ender672/gist:1449283

## gistfile1.rb
require 'nokogiri'

html = '<html><body><br/></body></html>'

# Nokogiri's new HTML encoding detection uses a custom SAX document handler to
# "peek" at an IO before parsing it.
#
# It interrupts the SAX parser by throwing from the context of a SAX document
# handler callback:
# https://github.com/tenderlove/nokogiri/blob/master/lib/nokogiri/html/document.rb#L144
#
# This causes a memory leak since the libxml2 parser does not expect its
# callbacks to longjump. Nokogiri leaks a little bit of memory every time we
# open an HTML document from an IO.

loop do
  doc = Nokogiri::HTML::Document::EncodingReader::SAXHandler.new(:foo)
  prs = Nokogiri::HTML::SAX::Parser.new(doc)
  ctx = Nokogiri::HTML::SAX::ParserContext.memory(html, 'UTF-8')
  catch(:foo) do
    ctx.parse_with(prs)
  end
end

# The above shows what is going on behind the scenes. Here is a much easier way
# to trigger this memory leak:
loop{ Nokogiri::HTML(StringIO.new(html)) }

# The proper fix for this issue is intrusive. I am unsure if we want to
# incorporate it into a stable release. It involves wrapping every rb_funcall in
# xml_sax_parser.c so that it:
#   * intercepts exceptions and throws
#   * stashes the exception in a new C struct associated with the handler or the
#     parser.
#   * tells libxml2 to stop parsing.
#   * re-throws the exception after libxml2 finishes its cleanup.
	require 'nokogiri'

	html = '<html><body><br/></body></html>'

	# Nokogiri's new HTML encoding detection uses a custom SAX document handler to
	# "peek" at an IO before parsing it.
	#
	# It interrupts the SAX parser by throwing from the context of a SAX document
	# handler callback:
	# https://github.com/tenderlove/nokogiri/blob/master/lib/nokogiri/html/document.rb#L144
	#
	# This causes a memory leak since the libxml2 parser does not expect its
	# callbacks to longjump. Nokogiri leaks a little bit of memory every time we
	# open an HTML document from an IO.

	loop do
	doc = Nokogiri::HTML::Document::EncodingReader::SAXHandler.new(:foo)
	prs = Nokogiri::HTML::SAX::Parser.new(doc)
	ctx = Nokogiri::HTML::SAX::ParserContext.memory(html, 'UTF-8')
	catch(:foo) do
	ctx.parse_with(prs)
	end
	end

	# The above shows what is going on behind the scenes. Here is a much easier way
	# to trigger this memory leak:
	loop{ Nokogiri::HTML(StringIO.new(html)) }

	# The proper fix for this issue is intrusive. I am unsure if we want to
	# incorporate it into a stable release. It involves wrapping every rb_funcall in
	# xml_sax_parser.c so that it:
	# * intercepts exceptions and throws
	# * stashes the exception in a new C struct associated with the handler or the
	# parser.
	# * tells libxml2 to stop parsing.
	# * re-throws the exception after libxml2 finishes its cleanup.