`),
|
||
"u": regexp.MustCompile(`ù|ú|û|ü|ũ|ū|ŭ|ů|ű|ų`),
|
||
"y": regexp.MustCompile(`ý|ÿ`),
|
||
}
|
||
|
||
// transforms latin characters like é -> e
|
||
func (rs *Ruleset) Asciify(word string) string {
|
||
for repl, regex := range lookalikes {
|
||
word = regex.ReplaceAllString(word, repl)
|
||
}
|
||
return word
|
||
}
|
||
|
||
var tablePrefix = regexp.MustCompile(`^[^.]*\.`)
|
||
|
||
// "something_like_this" -> "SomethingLikeThis"
|
||
func (rs *Ruleset) Typeify(word string) string {
|
||
word = tablePrefix.ReplaceAllString(word, "")
|
||
return rs.Camelize(rs.Singularize(word))
|
||
}
|
||
|
||
// "SomeText" -> "some-text"
|
||
func (rs *Ruleset) Dasherize(word string) string {
|
||
return rs.seperatedWords(word, "-")
|
||
}
|
||
|
||
// "1031" -> "1031st"
|
||
func (rs *Ruleset) Ordinalize(str string) string {
|
||
number, err := strconv.Atoi(str)
|
||
if err != nil {
|
||
return str
|
||
}
|
||
switch abs(number) % 100 {
|
||
case 11, 12, 13:
|
||
return fmt.Sprintf("%dth", number)
|
||
default:
|
||
switch abs(number) % 10 {
|
||
case 1:
|
||
return fmt.Sprintf("%dst", number)
|
||
case 2:
|
||
return fmt.Sprintf("%dnd", number)
|
||
case 3:
|
||
return fmt.Sprintf("%drd", number)
|
||
}
|
||
}
|
||
return fmt.Sprintf("%dth", number)
|
||
}
|
||
|
||
/////////////////////////////////////////
|
||
// the default global ruleset
|
||
//////////////////////////////////////////
|
||
|
||
var defaultRuleset *Ruleset
|
||
|
||
func init() {
|
||
defaultRuleset = NewDefaultRuleset()
|
||
}
|
||
|
||
func Uncountables() map[string]bool {
|
||
return defaultRuleset.Uncountables()
|
||
}
|
||
|
||
func AddPlural(suffix, replacement string) {
|
||
defaultRuleset.AddPlural(suffix, replacement)
|
||
}
|
||
|
||
func AddSingular(suffix, replacement string) {
|
||
defaultRuleset.AddSingular(suffix, replacement)
|
||
}
|
||
|
||
func AddHuman(suffix, replacement string) {
|
||
defaultRuleset.AddHuman(suffix, replacement)
|
||
}
|
||
|
||
func AddIrregular(singular, plural string) {
|
||
defaultRuleset.AddIrregular(singular, plural)
|
||
}
|
||
|
||
func AddAcronym(word string) {
|
||
defaultRuleset.AddAcronym(word)
|
||
}
|
||
|
||
func AddUncountable(word string) {
|
||
defaultRuleset.AddUncountable(word)
|
||
}
|
||
|
||
func Pluralize(word string) string {
|
||
return defaultRuleset.Pluralize(word)
|
||
}
|
||
|
||
func Singularize(word string) string {
|
||
return defaultRuleset.Singularize(word)
|
||
}
|
||
|
||
func Capitalize(word string) string {
|
||
return defaultRuleset.Capitalize(word)
|
||
}
|
||
|
||
func Camelize(word string) string {
|
||
return defaultRuleset.Camelize(word)
|
||
}
|
||
|
||
func CamelizeDownFirst(word string) string {
|
||
return defaultRuleset.CamelizeDownFirst(word)
|
||
}
|
||
|
||
func Titleize(word string) string {
|
||
return defaultRuleset.Titleize(word)
|
||
}
|
||
|
||
func Underscore(word string) string {
|
||
