These signals are meant to send to a process group, instead of a single
process: SIGINT is usually emitted by the terminal and sent to the process
group. SIGHUP usually happens to a process group if termination of a process
causes that process group to become orphaned.
Before this patch, chg will only forward these signals to the single server
process. This patch changes it to the server process group.
This will allow us to properly kill processes started by the forked server
process, like a ssh process. The behavior difference can be observed by
setting SSH_ASKPASS to a dummy script doing "sleep 100" and then run
"chg push ssh://dest-need-password-auth". Before this patch, the first Ctrl+C
will kill the hg process while ssh-askpass and ssh will remain alive. This
patch will make sure they are killed properly.
We recently discovered a case in production that chg uses 100% CPU and is
trying to read data forever:
recvfrom(4, "", 1814012019, 0, NULL, NULL) = 0
Using gdb, apparently readchannel() got wrong data. It was reading in an
infinite loop because rsize == 0 does not exit the loop, while the server
process had ended.
(gdb) bt
#0 ... in recv () at /lib64/libc.so.6
#1 ... in readchannel (...) at /usr/include/bits/socket2.h:45
#2 ... in readchannel (hgc=...) at hgclient.c:129
#3 ... in handleresponse (hgc=...) at hgclient.c:255
#4 ... in hgc_runcommand (hgc=..., args=<optimized>, argsize=<optimized>)
#5 ... in main (argc=...486922636, argv=..., envp=...) at chg.c:661
(gdb) frame 2
(gdb) p *hgc
$1 = {sockfd = 4, pid = 381152, ctx = {ch = 108 'l',
data = 0x7fb05164f010 "st):\nTraceback (most recent call last):\n"
"Traceback (most recent call last):\ne", maxdatasize = 1814065152,"
" datasize = 1814064225}, capflags = 16131}
This patch addresses the infinite loop issue by detecting continuously empty
responses and abort in that case.
Note that datasize can be translated to ['l', ' ', 'l', 'a']. Concatenate
datasize and data, it forms part of "Traceback (most recent call last):".
This may indicate a server-side channeledoutput issue. If it is a race
condition, we may want to use flock to protect the channels.
If the user press 'q' to leave the 'less' pager, it is expected to end the
hg process immediately. We currently rely on SIGPIPE for this behavior. But
SIGPIPE won't arrive if we don't write anything (like doing heavy
computation, reading from network etc). If that happens, the user will feel
that the hg process just hangs.
The patch address the issue by adding a SIGCHLD signal handler and sends
SIGPIPE to the server as soon as the pager exits.
This is also an issue with hg's pager implementation.
Our signal handlers forward signals to the server process, but it will
disappear soon after hgc_close(). So we should unregister handlers before
hgc_close(). Otherwise chg would abort due to kill(perrpid, sig) failure.
The problem is spotted by SIGWINCH while waiting pager termination.
Before this patch, chg will give up when it cannot connect to the new server
within 10 seconds. If the host has high load during that time, 10 seconds
is not enough.
This patch makes it adjustable using the CHGTIMEOUT environment variable.
Before this patch, chg uses the old pager behavior (pre 55f6f7fb60d2), which
executes pager in the main process. The user will see the exit code of the
pager, instead of the hg command.
Like 55f6f7fb60d2, this patch fixes the behavior by executing the pager in
the child process, and wait for it at the end of the main process.
As we don't use sockdirfd yet, this is the simplest workaround to compile chg
on old Unices where AT_FDCWD does not exist. Foozy pointed out Mac OS X 10.10
is required for AT_FDCWD as well as xxxat() functions.
Before this patch, if the user uses chg and ncurses interface, resizing the
terminal window will mess up its content.
This patch fixes the issue by forwarding SIGWINCH to the worker process.
Before this patch, if the server started by chg has exited with code 0 without
creating a connectable unix domain socket at the specified address, chg will
exit with code 0, which is not the correct behavior. It can happen, for
example, CHGHG is set to /bin/true.
This patch addresses the issue by checking the exit code of the server and
printing a new error message if the server exited normally but cannot be
reached.
We check for sockdirfd at freecmdserveropts but not lockfd, which is a bit
strange to people new to the code. Add a comment and an assert to make it
clear that lockfd should be closed earlier.
As part of the series to support long socket paths, we need to add the fd of
the directory to the cmdserveropts structure so we can use basenames instead
of full paths for sockname, redirectsockname, and lockfile.
This is a style fix. I was using tabstop=4 for some early patches, although
I realized we use tabstop=8 later but these early style issues remains. Let's
fix them.
Before this patch, chg always uses color in its debugmsg and abortmsg and
there is no way to turn it off.
This patch adds a global flag to control whether chg should use color or
not and only enables it when stderr is a tty and HGPLAIN is not set.
Before this patch, "connect to" debug message is printed repeatedly because
a previous patch changed how the chg client decides the server is ready to be
connected.
This patch revises the places we print connect debug messages so they are less
repetitive without losing useful information.
Before this patch, chg will fall back to "hg" if neither CHGHG nor HG are set.
This may have trouble if the "hg" in PATH is not compatible with chg, which
can happen, for example, an old hg is installed in a virtualenv.
Since it's very hard to do a quick hg version check from chg, after discussion
in IRC with smf and marmoute, the quickest solution is to build a package with
a hardcoded absolute hg path in chg. This patch makes it possible by adding a
C macro HGPATH.
Before this patch, if __GNUC__ is not defined, PRINTF_FORMAT_ will not be
defined and will cause compilation error.
