Do not ship vendored inflector anymore
This commit is contained in:
parent
71eb4452ac
commit
8874ab5c28
5 changed files with 0 additions and 430 deletions
|
@ -1,7 +0,0 @@
|
||||||
# Library namespace
|
|
||||||
module Inflector
|
|
||||||
end
|
|
||||||
|
|
||||||
require 'inflector/inflections'
|
|
||||||
require 'inflector/defaults'
|
|
||||||
require 'inflector/methods'
|
|
|
@ -1,62 +0,0 @@
|
||||||
module Inflector
|
|
||||||
Inflector::Inflections.instance.instance_eval do
|
|
||||||
plural(/$/, 's')
|
|
||||||
plural(/s$/i, 's')
|
|
||||||
plural(/us$/i, 'i')
|
|
||||||
plural(/(ax|test)is$/i, '\1es')
|
|
||||||
plural(/(octop|vir)us$/i, '\1i')
|
|
||||||
plural(/(octop|vir)i$/i, '\1i')
|
|
||||||
plural(/(alias|status)$/i, '\1es')
|
|
||||||
plural(/(bu)s$/i, '\1ses')
|
|
||||||
plural(/(buffal|tomat)o$/i, '\1oes')
|
|
||||||
plural(/([ti])um$/i, '\1a')
|
|
||||||
plural(/([ti])a$/i, '\1a')
|
|
||||||
plural(/sis$/i, 'ses')
|
|
||||||
plural(/(?:([^f])fe|([lr])f)$/i, '\1\2ves')
|
|
||||||
plural(/(hive)$/i, '\1s')
|
|
||||||
plural(/([^aeiouy]|qu)y$/i, '\1ies')
|
|
||||||
plural(/(x|ch|ss|sh)$/i, '\1es')
|
|
||||||
plural(/(matr|vert|ind)(?:ix|ex)$/i, '\1ices')
|
|
||||||
plural(/([m|l])ouse$/i, '\1ice')
|
|
||||||
plural(/([m|l])ice$/i, '\1ice')
|
|
||||||
plural(/^(ox)$/i, '\1en')
|
|
||||||
plural(/^(oxen)$/i, '\1')
|
|
||||||
plural(/(quiz)$/i, '\1zes')
|
|
||||||
|
|
||||||
singular(/s$/i, '')
|
|
||||||
singular(/i$/i, 'us')
|
|
||||||
singular(/(n)ews$/i, '\1ews')
|
|
||||||
singular(/([ti])a$/i, '\1um')
|
|
||||||
singular(/((a)naly|(b)a|(d)iagno|(p)arenthe|(p)rogno|(s)ynop|(t)he)ses$/i, '\1\2sis')
|
|
||||||
singular(/(^analy)ses$/i, '\1sis')
|
|
||||||
singular(/([^f])ves$/i, '\1fe')
|
|
||||||
singular(/(hive)s$/i, '\1')
|
|
||||||
singular(/(tive)s$/i, '\1')
|
|
||||||
singular(/([lr])ves$/i, '\1f')
|
|
||||||
singular(/([^aeiouy]|qu)ies$/i, '\1y')
|
|
||||||
singular(/(s)eries$/i, '\1eries')
|
|
||||||
singular(/(m)ovies$/i, '\1ovie')
|
|
||||||
singular(/(x|ch|ss|sh)es$/i, '\1')
|
|
||||||
singular(/([m|l])ice$/i, '\1ouse')
|
|
||||||
singular(/(bus)es$/i, '\1')
|
|
||||||
singular(/(o)es$/i, '\1')
|
|
||||||
singular(/(shoe)s$/i, '\1')
|
|
||||||
singular(/(cris|ax|test)es$/i, '\1is')
|
|
||||||
singular(/(octop|vir)i$/i, '\1us')
|
|
||||||
singular(/(alias|status)es$/i, '\1')
|
|
||||||
singular(/^(ox)en/i, '\1')
|
|
||||||
singular(/(vert|ind)ices$/i, '\1ex')
|
|
||||||
singular(/(matr)ices$/i, '\1ix')
|
|
||||||
singular(/(quiz)zes$/i, '\1')
|
|
||||||
singular(/(database)s$/i, '\1')
|
|
||||||
|
|
||||||
irregular('person', 'people')
|
|
||||||
irregular('man', 'men')
|
|
||||||
irregular('child', 'children')
|
|
||||||
irregular('sex', 'sexes')
|
|
||||||
irregular('move', 'moves')
|
|
||||||
irregular('cow', 'kine')
|
|
||||||
|
|
||||||
uncountable(%w(grass equipment information rice money species series fish sheep jeans))
|
|
||||||
end
|
|
||||||
end
|
|
|
@ -1,209 +0,0 @@
|
||||||
module Inflector
|
|
||||||
# A singleton instance of this class is yielded by Inflector.inflections, which can then be used to specify additional
|
|
||||||
# inflection rules. Examples:
|
|
||||||
#
|
|
||||||
# Inflector.inflections do |inflect|
|
|
||||||
# inflect.plural /^(ox)$/i, '\1\2en'
|
|
||||||
