module Hpricot
# Once you've matched a list of elements, you will often need to handle them as
# a group. Or you may want to perform the same action on each of them.
# Hpricot::Elements is an extension of Ruby's array class, with some methods
# added for altering elements contained in the array.
#
# If you need to create an element array from regular elements:
#
# Hpricot::Elements[ele1, ele2, ele3]
#
# Assuming that ele1, ele2 and ele3 contain element objects (Hpricot::Elem,
# Hpricot::Doc, etc.)
#
# == Continuing Searches
#
# Usually the Hpricot::Elements you're working on comes from a search you've
# done. Well, you can continue searching the list by using the same <tt>at</tt>
# and <tt>search</tt> methods you can use on plain elements.
#
# elements = doc.search("/div/p")
# elements = elements.search("/a[@href='http://hoodwink.d/']")
# elements = elements.at("img")
#
# == Altering Elements
#
# When you're altering elements in the list, your changes will be reflected in
# the document you started searching from.
#
# doc = Hpricot("That's my <b>spoon</b>, Tyler.")
# doc.at("b").swap("<i>fork</i>")
# doc.to_html
# #=> "That's my <i>fork</i>, Tyler."
#
# == Getting More Detailed
#
# If you can't find a method here that does what you need, you may need to
# loop through the elements and find a method in Hpricot::Container::Trav
# which can do what you need.
#
# For example, you may want to search for all the H3 header tags in a document
# and grab all the tags underneath the header, but not inside the header.
# A good method for this is <tt>next_sibling</tt>:
#
# doc.search("h3").each do |h3|
# while ele = h3.next_sibling
# ary << ele # stuff away all the elements under the h3
# end
# end
#
# Most of the useful element methods are in the mixins Hpricot::Traverse
# and Hpricot::Container::Trav.
class Elements < Array
# Searches this list for any elements (or children of these elements) matching
# the CSS or XPath expression +expr+. Root is assumed to be the element scanned.
#
# See Hpricot::Container::Trav.search for more.
def search(*expr,&blk)
Elements[*map { |x| x.search(*expr,&blk) }.flatten.uniq]
end
alias_method :/, :search
# Searches this list for the first element (or child of these elements) matching
# the CSS or XPath expression +expr+. Root is assumed to be the element scanned.
#
# See Hpricot::Container::Trav.at for more.
def at(expr, &blk)
if expr.kind_of? Fixnum
super
else
search(expr, &blk)[0]
end
end
alias_method :%, :at
# Convert this group of elements into a complete HTML fragment, returned as a
# string.
def to_html
map { |x| x.output("") }.join
end
alias_method :to_s, :to_html
# Returns an HTML fragment built of the contents of each element in this list.
#
# If a HTML +string+ is supplied, this method acts like inner_html=.
def inner_html(*string)
if string.empty?
map { |x| x.inner_html }.join
else
x = self.inner_html = string.pop || x
end
end
alias_method :html, :inner_html
alias_method :innerHTML, :inner_html
# Replaces the contents of each element in this list. Supply an HTML +string+,
# which is loaded into Hpricot objects and inserted into every element in this
# list.
def inner_html=(string)
each { |x| x.inner_html = string }
end
alias_method :html=, :inner_html=
alias_method :innerHTML=, :inner_html=
# Returns an string containing the text contents of each element in this list.
# All HTML tags are removed.
def inner_text
map { |x| x.inner_text }.join
end
alias_method :text, :inner_text
# Remove all elements in this list from the document which contains them.
#
# doc = Hpricot("<html>Remove this: <b>here</b></html>")
# doc.search("b").remove
# doc.to_html
# => "<html>Remove this: </html>"
#
def remove
each { |x| x.parent.children.delete(x) }
end
# Empty the elements in this list, by removing their insides.
#
# doc = Hpricot("<p> We have <i>so much</i> to say.</p>")
# doc.search("i").empty
# doc.to_html
# => "<p> We have <i></i> to say.</p>"
#
def empty
each { |x| x.inner_html = nil }
end
# Add to the end of the contents inside each element in this list.
# Pass in an HTML +str+, which is turned into Hpricot elements.
def append(str = nil, &blk)
each { |x| x.html(x.children + x.make(str, &blk)) }
end
# Add to the start of the contents inside each element in this list.
# Pass in an HTML +str+, which is turned into Hpricot elements.
def prepend(str = nil, &blk)
each { |x| x.html(x.make(str, &blk) + x.children) }
end
# Add some HTML just previous to each element in this list.
# Pass in an HTML +str+, which is turned into Hpricot elements.
def before(str = nil, &blk)
each { |x| x.parent.insert_before x.make(str, &blk), x }
end
# Just after each element in this list, add some HTML.
# Pass in an HTML +str+, which is turned into Hpricot elements.
def after(str = nil, &blk)
each { |x| x.parent.insert_after x.make(str, &blk), x }
end
# Wraps each element in the list inside the element created by HTML +str+.
# If more than one element is found in the string, Hpricot locates the
# deepest spot inside the first element.
#
# doc.search("a[@href]").
# wrap(%{<div class="link"><div class="link_inner"></div></div>})
#
# This code wraps every link on the page inside a +div.link+ and a +div.link_inner+ nest.
def wrap(str = nil, &blk)
each do |x|
wrap = x.make(str, &blk)
nest = wrap.detect { |w| w.respond_to? :children }
unless nest
raise "No wrapping element found."
