Re: [clisp-list] More problems related to MT => with-timeout/abort

[<don...@is...>]

Don Cohen writes:
 > Vladimir Tzankov writes:
 >  > Many CL libraries will have similar issues (imagine inplace nreverse  
 >  > implementation being interrupted by timeout).
 > Clearly this is an example of something that should be abort protected.
Just thought this might interest you:
  [1]> (setf *print-circle* t)
  T
  [2]> (setf l (list 1 2 3))
  (1 2 3)
  [3]> (setf (cdr (last l)) l)
  #1=(1 2 3 . #1#)
  [4]> (nreverse l)
  #1=(2 . #1#)
I was surprised that this finished.
  [5]> (reverse l)
This hangs, so I try to interrupt it.
  ^C
  Signal 6 while exiting on a signal; cleanup may be incomplete
  Aborted
  [2009-02-26 14:08:43 root@number11 ~/clisp/build-dir]

It seems clear to me that all such functions like nreverse or reverse
that could take arbitrarily long ought to be interruptible.  Of
course, they should also use something like with-deferred-abort to
warn the user that an abort will leave some data in an unexpected 
state.  

On the subject of the interpreter, I view this as similar to reverse 
in the sense that it's a limited amount of code, and should contain
checks for interrupts at places that guarantee that it cannot ignore
an interrupt request for very long.  I'd think that the beginning of
a form would be adequate.  That would include (go ...).  And in that
case, perhaps it would not even need to defer aborts.