I got the following bug report.  I've included the text of
the report more or less as sent.
-------------------------------------------------------------
Forward:
we fixed a problem with the French keyboard layout, and when I discovered
the problem, it also turned out to be the cause of two other problems:

  Keypad '+' was coming up as '='
  Caps lock was not working properly.

>   In ...src/unix/sun/canvas.c XK_KP_3 is being mapped to the character
>'3'. Somewhere Java detects if a French keyboard is being used. If so,
>it maps keypad 3 (i. e. XK_KP_3) to the character comma (or single quote?)
>which shares the key with the integer 3. (In American keyboards the key
>for the integer 3 is shared by the pound # character)
>   So we have a key mapping confusion in Java for the French keyboard.

N/A

We need a French keyboard for further evaluation.

  xxxxx@xxxxx   1998-09-24

A French keyboard is not required for this one. Simply do "export LANG=fr"
and then run the testcase (or any Java TextField/Area), and press "+" on the
Keypad and you'll the "=" appears in the textfield instead. The bug happens on
1.1.7 and 1.2, only on Solaris.

  xxxxx@xxxxx   1998-10-07

In fact, this bug is not specific to the French locale. In the default (English)
locale the "+" on the keyboard still gives a "=", and the "*" key gives a "8" when NumLock is on.

  xxxxx@xxxxx   1998-10-08

This bug verified as fixed in jdk118f via regression testing.  However, the test fails when running under JavaTest2.0 which means the test is not up to spec.  The test has no pass/fail buttons yet requires user interaction.
  xxxxx@xxxxx   1999-01-25

When the regression test was written it was thought it was applicable to only a French locale.  This is why the "ignore" keyword was used in the header.  As this test is actually potentially useful for any locale, I will replace the ignore so the test will be run by the harness and pass/fail buttons will be displayed. 

  xxxxx@xxxxx   1999-01-25


Note that there is a report of this bug filed against win32.  
This is probably a different bug which was fixed separately 
some time ago.  I tested this bug against 1.2.2-G on WinNT 4.0 SP3, 
and the bug did not exist there.  
  xxxxx@xxxxx   1999-02-16

Ported 1.1.8 fix to 1.2.2.  
Added regression test that can be run in the harness.  
See "Suggested Fix" section.  
  xxxxx@xxxxx   1999-02-16