Created attachment 163862 [details]
Tiny sample program reproducing the bug

Environment: Windows 7, using the Windows Look and Feel.

Scenario:
* Use the Windows Look and Feel in your Java application (with Metal: no problem).
* Create a ETable with a column header having characters in Chinese or Japanese (or any not supported by Tahoma font).
* When running the program, click or this header to sort the column => these characters will be replaced by a square.
(small Maven project attached, with 1 class and one main method)

Root cause:
The headers of the sorted columns will be rendered in bold by the ETable, which is done in "ETableHeader::getTableCellRendererComponent". This code takes the font of the JLabel used for the rendering, and creates a similar font in bold. For that, it does not uses "Font::deriveFont(...)" (because of a 2005 bug with Apple's JVM), but uses "(new Font (label.getFont().getName(), ...)".
Under Windows it is a problem:
* the default font (used here by the JLabel) is a composite font
* from what I understand, a composite font is a list of physical fonts, which will be successively used to try to render a given character.
* this list of the physical fonts contains one that supports Chinese characters.
* label.getFont().getName() retrieves the name of the first font of the composite font, in the case of the Windows Look and Feel it is "Tahoma" (which does not support Chinese).
* so the "new Font(...)" used as a header renderer will be only "Tahoma", hence the square characters.

Does this analysis makes sense for you?

My suggestion would be to revert the patch from 2005. I don't have a Mac to my disposition to test that its works though.

Old code:
// don't use deriveFont() - see #49973 for details
label.setFont (new Font (label.getFont ().getName (), Font.BOLD, label.getFont ().getSize ()));

New code:
label.setFont(label.getFont().deriveFont(Font.BOLD));