Last modified: 2013-06-11 23:37:06 UTC
Round-trip something like '<div></div></div>'. The treebuilder will strip end tags without a matching start tag, which will leave just the end-tag shadowing meta tags. This should allow us to figure out that an end tag was stripped in between. The tsr of the metas should give us the source of this tag.
This is low priority as the semantics (produced HTML DOM) don't change. The dirty diff portion can likely be mitigated with selective serialization.
We round-trip this in unmodified content using selective serialization. In modified content the introduction of a minor and local normalization diff is acceptable in our opinion. Closing as fixed.