[Xcb] second attempt of the keybinding

[Jamey Sharp]

On Sat, 2005-05-07 at 08:15 +0200, Vincent Torri wrote:
> Here is my second attempt. If you think it looks good, i can ad it into
> the cvs

The xcb-util module is all highly experimental stuff, at least for the
moment. Commit anything you want to it; that's what we gave you commit
rights for. :-)

That said, I'd encourage a couple more little adjustments, which I'll
list here.

Currently keybinding.h declares enum tag_t and struct _XCBKeySymbols.
I'd move both declarations to the .c file, as no caller of this library
needs them. (The "typedef struct _XCBKeySymbols XCBKeySymbols;"
obviously still needs to be in the .h file.)

The Alloc function should have a corresponding Free function.

The XCBKeySymbolsGetReply isn't needed now, at least not as a public
function. It may make sense to just turn it into a static function and
call it from the other public functions though.

The trouble with that is what to do with the XCBGenericError return; but
the answer is that the only error the server is specified to return is
BadValue, and that only if we query the mapping of a keycode outside the
range specified in the server connection setup data. As long as we just
don't do that, all errors may just as well be reported straight to the
application; we won't be interested in trapping any of them. So you
should ignore X errors everywhere in this code.

Functions like XCBKeySymbolsGetKeysym shouldn't fail when handed a
KeySymbols object with a cookie in it; instead they should convert the
cookie to a reply and process that.

Finally, XCBRefreshKeyboardMapping won't work as coded. It should
contain code more like this:

	if(syms->tag == TAG_VALUE)
	{
		free(syms->u.reply);
		syms->tag = TAG_COOKIE;
		syms->u.cookie = /* same as in Alloc function */
	}

That's how you transition back into having a cookie instead of a reply.

That's it! Your code looks good. Like I said, commit whenever you
want. :-)

--Jamey