# inflect.singular /^(ox)en/i, '\1'
|
|
||||||
#
|
|
||||||
# inflect.irregular 'octopus', 'octopi'
|
|
||||||
#
|
|
||||||
# inflect.uncountable "equipment"
|
|
||||||
# end
|
|
||||||
#
|
|
||||||
# New rules are added at the top. So in the example above, the irregular rule for octopus will now be the first of the
|
|
||||||
# pluralization and singularization rules that is runs. This guarantees that your rules run before any of the rules that may
|
|
||||||
# already have been loaded.
|
|
||||||
class Inflections
|
|
||||||
def self.instance
|
|
||||||
@__instance__ ||= new
|
|
||||||
end
|
|
||||||
|
|
||||||
attr_reader :plurals, :singulars, :uncountables, :humans
|
|
||||||
|
|
||||||
def initialize
|
|
||||||
@plurals, @singulars, @uncountables, @humans = [], [], [], []
|
|
||||||
end
|
|
||||||
|
|
||||||
# Specifies a new pluralization rule and its replacement. The rule can either be a string or a regular expression.
|
|
||||||
# The replacement should always be a string that may include references to the matched data from the rule.
|
|
||||||
def plural(rule, replacement)
|
|
||||||
@uncountables.delete(rule) if rule.is_a?(String)
|
|
||||||
@uncountables.delete(replacement)
|
|
||||||
@plurals.insert(0, [rule, replacement])
|
|
||||||
end
|
|
||||||
|
|
||||||
# Specifies a new singularization rule and its replacement. The rule can either be a string or a regular expression.
|
|
||||||
# The replacement should always be a string that may include references to the matched data from the rule.
|
|
||||||
def singular(rule, replacement)
|
|
||||||
@uncountables.delete(rule) if rule.is_a?(String)
|
|
||||||
@uncountables.delete(replacement)
|
|
||||||
@singulars.insert(0, [rule, replacement])
|
|
||||||
end
|
|
||||||
|
|
||||||
# Specifies a new irregular that applies to both pluralization and singularization at the same time. This can only be used
|
|
||||||
# for strings, not regular expressions. You simply pass the irregular in singular and plural form.
|
|
||||||
#
|
|
||||||
# Examples:
|
|
||||||
# irregular 'octopus', 'octopi'
|
|
||||||
# irregular 'person', 'people'
|
|
||||||
def irregular(singular, plural)
|
|
||||||
@uncountables.delete(singular)
|
|
||||||
@uncountables.delete(plural)
|
|
||||||
if singular[0,1].upcase == plural[0,1].upcase
|
|
||||||
plural(Regexp.new("(#{singular[0,1]})#{singular[1..-1]}$", "i"), '\1' + plural[1..-1])
|
|
||||||
plural(Regexp.new("(#{plural[0,1]})#{plural[1..-1]}$", "i"), '\1' + plural[1..-1])
|
|
||||||
singular(Regexp.new("(#{plural[0,1]})#{plural[1..-1]}$", "i"), '\1' + singular[1..-1])
|
|
||||||
else
|
|
||||||
plural(Regexp.new("#{singular[0,1].upcase}(?i)#{singular[1..-1]}$"), plural[0,1].upcase + plural[1..-1])
|
|
||||||
plural(Regexp.new("#{singular[0,1].downcase}(?i)#{singular[1..-1]}$"), plural[0,1].downcase + plural[1..-1])
|
|
||||||
plural(Regexp.new("#{plural[0,1].upcase}(?i)#{plural[1..-1]}$"), plural[0,1].upcase + plural[1..-1])
|
|
||||||
plural(Regexp.new("#{plural[0,1].downcase}(?i)#{plural[1..-1]}$"), plural[0,1].downcase + plural[1..-1])
|
|
||||||
singular(Regexp.new("#{plural[0,1].upcase}(?i)#{plural[1..-1]}$"), singular[0,1].upcase + singular[1..-1])
|
|
||||||
singular(Regexp.new("#{plural[0,1].downcase}(?i)#{plural[1..-1]}$"), singular[0,1].downcase + singular[1..-1])