end
x.parent.replace_child(x, wrap)
nest = nest.children.first until nest.empty?
nest.html([x])
end
end
# Gets and sets attributes on all matched elements.
#
# Pass in a +key+ on its own and this method will return the string value
# assigned to that attribute for the first elements. Or +nil+ if the
# attribute isn't found.
#
# doc.search("a").attr("href")
# #=> "http://hacketyhack.net/"
#
# Or, pass in a +key+ and +value+. This will set an attribute for all
# matched elements.
#
# doc.search("p").attr("class", "basic")
#
# You may also use a Hash to set a series of attributes:
#
# (doc/"a").attr(:class => "basic", :href => "http://hackety.org/")
#
# Lastly, a block can be used to rewrite an attribute based on the element
# it belongs to. The block will pass in an element. Return from the block
# the new value of the attribute.
#
# records.attr("href") { |e| e['href'] + "#top" }
#
# This example adds a <tt>#top</tt> anchor to each link.
#
def attr key, value = nil, &blk
if value or blk
each do |el|
el.set_attribute(key, value || blk[el])
end
return self
end
if key.is_a? Hash
key.each { |k,v| self.attr(k,v) }
return self
else
return self[0].get_attribute(key)
end
end
alias_method :set, :attr
# Adds the class to all matched elements.
#
# (doc/"p").add_class("bacon")
#
# Now all paragraphs will have class="bacon".
def add_class class_name
each do |el|
next unless el.respond_to? :get_attribute
classes = el.get_attribute('class').to_s.split(" ")
el.set_attribute('class', classes.push(class_name).uniq.join(" "))
end
self
end
# Remove an attribute from each of the matched elements.
#
# (doc/"input").remove_attr("disabled")
#
def remove_attr name
each do |el|
next unless el.respond_to? :remove_attribute
el.remove_attribute(name)
end
self
end
# Removes a class from all matched elements.
#
# (doc/"span").remove_class("lightgrey")
#
# Or, to remove all classes:
#
# (doc/"span").remove_class
#
def remove_class name = nil
each do |el|
next unless el.respond_to? :get_attribute
if name
classes = el.get_attribute('class').to_s.split(" ")
el.set_attribute('class', (classes - [name]).uniq.join(" "))
else
el.remove_attribute("class")
end
end
self
end
ATTR_RE = %r!\[ *(?:(@)([\w\(\)-]+)|([\w\(\)-]+\(\))) *([~\!\|\*$\^=]*) *'?"?([^'"]*)'?"? *\]!i # " (for emacs)
BRACK_RE = %r!(\[) *([^\]]*) *\]+!i
FUNC_RE = %r!(:)?([a-zA-Z0-9\*_-]*)\( *[\"']?([^ \)]*?)['\"]? *\)!
CUST_RE = %r!(:)([a-zA-Z0-9\*_-]*)()!
CATCH_RE = %r!([:\.#]*)([a-zA-Z0-9\*_-]+)!
def self.filter(nodes, expr, truth = true)
until expr.empty?
_, *m = *expr.match(/^(?:#{ATTR_RE}|#{BRACK_RE}|#{FUNC_RE}|#{CUST_RE}|#{CATCH_RE})/)
break unless _
expr = $'
m.compact!
if m[0] == '@'
m[0] = "@#{m.slice!(2,1).join}"
end
if m[0] == '[' && m[1] =~ /^\d+$/
m = [":", "nth", m[1].to_i-1]
end
if m[0] == ":" && m[1] == "not"
nodes, = Elements.filter(nodes, m[2], false)
elsif "#{m[0]}#{m[1]}" =~ /^(:even|:odd)$/
new_nodes = []
nodes.each_with_index {|n,i| new_nodes.push(n) if (i % 2 == (m[1] == "even" ? 0 : 1)) }
nodes = new_nodes
elsif "#{m[0]}#{m[1]}" =~ /^(:first|:last)$/
nodes = [nodes.send(m[1])]
else
meth = "filter[#{m[0]}#{m[1]}]" unless m[0].empty?
