Currently, we're clearing composition string from editor when WM_IME_COMPOSITION indicates neither committing nor composing. We should do nothing this time. I don't know actual case, though.
Created attachment 584404 [details] [diff] [review] Patch
This patch removes codes that are added at CVS revisions 3.219 and 3.215. http://bonsai.mozilla.org/cvsblame.cgi?file=mozilla/widget/src/windows/nsWindow.cpp&mark=&rev=3.219#3012 Are those comments no longer applied?
Thank you, it's nice information. The change should be for bug 13494. But looks like the change of 3.219 also fixed the bug by another approach. Even if we make our editor's composition string empty, WM_COMPOSITION_END is needed because nsEditor exits from composition mode only when it receives NS_COMPOSITION_END event. So, the change of 3.215 isn't enough for current nsEditor. The change of 3.219 makes the composition string empty before dispatching NS_COMPOSITION_END event. It is right way for current nsEditor. On the other hand, there might still be such IME. And some of them might not send WM_IME_ENDCOMPOSITION even when composition string becomes empty (e.g., keeps composing mode until hitting Enter or ESC). If so, my patch must make regression. But I don't think the current code is using correct way because the block *assumes* the composition string must be empty at the param. I'll post another patch tomorrow.
Created attachment 591709 [details] [diff] [review] Patch Sorry for the delay. I think that this is the right approach. Win7 + ATOK 2011 sends WM_IME_COMPOSITION with 0 for lParam when I delete all composing characters by BS key. When composition string is empty, we should dispatch empty text event. Otherwise, we should dispatch text event normally.
Comment on attachment 591709 [details] [diff] [review] Patch r=me assuming it passes the test.
Thank you, Kimura-san. https://hg.mozilla.org/integration/mozilla-inbound/rev/9ff7a8136813