Detect unavailable keys properly. Fixes #592.

[sarg]

xcb:keysyms:event->keysyms returns '((0 . 0)) when key is unavailable.

Using the same check as here to handle this: https://github.com/ch11ng/exwm/blob/9d035d713ee2692ea095dc6e0668ecabeb550845/exwm-input.el#L889-L890

Update: noticed that xcb:keysyms:event->keysyms returns nil when the keycode is not present in the current layout.

[medranocalvo]

Ok, so we were checking in different ways in different places, we now check consistently in both. I like this change, thank you very much. I have to study xcb:keysyms:event->keysyms before merging, will get to it as soon as possible. (We are also using xcb:keysyms:event->keysyms in exwm-xim; should be reviewed as well.)

[medranocalvo]

@Stebalien, @minad: I would appreciate your thoughts on this issue. The question is whether this patch is fixing the symptoms; see below.

The documentation function xcb:keysyms:event->keysyms says that the function returns a list of keysyms or '((0 . 0)).

@sarg notes that:

Update: noticed that xcb:keysyms:event->keysyms returns nil when the keycode is not present in the current layout.

Well, strictly speaking nil is a list. But does it make sense to differentiate between "keysym not recognized" and "no keycode for keysym"? Should both lead to '((0 . 0))?

Below a half-baked patch implementing that, for exposition.

diff --git i/xcb-keysyms.el w/xcb-keysyms.el
index 8999d71..8c0926a 100644
--- i/xcb-keysyms.el
+++ w/xcb-keysyms.el
@@ -683,7 +683,8 @@ (defconst xcb:keysyms:-xf86-keys
               modifiers (apply #'logior modifiers)))
       (let ((keycode (xcb:keysyms:keysym->keycode obj keysym))
             extra-modifiers)
-        (when (/= 0 keycode)
+        (if (zerop keycode)
+            '((0 . 0))
           (setq extra-modifiers (xcb:keysyms:keycode->keysym obj keycode nil)
                 ;; Always try without other modifier.
                 extra-modifiers (append '(0) extra-modifiers)
@@ -695,10 +696,10 @@ (defconst xcb:keysyms:-xf86-keys
                                                obj keycode modifier))
                                          keysym)
                                   modifier))
-                              extra-modifiers))))
-        (mapcar (lambda (extra-modifier)
-                  (cons keysym (logior (or modifiers 0) extra-modifier)))
-                extra-modifiers)))))
+                              extra-modifiers)))
+          (mapcar (lambda (extra-modifier)
+                    (cons keysym (logior (or modifiers 0) extra-modifier)))
+                  extra-modifiers))))))
 
 (cl-defmethod xcb:keysyms:keysym->event ((_obj xcb:connection) keysym
                                          &optional mask allow-modifiers)

[minad]

@medranocalvo I will look at this as soon as I can set aside some time for EXWM.

[Stebalien]

Personally, I'd just return nil in all cases and update the call sites to match. That should make it easier to correctly check for failure.

But that would be a breaking change.

But does it make sense to differentiate between "keysym not recognized" and "no keycode for keysym"?

The only case I can think of is if one wanted to re-implement xdotool and:

1. Try to convert to a keycode.
2. See that there is no keycode for the keysym. The tool should fail here if the keysym isn't recognized.
3. Bind a temporary keycode.
4. Send that temporary keycode.
5. Unbind the temporary keycode.

This, e.g., lets one type emoji.

However, that's a very contrived case.