See testcase, open it locally, I see these leaks with the nspr log tool and leak gauge: Results of processing log leak1.log : Leaked document at address 30ea400. ... with URI "jar:file:///C:/Documents%20and%20Settings/mw/Bureaublad/moz/firefox20081127/chrome/toolkit.jar!/content/global/bindings/scrollbar.xml". ... with URI "chrome://global/content/bindings/scrollbar.xml". Leaked document at address 30ea800. ... with URI "file:///C:/Documents%20and%20Settings/mw/Bureaublad/notexisting". Leaked content nodes associated with node info manager at address 312a490. ... with document URI "file:///C:/Documents%20and%20Settings/mw/Bureaublad/notexistingbindingleak.htm". Leaked content nodes associated with node info manager at address 3141160. ... with document URI "jar:file:///C:/Documents%20and%20Settings/mw/Bureaublad/moz/firefox20081127/chrome/toolkit.jar!/content/global/bindings/scrollbar.xml". ... with document URI "chrome://global/content/bindings/scrollbar.xml". Leaked content nodes associated with node info manager at address 3141a90. ... with document URI "file:///C:/Documents%20and%20Settings/mw/Bureaublad/notexisting". Summary: Leaked 0 out of 0 DOM Windows Leaked 2 out of 36 documents Leaked 0 out of 5 docshells Leaked content nodes in 3 out of 46 documents It doesn't seem to leak when opened from a network.

Do I need to enable some pref? I get Security Error: Content at file:///home/smaug/mozilla/tests/ may not load data from file:///home/smaug/mozilla/tests/notexisting#x.

Ah, apparently you need to have the security.fileuri.strict_origin_policy pref set to false.

I hate our XBL code.

Comment on attachment 351317 [details] [diff] [review] Fix Should we add these things to the cycle collector as well? If so, would be great with a new bug on that.

I think we have an existing bug on that, actually. The issue here was that we'd stick it in the hashtable too, so we want this patch anyway.

Pushed http://hg.mozilla.org/mozilla-central/rev/89dbc542270b. Do we need a mochitest here? In that case bug 424484 blocks the test. Or can leaks be detected in reftest?

Er, I guess for file:// in our default config we can't use reftest here, so we do need mochitest.

Verified fixed, using: Mozilla/5.0 (Windows; U; Windows NT 5.1; en-US; rv:1.9.2a1pre) Gecko/20081214 Minefield/3.2a1pre Thanks for fixing this, Boris, very useful to me.