bug#28211: Grafting code triggers GC/thread-safety issue on Guile 2.2.2

[Mark H Weaver <mhw@netris.org>]

Hi,

I've been studying the backtrace provided by Ludovic, trying to
determine where, in the Scheme code, this error happened in thread 1.

ludo@gnu.org (Ludovic Courtès) writes:
> Thread 1 (Thread 0x7f6fe6f5d700 (LWP 2856)):
> #0  0x00007f7019db0d79 in scm_is_pair (x=<error reading variable: ERROR: Cannot access memory at address 0x0>0x0) at ../libguile/pairs.h:159
> #1  scm_ilength (sx=<optimized out>) at list.c:190
> #2  0x00007f7019e0e2f6 in vm_regular_engine (thread=0x1425670, vp=0x144e6c0, registers=0x0, resume=16) at vm-engine.c:909
> #3  0x00007f7019e117da in scm_call_n (proc=proc@entry=#<program 1555fe0>, argv=argv@entry=0x0, nargs=nargs@entry=0) at vm.c:1257
> #4  0x00007f7019d94879 in scm_call_0 (proc=proc@entry=#<program 1555fe0>) at eval.c:481
> #5  0x00007f7019d85eb8 in scm_call_with_unblocked_asyncs (proc=#<program 1555fe0>) at async.c:400
> #6  0x00007f7019e0e17d in vm_regular_engine (thread=0x1425670, vp=0x144e6c0, registers=0x0, resume=16) at vm-engine.c:784
> #7  0x00007f7019e117da in scm_call_n (proc=#<program 13717c0>, argv=argv@entry=0x0, nargs=nargs@entry=0) at vm.c:1257
> #8  0x00007f7019d94879 in scm_call_0 (proc=<optimized out>) at eval.c:481
> #9  0x00007f7019dff496 in really_launch (d=0x149e520) at threads.c:794

The call to 'scm_call_with_unblocked_asyncs' in frame #5 is an important
clue, because it is used in very few places.  I see only two such places
in Guile (and none in Guix):

  https://git.savannah.gnu.org/cgit/guile.git/tree/module/ice-9/scm-style-repl.scm?h=v2.2.3#n80
  https://git.savannah.gnu.org/cgit/guile.git/tree/module/ice-9/threads.scm?h=v2.2.3#n158

I assume that it must be the second one in this case.  So, I believe
frames 5-9 correspond to lines 147 through 158 of ice-9/threads.scm:

  https://git.savannah.gnu.org/cgit/guile.git/tree/module/ice-9/threads.scm?h=v2.2.3#n147

which is within 'call-with-new-thread', called via the 'begin-thread'
macro used by 'n-par-for-each', here:

  https://git.savannah.gnu.org/cgit/guile.git/tree/module/ice-9/threads.scm?h=v2.2.3#n320

However, I would expect to see a 'catch' frame deeper in the stack,
since the procedure passed to 'n-par-for-each' by graft.scm is wrapped
by 'exit-on-exception'.

From these facts, I believe we can conclude that the error is happening
within the body of 'begin-thread' in 'n-par-for-each', but outside of
the body of 'proc' passed to 'n-par-for-each'.

There's not a lot of code in there.  My best guess at the moment is that
this error is happening within the (apply proc args) on line 335,
although I do not yet have an explanation of how that could happen.

To be continued...

       Mark