Reported by jteh on 2011-03-23 22:05
Some languages contain characters which do not have a single unicode character representation. Instead, pressing such a key produces a sequence of multiple unicode characters. This sequence looks like a single character and behaves like a single character in some respects; i.e. left/right arrow keys move the cursor past the entire compound character. However, the input actually contains several unicode characters.

Str:

1. Install the Tamil keyboard layout.
2. Open a Wordpad document, switch to the Tamil keyboard layout and enable speak typed characters.
3. Press "shift+.".
4. Expected: NVDA should speak one character. (Tamil: "sri")
5. Actual: NVDA individually speaks the four characters ("ஸ்ரீ") composing this compound character. (Tamil: "sa otru ra ii")

Other letters in Tamil like this:
||= Letter =||= Key =||
|| த்ர || shift+6 ||
|| க்ஷ || shift+7 ||
|| ஷ்ர || shift+8 ||
|| ஞ்ச || shift+ஜ ||
|| ந்த || shift+த ||
|| ஃப || shift+ப ||

I haven't checked this yet, but I'm pretty sure Windows actually sends separate WM_CHAR messages for these characters; that is the only way it could work, since WM_CHAR only handles one character. Unless there's something I don't know about which indicates whether a character is a compound character and when it ends, this is going to be difficult to fix. We can't use WM_KEYUP to determine when a character ends, as this would mean we'd have to wait for the key to be released to have it announced. Using a timer would incur delay in speaking of typed characters, which is unacceptable.
Blocking #4487