Created attachment 48724 [details] patch fix was partly explained in https://bugs.webkit.org/show_bug.cgi?id=15914#c58 comment from Kalle Vahlman Hovewer returning TextBreakDone instead of textBreakFirst was leading to a freeze on new lines. Attached patch fixing problem for me.
Created attachment 48725 [details] patch with a cnp fix in changelog
Created attachment 48726 [details] one more problem in changelog
Comment on attachment 48726 [details] one more problem in changelog Can we test this via a layout test? The Gtk port also has unit tests, but layout tests would be preferred. Something with a simple <input> field which used eventSender to do some input? Here are some extra docs: http://trac.webkit.org/wiki/Writing%20Layout%20Tests%20for%20DumpRenderTree http://webkit.org/quality/testwriting.html
glib unicode is not a default backend for webkit-gtk, so that seems will be tricky to test.
Comment on attachment 48726 [details] one more problem in changelog This fix doesn't look right. Why is UBRK_CHARACTER special here? Isn't returning TextBreakDone, simply exposing a different bug, maybe? Also, the ChangeLog needs some reasoning for the change, not just the boilerplate.
CCing Carlos Garcia Campos, who knows quite a bit about the GLib unicode backend.
We have removed the Glib unicode backend.