|
|
||||||
end
|
|
||||||
end
|
|
||||||
|
|
||||||
# Add uncountable words that shouldn't be attempted inflected.
|
|
||||||
#
|
|
||||||
# Examples:
|
|
||||||
# uncountable "money"
|
|
||||||
# uncountable "money", "information"
|
|
||||||
# uncountable %w( money information rice )
|
|
||||||
def uncountable(*words)
|
|
||||||
(@uncountables << words).flatten!
|
|
||||||
end
|
|
||||||
|
|
||||||
# Specifies a humanized form of a string by a regular expression rule or by a string mapping.
|
|
||||||
# When using a regular expression based replacement, the normal humanize formatting is called after the replacement.
|
|
||||||
# When a string is used, the human form should be specified as desired (example: 'The name', not 'the_name')
|
|
||||||
#
|
|
||||||
# Examples:
|
|
||||||
# human /_cnt$/i, '\1_count'
|
|
||||||
# human "legacy_col_person_name", "Name"
|
|
||||||
def human(rule, replacement)
|
|
||||||
@humans.insert(0, [rule, replacement])
|
|
||||||
end
|
|
||||||
|
|
||||||
# Clears the loaded inflections within a given scope (default is <tt>:all</tt>).
|
|
||||||
# Give the scope as a symbol of the inflection type, the options are: <tt>:plurals</tt>,
|
|
||||||
# <tt>:singulars</tt>, <tt>:uncountables</tt>, <tt>:humans</tt>.
|
|
||||||
#
|
|
||||||
# Examples:
|
|
||||||
# clear :all
|
|
||||||
# clear :plurals
|
|
||||||
def clear(scope = :all)
|
|
||||||
case scope
|
|
||||||
when :all
|
|
||||||
@plurals, @singulars, @uncountables = [], [], []
|
|
||||||
else
|
|
||||||
instance_variable_set "@#{scope}", []
|
|
||||||
end
|
|
||||||
end
|
|
||||||
end
|
|
||||||
|
|
||||||
# Yields a singleton instance of Inflector::Inflections so you can specify additional
|
|
||||||
# inflector rules.
|
|
||||||
#
|
|
||||||
# Example:
|
|
||||||
# Inflector.inflections do |inflect|
|
|
||||||
# inflect.uncountable "rails"
|
|
||||||
# end
|
|
||||||
def inflections
|
|
||||||
if block_given?
|
|
||||||
yield Inflections.instance
|
|
||||||
else
|
|
||||||
Inflections.instance
|
|
||||||
end
|
|
||||||
end
|
|
||||||
|
|
||||||
# Returns the plural form of the word in the string.
|
|
||||||
#
|
|
||||||
# Examples:
|
|
||||||
# "post".pluralize # => "posts"
|
|
||||||
# "octopus".pluralize # => "octopi"
|
|
||||||
# "sheep".pluralize # => "sheep"
|
|
||||||
# "words".pluralize # => "words"
|
|
||||||
# "CamelOctopus".pluralize # => "CamelOctopi"
|
|
||||||
def pluralize(word)
|
|
||||||
result = word.to_s.dup
|
|
||||||
|
|
||||||
if word.empty? || inflections.uncountables.include?(result.downcase)
|
|
||||||
result
|
|
||||||
else
|
|
||||||
inflections.plurals.each { |(rule, replacement)| break if result.gsub!(rule, replacement) }
