More experimental hackery for `win-dead-keys'. Works for me as well as it did

in the one case I can test, but I suspect this isn't the whole story. Bletch.

git-svn-id: svn://svn.tartarus.org/sgt/putty@3986 cda61777-01e9-0310-a592-d414129be87e

diff --git a/window.c b/window.c
index 14bedd8..fd977ed 100644 (file)
--- a/window.c
+++ b/window.c
@@ -3170,8 +3170,8 @@ static int TranslateKey(UINT message, WPARAM wParam, LPARAM lParam,
 
     HKL kbd_layout = GetKeyboardLayout(0);
 
-    /* keys is for ToAsciiEx; XXX do we know how big this needs to be? */
-    static BYTE keys[3];
+    /* keys is for ToAsciiEx. There's some ick here, see below. */
+    static WORD keys[3];
     static int compose_char = 0;
     static WPARAM compose_key = 0;
 
@@ -3862,12 +3862,27 @@ static int TranslateKey(UINT message, WPARAM wParam, LPARAM lParam,
            keystate[VK_CAPITAL] = 0;
        }
 
-       /* 'keys' parameter is declared in MSDN documentation as
-        * 'LPWORD lpChar'. In 0.54 and below we took that to mean that
-        * 'keys' should be an array of WORD, but an array of BYTE works
-        * better on keyboard with dead keys, at least for Win2K/US-
-        * International and WinXP/German. Bletch. */
-       r = ToAsciiEx(wParam, scan, keystate, (LPWORD)keys, 0, kbd_layout);
+       /* XXX how do we know what the max size of the keys array should
+        * be is? There's indication on MS' website of an Inquire/InquireEx
+        * functioning returning a KBINFO structure which tells us. */
+       if (osVersion.dwPlatformId == VER_PLATFORM_WIN32_NT) {
+           /* XXX 'keys' parameter is declared in MSDN documentation as
+            * 'LPWORD lpChar'.
+            * The experience of a French user indicates that on
+            * Win98, WORD[] should be passed in, but on Win2K, it should
+            * be BYTE[]. German WinXP and my Win2K with "US International"
+            * driver corroborate this.
+            * Experimentally I've conditionalised the behaviour on the
+            * Win9x/NT split, but I suspect it's worse than that.
+            * See wishlist item `win-dead-keys' for more horrible detail
+            * and speculations. */
+           BYTE keybs[3];
+           int i;
+           r = ToAsciiEx(wParam, scan, keystate, (LPWORD)keybs, 0, kbd_layout);
+           for (i=0; i<3; i++) keys[i] = keybs[i];
+       } else {
+           r = ToAsciiEx(wParam, scan, keystate, keys, 0, kbd_layout);
+       }
 #ifdef SHOW_TOASCII_RESULT
        if (r == 1 && !key_down) {
            if (alt_sum) {