In polybar-msg, the connection handle was captured as a reference to a
shared_ptr, but the shared_ptr went out of scope right after.
This probably always caused UB, but was only noticeable for two or more
instances.
* Return shared_ptr from eventloop
* Add -Wdeprecated-copy-dtor warning
Produces a warning if classes don't have explicit copy operations if
they have a user-defined constructor.
This helps us stick to the rule of 5 (kinda, no warnings for missing
move operators).
* Clean up eventloop
* Fix compiler warnings
* Fix fs_event_handle_t name
If two users start a bar with IPC and don't have XDG_RUNTIME_DIR set,
polybar will create the fallback directory `/tmp/polybar`.
However, that directory is only accessible by the user that created it
and so polybar running under the second user will fail to open its
socket there.
We add the UID to the fallback directory to prevent this.
Fixes#2683
If two processes call `mkdir` at the same time, the second one will fail
and wrongly assume the folder wasn't created properly.
We now first check if the folder really doesn't exist and we also catch
any IPC initialization errors and disable IPC instead of crashing the
whole bar.
* eventloop: Use eventloop namespace in cpp files
* changelog: Add missing deprecated hook message
* Make eventloop and ipc classes non-copyable and non-movable
* Remove functional.hpp
* eventloop: Don't close handles in error cases
Client should be responsible for closing handles.
* eventloop: Address invalidation of handle references
Deprecates not using `polybar-msg` for IPC.
Fixes#2532Closes#2465Fixes#2504
* Create FIFO specific NamedPipeHandle subclass to PipeHandle
* Prototype SocketHandle
* Move mainloop up to main.cpp
* Pass eventloop to ipc class
* Deprecate sending ipc over the named pipe
Unfortunately, we can only show the warning in the polybar log and not
give the user any feedback because the pipe is one-way
* Move eventloop into its own namespace
* Prototype ipc socket handling
* Remove handles from ipc_client
Should be independent from eventloop logic
* Remove ipc clients when finished
* Add tests for ipc_client decoding
* Add callback for complete ipc messages
* Remove template param from mixins
* Move signal handler to new callback system
* Move poll handle to new callback system
* Move FSEventHandle to new callback system
* Move TimerHandle and AsyncHandle to new callback system
* Move PipeHandle to new callback system
* Implement socket functionality in new callback system
* Correctly reset ipc named pipe handle
* Let client close handles in error callback
* Wrap client pipe and ipc::client in connection class
* Better decoder log messages
* Socket path logic
* Fix CI warnings
* Remove UVHandleGeneric
* Fix error when socket folder already exists
* Proof of concept message writeback
* Restructure ipc files
* polybar-msg: Use sockets
* polybar-msg: Better syntax for actions
* Fix memory leak with fifo
After EOF, the pipe wasn't closed and EOF was called all the time, each
time allocating a new pipe.
* Make polybar-msg compile on its own
* Rudimentary writeback for polybar-msg
* Fix payload reference going out of scope.
* Add IPC documentation
* Cleanup polybar-msg code
* Specify the v0 ipc message format
* Close ipc connection after message
* Fix ipc tests
* Properly close ipc connections
* Fix polybar-msg not working with action string
* Write polybar-msg manpage
* polybar-msg: Stop using exit()
* ipc: Print log message with PID
* Add tests for ipc util
* polybar-msg: Print PID with success message
* ipc: Propagate message errors
* Rename ipc::client to ipc::decoder
* Rename ipc.cpp to polybar-msg.cpp
* ipc: Write encoder function and fix decoder bugs
* ipc: Use message format for responses
* ipc: Handle wrong message types
* ipc: Write back error message if ipc message cannot be processed
This only happens for commands and empty actions.
Non-empty actions are not immediately executed, but deferred until the
next loop iteration.
* Remove TODO about deleting runtime directory
The socket file is not deleted after socket.close() is called, only
after libuv executes the close callback.
So we can't just call rmdir because it will probably always fail.
* CLeanup WriteRequest
* Update manpage authors
* Cleanup