|
|
||||||
result
|
|
||||||
end
|
|
||||||
end
|
|
||||||
|
|
||||||
# The reverse of +pluralize+, returns the singular form of a word in a string.
|
|
||||||
#
|
|
||||||
# Examples:
|
|
||||||
# "posts".singularize # => "post"
|
|
||||||
# "octopi".singularize # => "octopus"
|
|
||||||
# "sheep".singularize # => "sheep"
|
|
||||||
# "word".singularize # => "word"
|
|
||||||
# "CamelOctopi".singularize # => "CamelOctopus"
|
|
||||||
def singularize(word)
|
|
||||||
result = word.to_s.dup
|
|
||||||
|
|
||||||
if inflections.uncountables.any? { |inflection| result =~ /\b(#{inflection})\Z/i }
|
|
||||||
result
|
|
||||||
else
|
|
||||||
inflections.singulars.each { |(rule, replacement)| break if result.gsub!(rule, replacement) }
|
|
||||||
result
|
|
||||||
end
|
|
||||||
end
|
|
||||||
|
|
||||||
# Capitalizes the first word and turns underscores into spaces and strips a
|
|
||||||
# trailing "_id", if any. Like +titleize+, this is meant for creating pretty output.
|
|
||||||
#
|
|
||||||
# Examples:
|
|
||||||
# "employee_salary" # => "Employee salary"
|
|
||||||
# "author_id" # => "Author"
|
|
||||||
def humanize(lower_case_and_underscored_word)
|
|
||||||
result = lower_case_and_underscored_word.to_s.dup
|
|
||||||
|
|
||||||
inflections.humans.each { |(rule, replacement)| break if result.gsub!(rule, replacement) }
|
|
||||||
result.gsub(/_id$/, "").gsub(/_/, " ").capitalize
|
|
||||||
end
|
|
||||||
|
|
||||||
# Capitalizes all the words and replaces some characters in the string to create
|
|
||||||
# a nicer looking title. +titleize+ is meant for creating pretty output. It is not
|
|
||||||
# used in the Rails internals.
|
|
||||||
#
|
|
||||||
# +titleize+ is also aliased as as +titlecase+.
|
|
||||||
#
|
|
||||||
# Examples:
|
|
||||||
# "man from the boondocks".titleize # => "Man From The Boondocks"
|
|
||||||
# "x-men: the last stand".titleize # => "X Men: The Last Stand"
|
|
||||||
def titleize(word)
|
|
||||||
humanize(underscore(word)).gsub(/\b('?[a-z])/) { $1.capitalize }
|
|
||||||
end
|
|
||||||
|
|
||||||
# Create the name of a table like Rails does for models to table names. This method
|
|
||||||
# uses the +pluralize+ method on the last word in the string.
|
|
||||||
#
|
|
||||||
# Examples
|
|
||||||
# "RawScaledScorer".tableize # => "raw_scaled_scorers"
|
|
||||||
# "egg_and_ham".tableize # => "egg_and_hams"
|
|
||||||
# "fancyCategory".tableize # => "fancy_categories"
|
|
||||||
def tableize(class_name)
|
|
||||||
pluralize(underscore(class_name))
|
|
||||||
end
|
|
||||||
|
|
||||||
# Create a class name from a plural table name like Rails does for table names to models.
|
|
||||||
# Note that this returns a string and not a Class. (To convert to an actual class
|
|
||||||
# follow +classify+ with +constantize+.)
|
|
||||||
#
|
|
||||||
# Examples:
|
|
||||||
# "egg_and_hams".classify # => "EggAndHam"
|
|
||||||
# "posts".classify # => "Post"
|
|
||||||
#
|
|
||||||
# Singular names are not handled correctly:
|
|
||||||
# "business".classify # => "Busines"
|
|
||||||
def classify(table_name)
|
|
||||||
# strip out any leading schema name
|
|
||||||
camelize(singularize(table_name.to_s.sub(/.*\./, '')))
|
|
||||||
end
|
|
||||||
end
|
|
|
@ -1,149 +0,0 @@
|
||||||
# The Inflector transforms words from singular to plural, class names to table names, modularized class names to ones without,
|
|
||||||
# and class names to foreign keys. The default inflections for pluralization, singularization, and uncountable words are kept
