We can reduce the WMNotification API surface area further by getting rid of
WMCreateNotification and WMPostNotification. WMNotification remains a
first-class object, but it is no longer possible for client code to create a
notification object directly. Notifications must now be posted through
WMPostNotificationName, which handles notification creation and destruction on
its own.
This will simplify the notification lifecycle and make the Rust rewrite
simpler. (Notifications no longer need to be reference-counted, heap-allocated
objects that might be saved somewhere after they are dispatched.)
WTextField code which reused the WMNotification struct has been modified to take
parameters of the correct type directly, instead of through a WMNotification's
void* data field.
This reduces the notification API in a way that is helpful for rewriting in
Rust. This function is only used in one place, and the object that is being
deregistered is free'd immediately after WMRemoveNotificationObserverWithName is
called, so it should be safe just to use WMRemoveNotificationObserver (since I
hope it's an error to keep a notification to a free'd object registered).
This appears to have been used by now-defunct support for network
connections (WMConnection). No live code instantiates a notification queue or
pushes/dequeues notifications from a notification queue. (The global
NotificationCenter in WINGs/notification.c is still in use, so it is not going
anywhere in this commit.)
`WMRetainNotification`, `WMGetDefaultNotificationQueue`,
`WMDequeueNotificationMatching`, `WMEnqueueNotification`, and
`WMEnqueueCoalesceNotification`, are only used in WINGs/notification.c. To
reduce the API surface area that needs to be migrated to Rust, they are being
made private.
We should eventually get rid of this entirely, in favor of something along the
lines of a sorted Vec that is fully typed (so the payload isn't just a void*).
This is a bit of a red herring, since WMTree is only used in wmmenugen, which I
don't think I've ever directly used. See notes at the top of tree.rs for more
musings on whether this should be in wutil-rs at all.
WMTreeWalk is only called once, with depthFirst set to true, so we might as well
just hard-code that behavior.
Also, this parameter appears to have been misnamed (since the search is always DFS)
and should properly be named to indicate that it controls if this is a pre- or
post-order traversal.
This is a first step towards migrating WINGs/tree.c to Rust.
Happy finding: no parent pointer are actually needed, so we don't have to use a
shared structure for node pointers!
These functions were added to glibc 2.38, so we don't even need to check for
libbsd. There isn't a strong need to worry about supporting older systems
because use of these functions should go away as we rewrite string.c in
Rust. And, apparently, they are not held in high regard historically. That's at
least 2 reasons to be rid of them.