if meth and Traverse.method_defined? meth
args = m[2..-1]
else
meth = "filter[#{m[0]}]"
if Traverse.method_defined? meth
args = m[1..-1]
end
end
args << -1
nodes = Elements[*nodes.find_all do |x|
args[-1] += 1
x.send(meth, *args) ? truth : !truth
end]
end
end
[nodes, expr]
end
# Given two elements, attempt to gather an Elements array of everything between
# (and including) those two elements.
def self.expand(ele1, ele2, excl=false)
ary = []
offset = excl ? -1 : 0
if ele1 and ele2
# let's quickly take care of siblings
if ele1.parent == ele2.parent
ary = ele1.parent.children[ele1.node_position..(ele2.node_position+offset)]
else
# find common parent
p, ele1_p = ele1, [ele1]
ele1_p.unshift p while p.respond_to?(:parent) and p = p.parent
p, ele2_p = ele2, [ele2]
ele2_p.unshift p while p.respond_to?(:parent) and p = p.parent
common_parent = ele1_p.zip(ele2_p).select { |p1, p2| p1 == p2 }.flatten.last
child = nil
if ele1 == common_parent
child = ele2
elsif ele2 == common_parent
child = ele1
end
if child
ary = common_parent.children[0..(child.node_position+offset)]
end
end
end
return Elements[*ary]
end
def filter(expr)
nodes, = Elements.filter(self, expr)
nodes
end
def not(expr)
if expr.is_a? Traverse
nodes = self - [expr]
else
nodes, = Elements.filter(self, expr, false)
end
nodes
end
private
def copy_node(node, l)
l.instance_variables.each do |iv|
node.instance_variable_set(iv, l.instance_variable_get(iv))
end
end
end
module Traverse
def self.filter(tok, &blk)
define_method("filter[#{tok.is_a?(String) ? tok : tok.inspect}]", &blk)
end
filter '' do |name,i|
name == '*' || (self.respond_to?(:name) && self.name.downcase == name.downcase)
end
filter '#' do |id,i|
self.elem? and get_attribute('id').to_s == id
end
filter '.' do |name,i|
self.elem? and classes.include? name
end
filter :lt do |num,i|
self.position < num.to_i
end
filter :gt do |num,i|
self.position > num.to_i
end
nth = proc { |num,i| self.position == num.to_i }
nth_first = proc { |*a| self.position == 0 }
nth_last = proc { |*a| self == parent.children_of_type(self.name).last }
filter :nth, &nth
filter :eq, &nth
filter ":nth-of-type", &nth
filter :first, &nth_first
filter ":first-of-type", &nth_first
filter :last, &nth_last
filter ":last-of-type", &nth_last
filter :even do |num,i|
self.position % 2 == 0
end
filter :odd do |num,i|
self.position % 2 == 1
end
filter ':first-child' do |i|
self == parent.containers.first
end
filter ':nth-child' do |arg,i|
case arg
when 'even'; (parent.containers.index(self) + 1) % 2 == 0
when 'odd'; (parent.containers.index(self) + 1) % 2 == 1
else self == (parent.containers[arg.to_i - 1])
end
end
filter ":last-child" do |i|
self == parent.containers.last
end
filter ":nth-last-child" do |arg,i|
self == parent.containers[-1-arg.to_i]
end
filter ":nth-last-of-type" do |arg,i|
self == parent.children_of_type(self.name)[-1-arg.to_i]
end
filter ":only-of-type" do |arg,i|
parent.children_of_type(self.name).length == 1
end
filter ":only-child" do |arg,i|
parent.containers.length == 1
end
filter :parent do |*a|
containers.length > 0
end
filter :empty do |*a|
elem? && inner_html.length == 0
end
filter :root do |*a|
self.is_a? Hpricot::Doc
end
filter 'text' do |*a|
self.text?
end
filter 'comment' do |*a|
self.comment?
end
filter :contains do |arg, ignore|
html.include? arg
end
pred_procs =
{'text()' => proc { |ele, *_| ele.inner_text.strip },
'@' => proc { |ele, attr, *_| ele.get_attribute(attr).to_s if ele.elem? }}
oper_procs =
{'=' => proc { |a,b| a == b },
'!=' => proc { |a,b| a != b },
'~=' => proc { |a,b| a.split(/\s+/).include?(b) },
'|=' => proc { |a,b| a =~ /^#{Regexp::quote b}(-|$)/ },
'^=' => proc { |a,b| a.index(b) == 0 },
'$=' => proc { |a,b| a =~ /#{Regexp::quote b}$/ },
'*=' => proc { |a,b| idx = a.index(b) }}
pred_procs.each do |pred_n, pred_f|
oper_procs.each do |oper_n, oper_f|
filter "#{pred_n}#{oper_n}" do |*a|
qual = pred_f[self, *a]
oper_f[qual, a[-2]] if qual
end
end
end
filter 'text()' do |val,i|
self.children.grep(Hpricot::Text).detect { |x| x.content =~ /\S/ } if self.children
end
filter '@' do |attr,val,i|
self.elem? and has_attribute? attr
end
filter '[' do |val,i|
self.elem? and search(val).length > 0
end
end
end