|
|
||||||
# in inflections.rb.
|
|
||||||
#
|
|
||||||
# The Rails core team has stated patches for the inflections library will not be accepted
|
|
||||||
# in order to avoid breaking legacy applications which may be relying on errant inflections.
|
|
||||||
# If you discover an incorrect inflection and require it for your application, you'll need
|
|
||||||
# to correct it yourself (explained below).
|
|
||||||
module Inflector
|
|
||||||
extend self
|
|
||||||
|
|
||||||
# By default, +camelize+ converts strings to UpperCamelCase. If the argument to +camelize+
|
|
||||||
# is set to <tt>:lower</tt> then +camelize+ produces lowerCamelCase.
|
|
||||||
#
|
|
||||||
# +camelize+ will also convert '/' to '::' which is useful for converting paths to namespaces.
|
|
||||||
#
|
|
||||||
# Examples:
|
|
||||||
# "active_record".camelize # => "ActiveRecord"
|
|
||||||
# "active_record".camelize(:lower) # => "activeRecord"
|
|
||||||
# "active_record/errors".camelize # => "ActiveRecord::Errors"
|
|
||||||
# "active_record/errors".camelize(:lower) # => "activeRecord::Errors"
|
|
||||||
#
|
|
||||||
# As a rule of thumb you can think of +camelize+ as the inverse of +underscore+,
|
|
||||||
# though there are cases where that does not hold:
|
|
||||||
#
|
|
||||||
# "SSLError".underscore.camelize # => "SslError"
|
|
||||||
def camelize(lower_case_and_underscored_word, first_letter_in_uppercase = true)
|
|
||||||
if first_letter_in_uppercase
|
|
||||||
lower_case_and_underscored_word.to_s.gsub(/\/(.?)/) { "::#{$1.upcase}" }.gsub(/(?:^|_)(.)/) { $1.upcase }
|
|
||||||
else
|
|
||||||
lower_case_and_underscored_word.to_s[0].chr.downcase + camelize(lower_case_and_underscored_word)[1..-1]
|
|
||||||
end
|
|
||||||
end
|
|
||||||
|
|
||||||
# Makes an underscored, lowercase form from the expression in the string.
|
|
||||||
#
|
|
||||||
# Changes '::' to '/' to convert namespaces to paths.
|
|
||||||
#
|
|
||||||
# Examples:
|
|
||||||
# "ActiveRecord".underscore # => "active_record"
|
|
||||||
# "ActiveRecord::Errors".underscore # => active_record/errors
|
|
||||||
#
|
|
||||||
# As a rule of thumb you can think of +underscore+ as the inverse of +camelize+,
|
|
||||||
# though there are cases where that does not hold:
|
|
||||||
#
|
|
||||||
# "SSLError".underscore.camelize # => "SslError"
|
|
||||||
def underscore(camel_cased_word)
|
|
||||||
word = camel_cased_word.to_s.dup
|
|
||||||
word.gsub!(/::/, '/')
|
|
||||||
word.gsub!(/([A-Z]+)([A-Z][a-z])/,'\1_\2')
|
|
||||||
word.gsub!(/([a-z\d])([A-Z])/,'\1_\2')
|
|
||||||
word.tr!("-", "_")
|
|
||||||
word.downcase!
|
|
||||||
word
|
|
||||||
end
|
|
||||||
|
|
||||||
# Replaces underscores with dashes in the string.
|
|
||||||
#
|
|
||||||
# Example:
|
|
||||||
# "puni_puni" # => "puni-puni"
|
|
||||||
def dasherize(underscored_word)
|
|
||||||
underscored_word.gsub(/_/, '-')
|
|
||||||
end
|
|
||||||
|
|
||||||
# Removes the module part from the expression in the string.
|
|
||||||
#
|
|
||||||
# Examples:
|
|
||||||
# "ActiveRecord::CoreExtensions::String::Inflections".demodulize # => "Inflections"
|
|
||||||
# "Inflections".demodulize # => "Inflections"
|
|
||||||
def demodulize(class_name_in_module)
|
|
||||||
class_name_in_module.to_s.gsub(/^.*::/, '')
