got a corrupted external term

[zwizwa]

Hi,

I keep running into this error and I'm a bit lost regarding how to start debugging it. I'm not quite sure how it happens, but in very vague terms it seems to happen when I send too much logging data from Erlang to Emacs. Once this error triggers, the distel communication gets into a bad state, and the only way I've found to fix it is to restart emacs.

=WARNING REPORT==== 12-Apr-2020::10:37:00 ===
'exo@10.1.3.29' got a corrupted external term from 'distel_27211@panda' on distribution channel 28280
<<...,131,104,4,97,6,103,100,0,18,100,105,115,116,101,108,95,50,55,50,49,49,64,112,97,110,100,97,0,0,188,139,0,0,0,0,0,100,0,0,100,0,3,114,101,120,131,104,2,103,100,0,18,100,105,115,116,101,108,95,50,55,50,49,49,64,112,97,110,100,97,0,0,188,139,0,0,0,0,0,104,5,100,0,4,99,97,108,108,100,0,4,99,111,100,101,100,0,13,101,110,115,117,114,101,95,108,111,97,100,101,100,108,0,0,0,1,100,0,6,100,105,115,116,101,108,106,103,100,0,18,100,105,115,116,101,108,95,50,55,50,49,49,64,112,97,110,100,97,0,0,0,1,0,0,0,0,0>>
ATOM_CACHE_REF translations: none

[zwizwa]

Taking that apart gives a term

Term = <<
131, %% ETF
104,4, %% SMALL_TUPLE_EXT
97,6,%% SMALL_INTEGER_EXT
103, %% PID_EXT
100, %% ATOM_EXT
0,18,
100,105,115,116,101,108,95,50,55,50,49,49,64,112,97,110,100,97, %% Node: "distel_27211@panda"
0,0,188,139, %% ID
0,0,0,0, %% Serial
0, %% Creation
100, %% ATOM_EXT
0,0,
100, %% ATOM_EXT
0,3,
114,101,120>>.

followed by another term
131, ....

binary_to_term(Term). gives exception error: bad argument, as does binary_to_term(Pid) below, so it seems to be the PID_EXT.

Pid = <<
131,
103, %% PID_EXT
100, %% ATOM_EXT
0,18,
100,105,115,116,101,108,95,50,55,50,49,49,64,112,97,110,100,97, %% Node: "distel_27211@panda"
0,0,188,139, %% ID
0,0,0,0, %% Serial
0>>. %% Creation

The format seems to be ok, so what causes the term parse error?

[leoliu]

Based on #68 (comment):

the ID of the PID is 48267 which is 1011110010001011. ETF says ID "Only 15 bits are significant; the rest are to be 0." Looks like the 16th bit may be causing trouble.

[zwizwa]

If I change that Term to something that doesn't have bit 15 set then indeed it does parse with binary_to_term/1. Am I spawning too many processes? I do have a lot of transient processes in the logger I mentioned.

[leoliu]

It’s not your fault. I will see if I can send in a fix today. There is likely a bug in distel that does not handle ID wraparound correctly.

Could you try the PR and see if it has any ill effect. Thanks.

[zwizwa]

Thanks. I've switched to f9c5067 as my daily driver. I'll let you know how it goes. I can confirm that in the buggy version the problems indeed started to happen when erl-pid-counter went over 32k.

Just FYI: Maybe I have github misconfigured, but it seems to not send notifications for message edits. I just saw your fix when I checked back to see if I could do some debugging myself... It does seem to send notifications for new messages.

[leoliu]

My bad. I was not fully aware of how gh notifications work. I might amend the PR with some doc fix as follows.

--- a/elisp/erl.el
+++ b/elisp/erl.el
@@ -58,7 +58,7 @@ (defun make-erl-local-pid ()
     pid))
 
 (defun erl-get-counter (counter)
-  "Get and then advance counter."
+  "Get and then advance COUNTER."
   (cl-macrolet ((modf (place y) `(cl-callf mod ,place ,y))
                 (roundf (place) `(cl-callf round ,place)))
     (cl-labels ((increment (cells)
@@ -83,7 +83,7 @@ (defun erl-determine-hostname ()
 
 ;; (NUM SERIAL)
 (defvar erl-pid-counter '((0 . #x8000) (0 . #x2000))
-  "Counters for PIDs.")
+  "Counter for PIDs.")
 
 (defvar erl-node-name
   (intern (format "distel_%S@%s" (emacs-pid) (erl-determine-hostname)))

[zwizwa]

I'll keep an eye on the counter :)

Currently at:
erl-pid-counter
((8917 . 32768) (0 . 8192))

[zwizwa]

Things started to misbehave once it wrapped. Currently at
((9 . 32768) (1 . 8192))

My code is a horrible rube goldberg machine so I'm not really clear what exactly happens. I'll look into this some more to see if I can get you a decent error log or a fix.

[zwizwa]

I ran a test until the next wrap around and something went wrong again.

In the minibuffer, I get "Buffer "derl exo@10.1.3.29" has a running process; kill it? (yes or no) "

The beam says:
=ERROR REPORT==== 1-May-2020::18:38:55 ===
** Node distel_13251@panda not responding **
** Removing (timedout) connection **

I get an error message in the emacs Messages log:
erl-pid->buffer: No buffer for pid [TYPE erl-pid distel_13251@panda 34 1 0]

And the erl-pid-counter is at
((35 . 32768) (2 . 8192))

So it seems that the 34 1 0 in the error message is supposed to be 34 2 0.

[zwizwa]

I think I found the issue in:

(defun erl-pid->buffer (pid)
"Get PID's buffer."
(or (cdr (assoc (erl-pid-id pid) erl-process-buffer-alist))
(error "No buffer for pid %S" pid)))

That only uses erl-pid-id and not erl-pid-serial.

My code fails because right after wraparound 1->2 this no longer finds a registered process due to erl-pid->buffer failing for [TYPE erl-pid distel_13251@panda 34 1 0]

[zwizwa]

Also erl-unenroll-process removes processes if just the id matches. Seems like the fix would be to change the keying for erl-process-buffer-alist.