Afraid I can't reproduce this (it works fine for me with Firefox, Safari and Chrome from here - I get as-expected diffs, which also then save appropriately) - what wiki / version of the software was this on, and with what browser?

Chrome, OSX, enwiki. I can reproduce this reliably on my User page on enwiki.

Also repro'd on Chrome/Linux

OK, interesting. This is actually a quite complicated bug, and is not actually correctly-named.

The problem appears to be the '[http://barbarbar bar]' beforehand - if you remove this, the diff magically "works" - BUT it fails to show the '[http://barbarbar bar]' being removed, and indeed puts it after the header too, with another paragraph of 'āæǣ'. The diff is the behaviour that results from the save, even though this isn't a fair representation of the document model (the '[http://barbarbar bar]' is removed from the DM).

RK: this feels like a Parsoid bug (or two?), but might be on our side related to SelSer? - can you confirm before we sling it at them?

This would make a nice automated browser regression test

guanhao7 wrote:

I found it is impossible to use Sogou Input or other Input software to type Chinese with this tool.

You may download a Sogou Input(which is the most popular one in china, on the Windows） to test it.

Here is the download link: http://download.ime.sogou.com/sogou_pinyin_65f.exe?st=lTJ11fWyCsrUkj6oWiGI_Q&e=1359462829&fn=sogou_pinyin_65f.exe

The underlying issue (with Parsoid) has been resolved with the new version of Parsoid finally running on the cluster; marking as FIXED.

[I have forked comment 6 into bug 47436.]