SUGGESTED FIX

Suggested fix to this problem may and most probably will break some other functionality, so let's keep it conditional.
Essential changes go to src/solaris/native/sun/xawt/XWindow.c, and property handling go somewhere else: I don't know how it should be done properly as code is not our.

------- XWindow.c -------
72a73
> jfieldID composed_ascii_hackID;
941a943,945
>     jint composed_ascii_hack = 0;
>
>     composed_ascii_hack = (*env)->GetIntField(env, peer, composed_ascii_hackID);
944a949,950
>     printf("Entered handleKeyEvent: type=%d, xkeycode=0x%x, xstate=0x%x, keysym=0x%x, composed_ascii_hack=%d\n",
>       event->type, event->xkey.keycode, event->xkey.state, keysym, composed_ascii_hack);
969c975,985
<     keysym = XKeycodeToKeysym(event->xkey.display, event->xkey.keycode, 0);
---
>
>     /* Certain ascii keysyms come with keycode 0 out of IM. An example: dead_acute followed
>      *  by space is apostrophe, and keycode is 0.
>      * In this case, an additional XLookupString or XKeycodeToKeysym call would return zeros.
>      * Conditionally, we'll bypass these calls; it will fix 4799499 but most probably break
>      * some locales with direct input of accented keys. Or doesn't it?
>      */
>     if( !composed_ascii_hack || event->xkey.keycode != 0 ) {
>         keysym = XKeycodeToKeysym(event->xkey.display, event->xkey.keycode, 0);
>     }
>     printf("handleKeyEvent: keysym recalculated:0x%x; remember, NoSymbol is 0x%x\n", keysym, NoSymbol );
997,998c1013,1018
<
<         charcount = XLookupString(&(event->xkey), buffer, bufsize-1, &keysym, &status_in_out);
---
>
>         if( !composed_ascii_hack || event->xkey.keycode != 0 ) {
>             charcount = XLookupString(&(event->xkey), buffer, bufsize-1, &keysym, &status_in_out);
>         }else{
>             charcount = 0;
>         }
1338a1359,1360
>    composed_ascii_hackID =  (*env)->GetFieldID(env, clazz, "composed_ascii_hack", "I");
>



// And here I just get a property value from the System.

------- XWindow.java -------
*** /tmp/sccs.zUxEH9    2004-11-02 13:02:44.000000000 +0300
--- XWindow.java        2004-11-02 13:02:34.000000000 +0300
***************
*** 66,71 ****
--- 66,73 ----
      // fallback default font object
      final static Font defaultFont = new Font(Font.DIALOG, Font.PLAIN, 12);

+     private int composed_ascii_hack;
+

      native int getNativeColor(Color clr, GraphicsConfiguration gc);
      native void getWMInsets(long window, long left, long top, long right, long bottom, long border);
***************
*** 146,151 ****
--- 148,154 ----
              params.putIfNull(BOUNDS, new Rectangle(0, 0, MIN_SIZE, MIN_SIZE));
          }
          params.putIfNull(BORDER_PIXEL, new Long(0));
+         composed_ascii_hack = System.getProperty("sun.awt.composed.ascii") == null ? 0 : 1;
          getColorModel(); // fix 4948833: this call forces the color map to be initialized
          params.putIfNull(COLORMAP, gData.get_awt_cmap());
          params.putIfNull(DEPTH, gData.get_awt_depth());
###@###.### 11/2/04 10:04 GMT
                                     

CONVERTED DATA

BugTraq+ Release Management Values

COMMIT TO FIX:
mustang


                                     

EVALUATION

Commit to fix in Tiger.  
###@###.### 2003-01-06

This rurned out to be much more difficult than expected.  Our key event 
translation model has some serious deficiencies which require a rewrite. 
I added a bunch of debugging code and comments in the XAWT implementation 
(XWindow.c) explaining where the problems lie.  

Also, the work I tried with motif-based AWT and XLookupString was a failure. 
It seems that doing the translations to support lightweight (e.g. swing) 
components and motif components are mutually incompatible (probably since 
motif does its own translation), so we will probably just fix this in XAWT. 

We'll have to put this off until a later release.  
###@###.### 2004-01-08

A roots of our difficulties with this bug (and many similar bugs) are in unnecessary duplication - sometimes even triplication - of some functions belonging to Input Method code. For instance, we try to determine keychar using key event data _after_ IM has it processed; however IM code changes some fields of original event, by convention; as a result, certain keys are swallowed. 
We need to streamline key processing again, in cooperation with IM team. In the meantime, if a fix to this bug is so urgently needed, we can provide yet another patch; as I don't know what exactly will be broken by this patch, I'd propose to make it conditional, a property-driven. We invent a runtime property and if it is set, switch off some redundant checks in our old code.
###@###.### 11/2/04 09:55 GMT

After fix to 4360364, XAWT version should work properly in Mustang.

###@###.### 2005-05-25 10:17:48 GMT