[gnugo-devel] possible bug in dfa matching of GnuGo

[mike polyakov]

Hi,     
   I'm using GnuGo's dfa code as part of my own evolutionary-oriented Go 
program (a student project).  Basically i copied dfa.c, dfa.h and 
patterns.h to my directory without modifying them.  I put the match_pattern 
function into my own code and modified it to be compatible.
For the most part it works fine but occasionally an assertion in 
do_sync_product (int l, int r)  fails:

      nextr = gpright->states[r].next[c];
      gg_assert (nextr < gpright->lastState + 1);

because r is being passed in as -1 (by another instance of do_sync_product).
The strings that I pass to the dfa's are 25 characters or less, and there's 
only a few of them.  The strings that cause the assertion to fail (the root 
caller of do_sync_product is dfa_add_string()) are no different than other 
strings that work harmlessly.  When it does happen, it's always when 
maxStates (of some dfa used in construction) goes from 256 to 131328 - 
though at other times this causes no problems.
I am compiling in cygwin(a recent version) with g++ 2.95.3-5.  It could be 
a bug on my part though I don't see how.
If you have any ideas, or have come across similar problems in dfa 
constructions please let me know
mike