Re: [libxml-devel] Probably Memory Leak in XML::Reader#read_string

[Sean Geoghegan]

After some digging I found the memory leak in ruby_xml_reader.c. I've attached a patch for this. It is similar to the patch at http://rubyforge.org/tracker/index.php?func=detail&aid=26014&group_id=494&atid=1971 but also fixes read_inner_xml and read_outer_xml which have the same problem and it prevents an attempt to deallocate NULL pointers.

Thanks

Sean

I couldn't get Rubyforge to behave when creating the patch though so I've just sending it to the list.

memory_leak_patch.diff
Description: Binary data

On 05/06/2009, at 8:42 AM, Sean Geoghegan wrote:

Hi libxml-ruby folk



require 'rubygems'
gem 'libxml-ruby', '>= 1.1.2'
require 'xml/libxml'

def parse()
 x = XML::Reader.string("<foo>this is bar</foo>")
 x.next
 x.read_string  # This line causes the leak
 x.close
end

1.upto(2000000) do |i|
 parse
 GC.start if i % 1000 == 0
end



Versions are:

libxml = 2.6.32
libxml-ruby = 1.1.2
Ruby = 1.8.6 p287





Thanks

Sean Geoghegan

I have some long running scripts that parse lots of XML files over a long period of time using XML:Reader and I've been having what appears to be memory leak problems with them. It seems that the read_string method of XML::Reader is the cuplrit. I've narrowed it down to simple test case. Here is the code: If you comment out the "x.read_string" line everything is fine, it will only use about 20MB of memory, uncomment it however and it will quickly race to 1GB of memory usage. I've tried various ObjectSpace memory inspectors such as Dike but it doesn't show anything that could account for the rapid growth in usage, which makes me think it is a leak in the C extension. I've only tested the above script on OSX but I have experience the leak on various flavours of Linux with Ruby 1.8.5 and 1.8.6. I'll keep investigating and start looking at the C code of libxml- ruby next, but I thought I'd let someone know about it at this stage.

smime.p7s
Description: S/MIME cryptographic signature

_______________________________________________
libxml-devel mailing list
libxml-devel@rubyforge.org
http://rubyforge.org/mailman/listinfo/libxml-devel