Re: SNC crashes

[Thomas Birke <birke@bii.bessy.de>]

>>>>> "William" == William Lupton <wlupton@keck.hawaii.edu> writes:
>>>>> "Benjamin" == Benjamin Franksen <franksen@bii.bessy.de> writes:

 William> Ben, You wrote:
 Benjamin>> I have a problem with the State Notation Compiler...

 William> Which version? Your example seems to compile fine with the
 William> R3.13.0Beta11 version of the sequencer.

Hi,

I managed to get the snc source-code into ObjectCenter, and ObjectCenter
came to the conclusion that when compiling Benjamin's state-machine the snc
performs a "write past end of allocated array".

The exact position and error-message was:

        FILE: sequencer/gen_tables.c
        LINE: 462
        FUNCTION: eval_event_mask
        Illegal array index 2 into allocated data at <0x...>.
        Maximum array index is 1.

it's the line

462:    bitSet(pEventWords, cp->index + num_events + 1);

whith   cp->index       set to   58  and
        num_events      set to    5

So the bit-number evaluates to 64, which is *not* in the 2 allocated
words. Allocation is performed in line 215 of the same file
        
214:    numEventWords = (num_events + num_channels + NBITS - 1)/NBITS;
215:    pEventMask = (bitMask *)calloc(numEventWords, sizeof (bitMask));

where   num_events      is    6
        num_channels    is   58 and
        NBITS           is   32

Why is there a 1 added to the bitno in line 462? Is bit 0 used for special
purpose? If so, the number of used bits should be incremented in line 214.

So I applied the following patch to our snc:

--- gen_tables.c.ORIG  Wed Mar 11 15:05:44 1998
+++ gen_tables.c Wed Mar 11 15:24:26 1998
@@ -211,7 +211,7 @@
 
 
   /* Allocate an array for event mask bits */
-  numEventWords = (num_events + num_channels + NBITS - 1)/NBITS;
+  numEventWords = (num_events + num_channels + NBITS)/NBITS;
   pEventMask = (bitMask *)calloc(numEventWords, sizeof (bitMask));
 
   /* for all state sets ... */

The .c-files produced by snc are identical in the case the "old" snc (we're
still talking about beta11) doesn't crash. And the "new" snc doesn't crash
with Benjamin's state-machine.

So long,
Thomas
-- 
_______________________________________________________________________________
 /homas Birke  BESSY II   (+49)-30-6392-4934 (fax -4859)    birke@bii.bessy.de