Re: [Bug-gnubg] Crash in 1.06.002

[Terje Pedersen]

Hi!

No the command file just sets up a single position from an XGID and
evaluates a move.

Ah. bugger.. I sent you the wrong command file. Attaching the correct one.

I'll give the other suggestions a try tomorrow.

Best regards,
TP


On Mon, Nov 26, 2018 at 11:30 PM Philippe Michel
<address@hidden> wrote:
>
> Hello Terje,
>
> I agree with your general diagnostic (the problem is elsewhere but the
> analysis printing code shouldn't crash anyway). But I cannot reproduce
> the problem with the command file attached.
>
> The command file amounts to the analysis of a 43 opening roll while the
> log file shows a bearoff move.
>
> You write about "numbers growing larger until it crashes". Is the actuel
> command file supposed to be the analysis of an entire game ?
>
> Since you say you tried a version you built yourself, you may want to
> add "--enable-gasserts" to your configure command. It could catch
> silly numerical values in evaluations earlier.
>
> Another thing that could be tried is to build gnubg with
> "-fsanitize=address" addef to CFLAGS. This is more cumbersome : you may
> need to add some prerequisite packages (libasan ?) and it will create a
> noticeably slower binary (while --enable-gasserts is not very costly).
> It could catch cases where the array of evaluations is overwritten by
> something else, another possible cause of absurd values, although it
> doesn't really fit with the "numbers growing larger" you report.
>
> The latter is something I did at some point, and it allowed to fix many
> memory issues, but I'm sure I didn't test everything.
>

commands2.txt
Description: Text document