Re: [gnugo-devel] More bad DRAGON2 accesses

[Trevor Morris]

At 10:43 PM 3/9/2002 -0800, Daniel Bump wrote:
>
>> Here's a patch to one of at least a few problems exposed
>> by this re-definition of DRAGON2:
>> 
>> http:/www.public32.com/games/go/trevor_1_27.1
>>  - re-order conditions in revise_semeai to insure valid DRAGON2 data
>
>Trevor's patch replaced the DRAGON2 macro by a function call.
>I changed this so this isn't the default, i.e. unless you edit
>one line of liberty.h you have the old definition of DRAGON2,
>on the theory that it's easy enough to compile it the other
>way for testing but it shouldn't be the default.
>
>Your comment seems to imply there may have been other problems
>exposed by the redefinition. But I suppose that must not be
>true since 3.1.27 (which included the redefinition) doesn't
>hit the ASSERTs in dragon2_func.

Yes, I believe there were three different places that had 
bad dragon2 accesses.  It should be fine to disable the bounds
checking now, as they've all been fixed.  The only risk is that
there is a rare bad access that hasn't shown up in the last
couple of weeks.

-Trevor