Re: [gnugo-devel] about one GNU Go 3.7.7 windows build's test case( crash )

[Gunnar Farnebäck]

Cai wrote:
>     I build gnugo with VC++ 7 in debug mode
>     
>     When crashs, it prompts:
> 
> Debug Error!
> 
> Program: gnugo-3.7.7\interface\gnugo.exe
> Module: gnugo-3.7.7\interface\gnugo.exe
> File: gnugo-3.7.7\engine\optics.c
> Line: 1552
> 
> Run-Time Check Failure #2 - Stack around the variable 'vpos' was corrupted.

No idea what this may be. I can't reproduce it on linux, neither does
valgrind indicate any problem in the optics code. Do you have tools
and knowledge to debug this any further?

> During my building, GNU Go 3.7.7 's windows build seems broken. I
> manual built the uncompress_fuseki.c and generated fuseki[9,13,19].c

There are some windows build fixes in CVS after 3.7.7 but not related
to uncompress_fuseki, as far as I know. In general windows build fixes
need to be contributed by the people who have the necessary tools.

> I also built 3.7.7 in my linux box and executed the test case. It
> workd fine(produce "A4", though I think the best move should be B3).

Yes, B3 is a simpler and safer move.

> By the way, I think the move 102 is not best. It made left-bottom
> group gains zero points. It should be A4|A3.

B5 suffices to live with territory too.

> But it's strange that gnugo produce B5 at game and when I use
> test.sh to reproduce it, it produce C6 though I use same command
> line to call gnugo.

That might be a persistent cache problem. Potentially it could also be
a question of different random seeds in case the moves are identically
valued, but that seems less likely.

/Gunnar