This is for Linux only. For Asian input methods, some lM engines use vertical IM candidate window, for example, "Anthy" for Japanese, "SunPinYin" for Chinese. When choosing these IMs, the vertical IM candidate window can obscure the composed text. It seems that the candidate window is aligned to the bottom-left of the application window, instead of aligned to the input caret.
This happens with Java application as well.