Imagine you are a happy i3 user and want to also write patches for i3.
You use "Xephyr :1" to get another X11 server and then start your newly
build i3 in it with "DISPLAY=:1 ./i3". You test your changes and
everything seems fine. You are happy. Later that day, you try to log
out, but the $mod+Shift+e key binding from the default config no longer
works. i3-msg cannot connect to the IPC socket because "No such file or
directory". What is going on?
The problem boils down to $I3SOCK having something like two meanings.
When i3 starts, it sets the environment variable $I3SOCK to the path of
its IPC socket. That way, any process started from i3 inherits this and
i3-msg knows how to talk to i3. However, when this variable is already
set when i3 starts, then i3 will replace the existing socket. Thus, in
the earlier experiments, the "separate i3" that was used for
experimenting stole the "main i3"'s socket and replaced it with its own.
When it exited, it deleted that socket.
This commit adds half a work around to this problem: When creating the
IPC socket, i3 will now first try to connect() to the socket. If this
succeeds, it will complain and refuse to use this socket. If the
connect() call fails, it will proceed as usual and create the socket.
Note that trying to connect() to a socket that no process listens on
will fail. Thus, this new code only really "triggers" when some process
is actively listening on this socket and accepting connections.
Example output for when the socket is already in use:
$ I3SOCK=/tmp/sdfdsf DISPLAY=:2 ./i3
31.10.2021 17:03:55 - [libi3] ERROR: Refusing to create UNIX socket at /tmp/sdfdsf: Socket is already in use
31.10.2021 17:03:55 - ERROR: Could not create the IPC socket, IPC disabled
This commit sadly only provides part of the solution. i3 will still
delete the socket when shutting down, even if it failed to create the
IPC socket. Thus, the ipc socket will still break, but now only later.
This will be fixed separately.
First-step-towards-fixing: https://github.com/i3/i3/issues/4381
Signed-off-by: Uli Schlachter <psychon@znc.in>
Including config.h is necessary to get e.g. the _GNU_SOURCE define and
any other definitions that autoconf declares. Hence, config.h needs to
be included as the first header in each file.
This is done either via:
1. Including "common.h" (i3bar)
2. Including "libi3.h"
3. Including "all.h" (i3)
4. Including <config.h> directly
Also remove now-unused I3__FILE__, add copyright/license statement
where missing and switch include/all.h to #pragma once.
This has multiple effects:
1) The i3 codebase is now consistently formatted. clang-format uncovered
plenty of places where inconsistent code made it into our code base.
2) When writing code, you don’t need to think or worry about our coding
style. Write it in yours, then run clang-format-3.5
3) When submitting patches, we don’t need to argue about coding style.
The basic idea is that we don’t want to care about _how_ we write the
code, but _what_ it does :). The coding style that we use is defined in
the .clang-format config file and is based on the google style, but
adapted in such a way that the number of modifications to the i3 code
base is minimal.
Without this fix, children of i3bar would inherit the file descriptor of
the IPC connection to i3. Therefore, even if i3bar exits with SIGSEGV,
the connection to i3 stays open. Because nobody actually reads any
messages by i3, the buffer will fill up and i3 can’t deliver any more
messages, and thus busy-loops at that point.
fixes#995
