How to debug emacsclient?

Question

Sometimes emacsclient fails with an error like:

$ emacsclient .gitlab-ci.yml
Waiting for Emacs...
*ERROR*: Args out of range: 63695, 1, 1173

How can I debug this? emacsclient has no debug option, no backtrace is shown, and nothing is logged in the *Messages* buffer. Running the daemon with --debug-init does not help.

strace shows what the client sends the server over the socket, so if there is a bug, it is probably on the server side:

$ strace --trace=sendto,recvfrom -s 100 emacsclient ../.gitlab-ci.yml
sendto(3, "-dir /home/sam/Downloads/emacs/lisp/ -current-frame -file ../.gitlab-ci.yml \n", 77, 0, NULL, 0) = 77
Waiting for Emacs...
recvfrom(3, "-emacs-pid 33810\n-error Args&_out&_of&_range:&_116433,&_1,&_1173", 8192, 0, NULL, NULL) = 64

*ERROR*: Args out of range: 116433, 1, 1173
+++ exited with 1 +++

I ran the daemon in GDB with breakpoints on xsignal0, xsignal1, xsignal2, and xsignal3. Then when I invoke the emacsclient, I get:

(gdb) bt
#0  0x00000000004248da in xsignal3 ()
#1  0x0000000000423bc4 in args_out_of_range_3 ()
#2  0x0000000000597ade in Fline_number_at_pos ()
#3  0x0000000000590b6b in Ffuncall ()
#4  0x00000000005cc740 in exec_byte_code ()
#5  0x0000000000593480 in funcall_lambda ()
#6  0x0000000000590a97 in Ffuncall ()
#7  0x00000000005cc740 in exec_byte_code ()
#8  0x0000000000593480 in funcall_lambda ()
#9  0x0000000000590a97 in Ffuncall ()
#10 0x0000000000590b79 in funcall_nil ()
#11 0x000000000059026d in run_hook_with_args ()
#12 0x0000000000590378 in run_hook ()
...snip...
#40 0x0000000000429f56 in main ()

see full stack

The problem is that I am invoking emacsclient from inside *ansi-term* inside emacs. For some reason, the emacs server tries to use the position in the *ansi-term* buffer (63695) in the input file (only 1173 bytes).

This can be fixed by running:

$ emacsclient --eval '(find-file ".gitlab-ci.yml")'

but find-file doesn't wait for the file to be closed, so I still want to find out the root of the problem.

It appears (from the stacktrace) some hook, perhaps 'server-after-make-frame-hook, is causing the faulty behavior. But Emacs in my distro is not compiled with debug symbols, so I can't poke around in that frame easily. Is there an elisp way of seeing what functions are bound to a hook?

The problem seems to go away when I disable (indent-guide-global-mode), but I'm not sure why.

Answer 1

Sadly that could be a lot of things.

I think you will need to run the emacs inside of gdb

1. Run gdb --args emacs -nw, then start the server in that emacs.

   a. Alternatively, start the server normally, pgrep emacs | xargs ps fp | grep "emacs ". Note the space in the last grep command identifies the emacs server as opposed to the emacsclient. Then run sudo gdb -ex 'attach $PID' where $PID is the server's process ID.

2. Set a breakpoint on the xsignal-functions: type break xsignal0, break xsignal1, break xsignal2, break xsignal3.

3. Type c or continue

4. Do whatever invocation causes you to hit an error.

5. Switch back to gdb and type bt or backtrace. This shows the place in C code where the emacs server broke.

The emacs server/client protocol is implemented in emacs/lisp/server.el and emacs/lib-src/emacsclient.c. When emacsclient asks the emacs server to open $FILE, it will send -file $FILE. You can view the protocol trace by running:

$ strace -s 500 --trace=sendto,recvfrom emacsclient $FILE

The backtrace explains what went wrong, while the protocol trace explains how the server got there. With these in hand, you can begin to debug emacsclient.