summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorMatthias Clasen <mclasen@redhat.com>2020-04-02 17:40:44 -0400
committerMatthias Clasen <mclasen@redhat.com>2020-04-02 17:43:14 -0400
commitcffa45d5ff97b3b6107bb9d563a84a529014342a (patch)
tree46374e83cf20937de4247fb607b4706de3172bb3
parent14166b09b5d62ed7a8b78179d6fe655250b4083a (diff)
downloadgtk+-cffa45d5ff97b3b6107bb9d563a84a529014342a.tar.gz
wayland: Add a comment about keyboard state handling
Clarify a point that took me a few hours to fully track down, so lets preserve what I found for the next poor sould coming this way.
-rw-r--r--gdk/wayland/gdkdevice-wayland.c15
1 files changed, 15 insertions, 0 deletions
diff --git a/gdk/wayland/gdkdevice-wayland.c b/gdk/wayland/gdkdevice-wayland.c
index 0e1ac1d283..dcb38c2953 100644
--- a/gdk/wayland/gdkdevice-wayland.c
+++ b/gdk/wayland/gdkdevice-wayland.c
@@ -2160,6 +2160,21 @@ keyboard_handle_modifiers (void *data,
direction = gdk_keymap_get_direction (keymap);
xkb_state = _gdk_wayland_keymap_get_xkb_state (keymap);
+ /* Note: the docs for xkb_state_update mask state that all parameters
+ * must be passed, or we may end up with an 'incoherent' state. But the
+ * Wayland modifiers event only includes a single group field, so we
+ * can't pass depressed/latched/locked groups.
+ *
+ * We assume that the compositor is sending us the 'effective' group
+ * (the protocol is not clear on that point), and pass it as the depressed
+ * group - we are basically pretending that the user holds down a key for
+ * this group at all times.
+ *
+ * This means that our xkb_state would answer a few questions differently
+ * from the compositors xkb_state - e.g. if you asked it about the latched
+ * group. But nobody is asking it those questions, so it does not really
+ * matter. We hope.
+ */
xkb_state_update_mask (xkb_state, mods_depressed, mods_latched, mods_locked, group, 0, 0);
seat->key_modifiers = gdk_keymap_get_modifier_state (keymap);