mirror of
https://github.com/discourse/discourse.git
synced 2024-11-30 14:46:18 +08:00
ec0212e56b
Admins could specify category slug with upper case characters and same slug, but with different cases could be used simultaneously.
65 lines
1.9 KiB
Ruby
65 lines
1.9 KiB
Ruby
# encoding: utf-8
|
|
# frozen_string_literal: true
|
|
|
|
module Slug
|
|
|
|
CHAR_FILTER_REGEXP = /[:\/\?#\[\]@!\$&'\(\)\*\+,;=_\.~%\\`^\s|\{\}"<>]+/ # :/?#[]@!$&'()*+,;=_.~%\`^|{}"<>
|
|
MAX_LENGTH = 255
|
|
|
|
def self.for(string, default = 'topic', max_length = MAX_LENGTH, method: nil)
|
|
string = string.gsub(/:([\w\-+]+(?::t\d)?):/, '') if string.present? # strip emoji strings
|
|
method = (method || SiteSetting.slug_generation_method || :ascii).to_sym
|
|
max_length = 9999 if method == :encoded # do not truncate encoded slugs
|
|
|
|
slug =
|
|
case method
|
|
when :ascii then self.ascii_generator(string)
|
|
when :encoded then self.encoded_generator(string)
|
|
when :none then self.none_generator(string)
|
|
end
|
|
|
|
slug = self.prettify_slug(slug, max_length: max_length)
|
|
(slug.blank? || slug_is_only_numbers?(slug)) ? default : slug
|
|
end
|
|
|
|
private
|
|
|
|
def self.slug_is_only_numbers?(slug)
|
|
(slug =~ /[^\d]/).blank?
|
|
end
|
|
|
|
def self.prettify_slug(slug, max_length:)
|
|
# Reject slugs that only contain numbers, because they would be indistinguishable from id's.
|
|
slug = (slug_is_only_numbers?(slug) ? '' : slug)
|
|
|
|
slug
|
|
.tr("_", "-")
|
|
.truncate(max_length, omission: '')
|
|
.squeeze('-') # squeeze continuous dashes to prettify slug
|
|
.gsub(/\A-+|-+\z/, '') # remove possible trailing and preceding dashes
|
|
end
|
|
|
|
def self.ascii_generator(string)
|
|
I18n.with_locale(SiteSetting.default_locale) do
|
|
string.tr("'", "").parameterize
|
|
end
|
|
end
|
|
|
|
def self.encoded_generator(string, downcase: true)
|
|
# This generator will sanitize almost all special characters,
|
|
# including reserved characters from RFC3986.
|
|
# See also URI::REGEXP::PATTERN.
|
|
string = string.strip
|
|
.gsub(/\s+/, '-')
|
|
.gsub(CHAR_FILTER_REGEXP, '')
|
|
|
|
string = string.downcase if downcase
|
|
|
|
CGI.escape(string)
|
|
end
|
|
|
|
def self.none_generator(string)
|
|
''
|
|
end
|
|
end
|