| Commit message (Collapse) | Author | Age | Files | Lines |
... | |
|
|
|
|
|
|
|
|
|
| |
This big patch replaces our custom main loop with the event based glib
main loop. This has several advantages: we can make all the tiny code
bits in the main loop independent from each others, we can add
additional file descriptors for polling (e.g. the mpdclient socket).
We don't need the ncurses timeout() anymore, because glib will poll
stdin for us.
|
|
|
|
|
|
| |
Fix several segmentation faults: when the connection to the MPD server
is lost, there were NULL pointer dereferences because
client->status==NULL. Check before accessing it.
|
|
|
|
|
|
| |
Move all command handlers which need a connection to the MPD server to
screen_client_cmd(). This will allow us to unify error handling
later.
|
|
|
|
|
|
|
|
| |
Display the UI always, and show connection errors there. This fixes a
lot of consistency problems: error messages used to be printed to
stderr, but afterwards, the atexit() handlers deinitialized ncurses
and the screen was cleared. This patch is still problematic, but
constitutes a good step forward.
|
|
|
|
|
|
| |
Create generic utility functions for doint I/O: two wait functions,
and mpd_recv() fills the input buffer. These functions are used in
mpd_newConnection() and mpd_getNextReturnElement().
|
|
|
|
|
| |
Even for large responses, 16kB should be enough. There is no
performance gain for larger buffers, even if MPD is local.
|
|
|
|
|
|
|
| |
512 characters should be well enough for error messages which are
actually being displayed somewhere. Use sizeof(errorStr) instead of
MPD_BUFFER_MAX_LENGTH everywhere. Do proper length checking when
copying the output buffer to errorStr.
|
|
|
|
|
|
|
| |
In general, don't treat the input buffer as a null-terminated string:
don't append '\0', don't use strchr() and strtok(). To delete
consumed portions of the buffer, strcpy() is bad anyway, because it
does not allow overlapping buffers.
|
|
|
|
|
| |
Why bother duplicating the buffer, when we can simply parse the
response directly from the input buffer?
|
|
|
|
|
|
| |
When working with buffer lengths and recv() results, we should declare
size_t and ssize_t variables, otherwise the code may fail on 64 bit
machines.
|
|
|
|
|
| |
Declaring empty methods is of no use. If a method is empty, we should
set it to NULL in the screen_functions struct.
|
|
|
|
|
| |
Since the screen's initialization is guaranteed, we can remove the
NULL check in center_playing_item().
|
|
|
|
|
|
| |
Unfortunately when I fixed the GCC warnings and refactored lots of the
code, the "artist" screen was disabled. This patch applies all API
changes and fixes all warnings.
|
|
|
|
| |
Follow the same code style als MPD itself.
|
|
|
|
|
| |
When deselecting a song in the file browser, the local variable "idx"
was initialized and overwritten. Remove the initialization.
|
|
|
|
|
|
|
|
|
|
| |
playlist_remove_reuse() removes a song from the playlist, but does not
free it.
Use this function in mpdclient_cmd_delete() to fix a segmentation
fault: the song pointer was still used after it was freed, by passing
it to mpdclient_playlist_callback(). Free the song manually now.
Reorganize and simply some code in mpdclient_cmd_delete().
|
| |
|
| |
|
|
|
|
|
| |
Since the list_window code sets highlight to 0 before it calls the
list_callback, the callback itself doesn't have to reset the flag.
|
|
|
|
|
| |
current.lines is always initialized, and cannot be NULL, no need to
check it.
|
| |
|
|
|
|
|
| |
Why bother counting the elements of help_text[] (with sentinel) at
runtime, when the compiler can do it at compile time?
|
|
|
|
|
| |
lyrics_text_rows wasn't set anyway, I somehow removed the code in a
patch some time ago.
|
|
|
|
|
| |
Merge code which used to be duplicated in the screens "play", "help",
"lyrics".
|
|
|
|
|
|
| |
It is similar to list_window_cmd(), but it scrolls the window instead
of moving the selection. It is used by the "lyrics" and the "help"
screen.
|
| |
|
|
|
|
|
| |
When the selected screen does not exist, return before calling
close().
|
|
|
|
|
| |
Since the number of screens is known at compile time, define a macro
which calculates this, instead of having a sentinel element.
|
|
|
|
|
| |
Screens which are not active have already been closed, or have never
been opened. Don't call close again on them.
|
|
|
|
|
|
| |
There can only be onle screen struct instance at a time. Declare it
as a static struct, instead of a pointer to the struct. This
simplifies all accesses, and avoids the dynamic allocation.
|
| |
|
|
|
|
|
| |
At startup, mode_fn is initialized with &screen_playlist, and it will
never be reset to NULL. Remove all checks.
|
|
|
|
|
| |
That seems to be a fn->paint was checked, then the code called
fn->update(). Fix this.
|
|
|
|
|
|
| |
If it is really important to set the curser, we should think of a
better way, instead of exporting one non-generic method. For now,
just set the cursor to the origin.
|
|
|
|
|
|
| |
Compile time initialization saves resources, compared to run-time
initialization. Declare all screen_function structs as global
variables, and remove all get functions.
|
|
|
|
| |
Follow the same code style als MPD itself.
|
|
|
|
|
|
| |
Instead of freeing the playlist with mpdclient_playlist_free(), call
playlist_clear() to empty the song list. This fixes a segmentation
fault which occured when you cleared the playlist.
|
|
|
|
|
| |
Due to a missing refresh() call, ncmpc painted the screen after one
second. Add refresh(), make ncmpc display immediately.
|
|
|
|
|
| |
Let us declare struct names in addition to typedef names, so we can
forward-declare them.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
In-process plugins are very problematic. It is much easier and
flexible to move the lyrics plugins to external programs, with a
trivial protocol. This is work in progress, among the things missing:
- protocol specification, including exit codes
- plugin installation
- plugin search directory
- run-time configuration (currently hard coded)
- automatic polling (using glib's main loop?)
- better and more robust error handling
|
| |
|
|
|
|
|
|
| |
Added functions like playlist_length(), playlist_get(),
playlist_replace(), playlist_remove(). Don't access the
mpdclient_playlist struct directly.
|
|
|
|
|
|
| |
Moved code from mpdclient_new() and mpdclient_playlist_free(). In
mpdclient_disconnect(), call mpdclient_playlist_clear() instead of
mpdclient_playlist_free() (which is now called in mpdclient_free()).
|
|
|
|
|
|
|
|
|
|
| |
Move everything which manipulates the mpdclient_playlist struct to
playlist.c. Many of the functions get a mpdclient pointer instead of
a playlist; this will be changed later.
The functions mpdclient_playlist_update() and
mpdclient_playlist_update_changes() stay in mpdclient.c for now, since
they are tightly connected to the client code.
|
|
|
|
|
|
| |
Since we compile ncmpc with glib, we can use its slice allocator for
efficient song allocation. I have added a way to disable this with a
macro.
|
|
|
|
|
| |
Use a forward declared struct mpd_song instead of the typedef
mpd_Song.
|
|
|
|
|
| |
Move everything which manipulates the mpd_Song struct to song.c.
Thousands of lines of too much for libmpdclient.c!
|
|
|
|
|
|
|
| |
To relax the include dependencies, give all structs real names, in
addition to the typedef names. This way, other headers can
forward-declare them, and do not need to include mpdclient.h. In the
future, the typedefs will be removed.
|
|
|
|
| |
Pass const pointers whenever possible.
|