Re: [gnugo-devel] Segmentation fault, dragon2 array appears to be corrupted

[Gunnar Farneback]

Dan wrote:
> > Gunnar Farneback wrote:
> > 
> > > I'd recommend anyone who has a CVS working copy which was checked out
> > > before 3.1.19 (assuming that was when the change to automake 1.5 was
> > > done) to do a "make clean" in order to avoid strange surprises.
> > 
> > Good advise. Indeed gnugo --version erroneously reported 3.1.19
> I commented on this phenomenon in:
> 
> http://mail.gnu.org/pipermail/gnugo-devel/2002-January/001038.html

Yes, that was the message which started this thread.

> Is the automake 1.5 scheme for calculating dependencies buggy?
> 
> This is going to be a recurring problem if files depending
> on config.h are not recompiled when you reconfigure.

As I wrote in my reply to your message this may very well be a
transitional problem. I.e., assume you have a working copy which was
checked out while automake 1.4 was used. The dependencies would be set
up correctly during the build. Then when automake 1.5 was introduced
you updated your working copy and built, but if the dependency system
was changed it's quite possible that your tree is now in a state with
mixed dependencies from automake 1.4 and automake 1.5, which may very
well not work properly together. That was why I recommended people to
do a "make clean" to try to throw away the old stuff. Maybe they
should even do "make distclean; configure" to be on the safe side.

I doubt there's anything wrong with the automake 1.5 dependency
scheme. Updating from 3.1.19 to 3.1.20 went fine in the working copy
where I did "make clean" after the update from 3.1.18 to 3.1.19 didn't
have any effect on the version number.

/Gunnar