return defaultRuleset.Underscore(word)
|
||
}
|
||
|
||
func Humanize(word string) string {
|
||
return defaultRuleset.Humanize(word)
|
||
}
|
||
|
||
func ForeignKey(word string) string {
|
||
return defaultRuleset.ForeignKey(word)
|
||
}
|
||
|
||
func ForeignKeyCondensed(word string) string {
|
||
return defaultRuleset.ForeignKeyCondensed(word)
|
||
}
|
||
|
||
func Tableize(word string) string {
|
||
return defaultRuleset.Tableize(word)
|
||
}
|
||
|
||
func Parameterize(word string) string {
|
||
return defaultRuleset.Parameterize(word)
|
||
}
|
||
|
||
func ParameterizeJoin(word, sep string) string {
|
||
return defaultRuleset.ParameterizeJoin(word, sep)
|
||
}
|
||
|
||
func Typeify(word string) string {
|
||
return defaultRuleset.Typeify(word)
|
||
}
|
||
|
||
func Dasherize(word string) string {
|
||
return defaultRuleset.Dasherize(word)
|
||
}
|
||
|
||
func Ordinalize(word string) string {
|
||
return defaultRuleset.Ordinalize(word)
|
||
}
|
||
|
||
func Asciify(word string) string {
|
||
return defaultRuleset.Asciify(word)
|
||
}
|
||
|
||
// helper funcs
|
||
|
||
func reverse(s string) string {
|
||
o := make([]rune, utf8.RuneCountInString(s))
|
||
i := len(o)
|
||
for _, c := range s {
|
||
i--
|
||
o[i] = c
|
||
}
|
||
return string(o)
|
||
}
|
||
|
||
func isSpacerChar(c rune) bool {
|
||
switch {
|
||
case c == '_':
|
||
return true
|
||
case c == ':':
|
||
return true
|
||
case c == '-':
|
||
return true
|
||
case unicode.IsSpace(c):
|
||
return true
|
||
}
|
||
return false
|
||
}
|
||
|
||
func splitAtCaseChange(s string) []string {
|
||
words := make([]string, 0)
|
||
word := make([]rune, 0)
|
||
for _, c := range s {
|
||
spacer := isSpacerChar(c)
|
||
if len(word) > 0 {
|
||
if unicode.IsUpper(c) || spacer {
|
||
words = append(words, string(word))
|
||
word = make([]rune, 0)
|
||
}
|
||
}
|
||
if !spacer {
|
||
word = append(word, unicode.ToLower(c))
|
||
}
|
||
}
|
||
words = append(words, string(word))
|
||
return words
|
||
}
|
||
|
||
func splitAtCaseChangeWithTitlecase(s string) []string {
|
||
words := make([]string, 0)
|
||
word := make([]rune, 0)
|
||
for _, c := range s {
|
||
spacer := isSpacerChar(c)
|
||
if len(word) > 0 {
|
||
if unicode.IsUpper(c) || spacer {
|
||
words = append(words, string(word))
|
||
word = make([]rune, 0)
|
||
}
|
||
}
|
||
if !spacer {
|
||
if len(word) > 0 {
|
||
word = append(word, unicode.ToLower(c))
|
||
} else {
|
||
word = append(word, unicode.ToUpper(c))
|
||
}
|
||
}
|
||
}
|
||
words = append(words, string(word))
|
||
return words
|
||
}
|
||
|
||
func replaceLast(s, match, repl string) string {
|
||
// reverse strings
|
||
srev := reverse(s)
|
||
mrev := reverse(match)
|
||
rrev := reverse(repl)
|
||
// match first and reverse back
|
||
return reverse(strings.Replace(srev, mrev, rrev, 1))
|
||
}
|
||
|
||
func abs(x int) int {
|
||
if x < 0 {
|
||
return -x
|
||
}
|
||
return x
|
||
}
|