This patch solves the issue by making sure PRINTF_FORMAT_ is defined. It
allows chg to be compiled with tcc (http://bellard.org/tcc/) by:
tcc -o chg *.c
Redirecting stdout to /dev/null has unwanted side effects, namely ui.write
will stop working. This patch removes the redirection code and helps chg to
pass test-bad-extension.t.
If the server has an uncaught exception, it will exit without being able to
write the channel information. In this case, the client is likely to complain
about "failed to read channel", which looks inconsistent with original hg.
This patch silences the error message and makes uncaught exception behavior
more like original hg. It will help chg to pass test-fileset.t.
In some rare cases (next patch), we may want validate to do "unlink" without
forcing the client reconnect. This patch addes a new "reconnect" instruction
and makes "unlink" not to reconnect by default.
Currently the validate command in chgserver expects config can be loaded
without issues but the config can be broken and chg will print a stacktrace
instead of the parsing error, if a chg server is already running.
This patch adds a handler for ParseError in validate and a new instruction
"exit" to make the client exit without abortmsg. A test is also added to make
sure it will behave as expected.
See the previous patch for details. Since the socket will be closed by the
server, handleresponse() will never return:
Traceback (most recent call last):
...
chg: abort: failed to read channel
If chgserver aborts during startup, for example, error.ParseError when parsing
a config file, chg client probably just wants to exit with a same exit code
without printing other unrelated text. This patch changes the text "cmdserver
exited with status" from abortmsg to debugmsg and exits with a same exit code.
Current pidfile logic will only keep the pid of the newest server, which is
not very useful if we want to kill all servers, and will become outdated if
the server auto exits after being idle for too long.
Besides, the server-side pidfile writing logic runs before chgserver gets
confighash so it's not trivial to append confighash to pidfile basename like
we did for socket file.
This patch removes --pidfile from the command starting chgserver and switches
to an alternative way (unlink socket file) to stop the server.
chgserver now validates and reloads configs automatically. Manually reloading
is no longer necessary. Besides, we are deprecating pid files since the server
will periodically check its ownership of the socket file and exit if it does
not own the socket file any longer, which works more reliable than a pid file.
This patch removes the SIGHUP reload logic from both chg server and client.
The chgserver is designed to load repo config from current directory. "--cwd /"
will prevent chgserver from loading repo config and generate a wrong
confighash, which will result in a redirect loop. This patch removes "--cwd /"
and uses "--daemon-postexec chdir:/" instead.
Some users may have hg as a wrapper script which sets sensitive environment
variables (like setting up virtualenv). This will make chg redirect forever
because the environment variables are never considered up to date.
This patch adds a limit (10) for reconnect attempts and warn the user with
a possible solution if the limit is exceeded.
This patch uses the newly added validate method to make sure the server has
loaded the up-to-date config and extensions. If the server cannot validate
itself, the client will receive instructions and follow them to try to reach
another server that is more likely to validate itself. The instructions can
be a redirect (connect to another server address) and/or an unlink (stops an
out-dated server).
It was necessary to go through progress.uisetup() to set up the progressui
wrapper. Since the progress extension has got into the core, progress.assume-tty
is no longer necessary.
Sometimes people may create a symbol link from hg to chg, or write a wrapper
script named hg calling chg. Without $HG and $CHGHG set, this will lead to
chg executes itself causing deadlock. The user will notice chg hangs for some
time and aborts with a timed out message, without knowing the root cause and
how to solve it.
This patch sets a dummy environment variable before executing hg to detect
this situation, and print a fatal message with some possible solutions.
CHGINTERNALMARK is set by chg client to detect the situation that chg is
started by chg. It is temporary and should be dropped to avoid possible
side effects.
There are some known unsupported commands or flags for chg, such as hg serve -d
and hg foo --time. This patch detects these situations and transparently fall
back to the original hg. So the users won't bother remembering what chg can and
cannot do by themselves.
The current detection is not 100% accurate since we do not have an equivalent
command line parser in C. But it tries not to cause false positives that
prevents people from using chg for legit cases. In the future we may want to
implement a more accurate "unsupported" check server-side.
This is a part of the one server per config series. In multiple-server setup,
multiple clients may try to start different servers (on demand) at the same
time. The old lock will not guarantee a client to connect to the server it
just started, and is not crash friendly.
This patch addressed above issues by using flock and does not release the lock
until the client actually connects to the server or times out.
Initially we use --daemon-pipefds to pass file descriptors for synchronization.
Later, in order to support Windows, --daemon-pipefds is changed to accept a
file path to unlink instead. The name is outdated since then.
chg client is designed to use flock, which will be held before starting a
server and until the client actually connects to the server it started. The
unlink synchronization approach is not so helpful in this case.
To address the issues, this patch renames pipefds to postexec and the following
patch will allow the value of --daemon-postexec to be things like
'unlink:/path/to/file' or 'none'.
We are going to make chgserver load repo config, remember what it is, and
load repo config again to detect config change. This is the first step that
passes config, repo, cwd options to server. Traceback is passed as well to
cover errors before hitting chgserver.runcommand.
They are like malloc and realloc but will abort the program on error.
A lot of places use {m,re}alloc and check their results. This patch
can simplify them.
This is necessary to suspend/resume long pulls, interactive curses session,
etc.
The implementation is based on emacsclient, but our version doesn't test if
chg process is foreground or not before propagating SIGCONT. This is because
chg isn't always an interactive session. If we copy the SIGTTIN/SIGTTOU
emulation from emacsclient, non-interactive session can't be moved to a
background job.
$ chg pull
^Z
suspended
$ bg %1
[1] continued
[1] suspended (tty input) # wrong
https://github.com/emacs-mirror/emacs/blob/0e96320/lib-src/emacsclient.c#L1094
