[gnugo-devel] Re: Code removal (#25) (reorientation code)

[Teun Burgers]

Arend wrote:

> I understand its purpose, but:
>
> 1. Since I have joined GNU Go 4 years ago, there has never been a bug
> discovered by this.

True. The same argument holds for many test cases as well.

> 2. There are so many expected rotational dependencies all around the code,
> that I don't think any developer would track down a rotational depency
> of a test case and see whether it really is a bug (since 99% of the time
> it won't be).

The exception are the reading tests. They are currently invariant
under reorientation. It might be an idea to drop the owl_rotall
test and replace it with a reading_rotall test.

> 3. This code does not come for free, as it obfuscates rather
> central code. (Try 'grep orientation' or 'grep rot' in interface/.)

I tried but don't see much obfuscation. grep orientation only matches
on the reorientation code. If the rotate* functions would be renamed
to reorient* that would further reduce obfuscation.

> 4. This functionality is easily implemented with external scripts.
> (I have a script to convert .sgf-files to a GTP stream that could
> easiliy be adopted to convert the loadsgf commands in the .tst-files 
> on the fly to GTP stream that already includes the orientation.)

It can be, but the script based approach before the reorientation code
was introduced was messy, leading to the reorientation code. May I 
suggest to remove the reorientation code once the new script based 
approach has been added to CVS and works well.

Teun