[Pachi] Pachi 10.0 soon

Petr Baudis pasky at ucw.cz
Wed Aug 8 21:55:35 CEST 2012


On Wed, Aug 08, 2012 at 08:09:55PM +0100, Neil Mclean wrote:
> I've tidied up the gcc_win32_cleanup branch with your suggestions, let
> me know if you can use it or not for pachi 10.0

  Thanks! I have merged the commits that are bugfixes, but I would
rather postpone merging the cleanups after the release in order not to
break anything by accident, hope that's ok by you.

> * 1e5a3894bbda8dae68bdb11b3bfd81ee23fe3f7b
> Win32 gcc compiler warning fix. Added util.h prototype...

  I've taken this one, rewriting it to fix the commit message that
contained an extra unrelated line.

> * 8066a0db3daba07b67a5933f30e612438fda12a4
> Change type of mercymin to 'int' to avoid gcc sign...

  I've taken this one, splitting it to three commits since the changes
didn't seem related.

  However, something went wrong during your Git operations - the whole
gcc_win32_cleanup branch is based on commit
89141961c1c281e82f3e0c6d2f45eaf3a6ba891d that is a rewritten form of
15d5b939c91d61b539ca28786917220c5c806c8b! This was a little confusing...
In the end, I cherrypicked the commits.

> * fa41772db9aea812c518dd5c38789634232a9f41
> Add -Wextra flag for gcc. gcc_win32_cleanup
> * 0e47e27f6e375f47ec7f5d0f67b49ecb9e90564d
> Use of (void) to suppress compiler warnings about unuse...
> * c0ece51cf6d53abc047d823df45147a98a3d9b69
> GCC: new macro in util.h 'UNUSED' to suppress unused...

  These three commits look great, thanks! I will be happy to merge them
when 10.0 is tagged. Could you please rebase them on top of current
master so that this time, we get a clean merge?

> >> I've split up the gtp commit that 'moved files around and also changed files'.
> >> Smaller commits is the way to go. I'm surprised how nice git is to
> >> use, wow, very powerful!
> >  Great. :) However, the state that I see in the gtp branch still seems
> > to mix code changes and movement? Perhaps you forgot to push out the
> > last state of your branch?
> Was it the gtp branch in nmclean you looked at? I'm only a beginner at
> git so, I'm not too savvy yet.


> This commit:
> * 827a933e88da173979da2fa8bcd54c41cc1fb2bf
> I guess i could take the gtp.c and gtp.h file and split it to the
> different files, but it would'nt compile... is that what you mean?

Yes, this commit changes the tree to a strange state where things are
not in place, and splits to files are still mixed with code changes
(splits to functions). Also, the new code doesn't actually work either,
you are fixing that in further commits.

For example, having commit


separate is fine as it's a gradual improvement of existing code. But
having original commit introduce gtp_pachi-result() and then fixing that
up to gtp_pachi_result() in another commit is less ok since the
intermediate state is not consistent.

Let me know if fixing this goes over your head and we can try to work on
refactoring the branch together.

(A minor nit - valid_color() is preferrable to valid_colour() since we
already use the color spelling. But again - fixing this up in a separate
commit (possibly later in the future) would be fine as the intermediate
states can be roughly consistent.)

> It'd be nice if you could include the gogui branch which uses the gtp
> branch. The gogui engine parameters are working. it may be too close
> to the 10.0 tag though and i guess I may have to rework the code a
> bit...

Great, it seems we will get a really decent gogui support!

I think this would also be better to include early in 10.99-devel
branch. 11.00 needn't be very far away, after all.

Again, perhaps the intermediate experiments in the history should be
squashed to a few internally consistent commits here. I have some
comments about the code as well, it seems that the UCT engine details
unnecessarily spilled out to the GTP file. But it shouldn't be hard to
deal with this; let's focus on it when the other branches are merged.

> I'll make a new .exe for the website once you've tagged it. I'll be
> using the linux version :)

Awesome, thanks.

				Petr "Pasky" Baudis
	Smart data structures and dumb code works a lot better
	than the other way around.  -- Eric S. Raymond

More information about the Pachi mailing list