[Freeciv-Dev] Re: (PR#10085) Game proceeds differently on different systems

["Gregory Berkolaiko" <Gregory.Berkolaiko@xxxxxxxxxxxxx>]

<URL: http://rt.freeciv.org/Ticket/Display.html?id=10085 >

On Mon, 13 Sep 2004, Marko Lindqvist wrote:

>   There's no much hope to really catch this one, but I wanted to open
> ticket just in case. Data collected should be saved somewhere in case it
> someday turns useful.
>
>   Jason cannot reproduce #10061 while I can. We definitely have very
> different systems, but low-level differences are not supposed to affect
> how game proceeds. I have confirmed that year -3950 save games are
> identical.
>
>   All this is probably quite harmless. Normal users don't care if game
> would run slightly differently on somebody else's machine. Also, problem
> can be cygwin (<->other systems) specific, meaning that developers
> (except me for now) are not losing bug reproducibility because of this.
>
>   So far all my attempts _not_ to reproduce #10061 using freeciv built
> from late-Sep-11 - midday-Sep-12 sources have failed. Now I have
> downloaded cvs-Sep-11 snapshot from freeciv.org. This way I have known
> set of source & data files, with autogen.sh already run. I run
> ./configure without any parameters (or sometimes with
> --enable-maintainer-mode, but this has no affect - save games up to
> assert in 1545 are identical).
>
>   Unless it turns out that Jason can reproduce #10061 with above
> setting, I guess that only way to proceed with this (if we bother) is to
> compare more save games in order to find point where they start to
> differ. My first guess is that -375 saves are identical, -350 are not
> (auto saves are made in the beginning of turn, right? Otherwise I have
> off-by-turn error here)

some time ago I did a major divergence bug hunt and I wanted to write a
howto but never got to it.  the basic idea was to turn on LOG_DEBUG
messages (including the one in myrand() which is currently commented out)
and compare the resulting HUGE files.  you can also add some custom
freelogs to crucial places like handle_*_request and switch off the bulk
(which is too much).  Once you find the divergence in logs, you can quite
easily pinpoint the divergence in the code and then some gdb work will
find you the cause.

It's preferable to have access to both computers in your case, of course.

Gr.