|
|
||||||
end
|
|
||||||
|
|
||||||
# Creates a foreign key name from a class name.
|
|
||||||
# +separate_class_name_and_id_with_underscore+ sets whether
|
|
||||||
# the method should put '_' between the name and 'id'.
|
|
||||||
#
|
|
||||||
# Examples:
|
|
||||||
# "Message".foreign_key # => "message_id"
|
|
||||||
# "Message".foreign_key(false) # => "messageid"
|
|
||||||
# "Admin::Post".foreign_key # => "post_id"
|
|
||||||
def foreign_key(class_name, separate_class_name_and_id_with_underscore = true)
|
|
||||||
underscore(demodulize(class_name)) + (separate_class_name_and_id_with_underscore ? "_id" : "id")
|
|
||||||
end
|
|
||||||
|
|
||||||
# Ruby 1.9 introduces an inherit argument for Module#const_get and
|
|
||||||
# #const_defined? and changes their default behavior.
|
|
||||||
if Module.method(:const_get).arity == 1
|
|
||||||
# Tries to find a constant with the name specified in the argument string:
|
|
||||||
#
|
|
||||||
# "Module".constantize # => Module
|
|
||||||
# "Test::Unit".constantize # => Test::Unit
|
|
||||||
#
|
|
||||||
# The name is assumed to be the one of a top-level constant, no matter whether
|
|
||||||
# it starts with "::" or not. No lexical context is taken into account:
|
|
||||||
#
|
|
||||||
# C = 'outside'
|
|
||||||
# module M
|
|
||||||
# C = 'inside'
|
|
||||||
# C # => 'inside'
|
|
||||||
# "C".constantize # => 'outside', same as ::C
|
|
||||||
# end
|
|
||||||
#
|
|
||||||
# NameError is raised when the name is not in CamelCase or the constant is
|
|
||||||
# unknown.
|
|
||||||
def constantize(camel_cased_word)
|
|
||||||
names = camel_cased_word.split('::')
|
|
||||||
names.shift if names.empty? || names.first.empty?
|
|
||||||
|
|
||||||
constant = Object
|
|
||||||
names.each do |name|
|
|
||||||
constant = constant.const_defined?(name) ? constant.const_get(name) : constant.const_missing(name)
|
|
||||||
end
|
|
||||||
constant
|
|
||||||
end
|
|
||||||
else
|
|
||||||
def constantize(camel_cased_word) #:nodoc:
|
|
||||||
names = camel_cased_word.split('::')
|
|
||||||
names.shift if names.empty? || names.first.empty?
|
|
||||||
|
|
||||||
constant = Object
|
|
||||||
names.each do |name|
|
|
||||||
constant = constant.const_defined?(name, false) ? constant.const_get(name) : constant.const_missing(name)
|
|
||||||
end
|
|
||||||
constant
|
|
||||||
end
|
|
||||||
end
|
|
||||||
|
|
||||||
# Turns a number into an ordinal string used to denote the position in an
|
|
||||||
# ordered sequence such as 1st, 2nd, 3rd, 4th.
|
|
||||||
#
|
|
||||||
# Examples:
|
|
||||||
# ordinalize(1) # => "1st"
|
|
||||||
# ordinalize(2) # => "2nd"
|
|
||||||
# ordinalize(1002) # => "1002nd"
|
|
||||||
# ordinalize(1003) # => "1003rd"
|
|
||||||
def ordinalize(number)
|
|
||||||
if (11..13).include?(number.to_i % 100)
|
|
||||||
"#{number}th"
|
|
||||||
else
|
|
||||||
case number.to_i % 10
|
|
||||||
when 1; "#{number}st"
|
|
||||||
when 2; "#{number}nd"
|
|
||||||
when 3; "#{number}rd"
|
|
||||||
else "#{number}th"
|
|
||||||
end
|
|
||||||
end
|
|
||||||
end
|
|
||||||
end
|
|
|
@ -1,3 +0,0 @@
|
||||||
module Inflector
|
|
||||||
VERSION = '0.0.1'.freeze
|
|
||||||
end
|
|
Loading…
Reference in a new issue