[MIT-Scheme-devel] Edwin cannot run `M-x shell' on Mac OS 10.8

[Rich Loveland]

   Date: Sat, 22 Jun 2013 22:18:42 +0000
   From: Taylor R Campbell <address@hidden>

   M-x set-variable RET debug-on-internal-error RET #t RET

Ah, thanks. Here it is:

The error that started the debugger is:

  The primitive channel-nonblocking, while executing the fcntl-setfl system 
call, received the error: Inappropriate ioctl for device.

 S0  (channel-nonblocking 2)
 S1  (channel-nonblocking channel)
 S2  (let ((channel (subprocess-input-channel subprocess))) (if channel (be ...
 S3  (let ((program (or (%record-ref edwin-variable$explicit-shell-file-nam ...
 S4  (apply procedure (interactive-arguments command record?))
 S5  (apply procedure (interactive-arguments command record?))
 S6  (if (queue-empty? command-reader-override-queue) (let ((input (get-nex ...
 S7  (bind-abort-editor-command (lambda () (let do-loop () (reset-command-s ...
 S8  (bind-condition-handler (cons condition-type:error (quote ())) interna ...
 S9  (bind-abort-restart cmdl (lambda () (deregister-all-events) (with-inte ...
 S10 (procedure (car l))
 S11 (for-each (lambda (act) (act)) (reverse after-parsing-actions))
 S12 (apply (car receivers) (cdr event))
 S13 (procedure item)
 S14 (procedure (car l))
 S15 (ignore-errors (lambda () ((->cmdl-message message) cmdl)))
>S16 (bind-abort-restart cmdl (lambda () (deregister-all-events) (with-inte ...

   That's not the error in question -- look for fcntl(..., F_SETFL, ...),
   or the numeric value from <sys/fcntl.h> instead of F_SETFL.  It looks
   like this is probably the failure:

   fcntl(0x5, 0x4, 0x6)          = -1 Err#25

I've attached some more dtruss output. I think this is the bit (line
1197) where I `M-x shell' for the first time:

access("/bin/bash\0", 0x1, 0x7800000000000000)           = 0 0
open("/dev/ptmx\0", 0x20002, 0x0)                = 15 0
ioctl(0xF, 0x20007454, 0x0)              = 0 0
ioctl(0xF, 0x20007452, 0x0)              = 0 0
ioctl(0xF, 0x40807453, 0x7FFF7AB72DC0)           = 0 0
stat64("/dev/ttys012\0", 0x7FFF5FBFF890, 0x0)            = 0 0

Later (on line 1254), an operation on a file descriptor with the same
number fails (debugger opens):

fcntl(0xF, 0x3, 0x0)             = 2 0
fcntl(0xF, 0x4, 0x6)             = -1 Err#25

If I keep trying `M-x shell', I fail and fail as expected (Tried
dozens of times in a row to be reasonably sure).

However, when I attach dtruss to the Scheme process, I can retry a few
times, and eventually the inferior shell will start, albeit more
slowly than normal. This is represented around line 4152, when I try
to start the shell again:

access("/bin/bash\0", 0x1, 0x7800000000000000)           = 0 0
open("/dev/ptmx\0", 0x20002, 0x0)                = 16 0
ioctl(0x10, 0x20007454, 0x0)             = 0 0
ioctl(0x10, 0x20007452, 0x0)             = 0 0
ioctl(0x10, 0x40807453, 0x7FFF7AB72DC0)          = 0 0
stat64("/dev/ttys013\0", 0x7FFF5FBFF890, 0x0)            = 0 0

This time, it works and the inferior shell starts up (line 4211):

fcntl(0x10, 0x3, 0x0)            = 2 0
fcntl(0x10, 0x4, 0x6)            = 0 0

I verified this ``intermittent start with dtruss running'' behavior
several more times.  My guess is that for some reason dtruss is
slowing Scheme down enough for this command to work. Of course I don't
know enough to know why, that's why I'm here. :-}

Hope that helps. Thanks for looking into this.

edwin-dmg-dtruss-output2.txt
Description: Text document