[cvsnt] Re: Converting from -kb to -kB (and defaulting to -kB in the future)

[Bo Berglund]

On Mon, 27 Mar 2006 16:14:23 -0500, "Jesse" <cvs at gamesthatwork.com>
wrote:

>Hi,
>I work for a small company, and set up a CVSNT 2.5.01.1969 server.  It was 
>primarilly set up for our programming staff, but the artists use it too.  I 
>told them to simply flag all their files as binary in tortoise (they never 
>use ASCII files), but then they got huge.
>
>So I finished RTFMing, and I noticed that it's not using binary diffs 
>between photoshop, etc. files.
>So I went in, and tried admin -kB filename, but that didn't work.  I assume 
>it's because there are already several revisions with the ascii type of 
>diffs.  Is there a way to (hopefully batch) convert the months of files and 
>revisions to use binary deltas instead?
>
>Or does this not make the huge difference I think it will?
>
>Also, is there a way to have tortoise (2.0.62.1817) default to using -kB for 
>binary files instead?
>
>The only other solution is going to be me spending three days rebuilding 
>each file by hand.  And there getting the old versions of some of these 
>files will be a long process.
>
>Thanks in advance! 

I am not sure, but the way the binary diff could ever work must be by
recognizing some part of the committed binary file as non-changing and
then it needs only extract and save the part that did change.
This would work well with files like Microsoft Office documents being
worked on since typically only a small part of the file is changed on
each session (mostly text too). So the binary diff will be quite small
and probably compressible as well.

But you are talking about "artists" and "photoshop" which indicates
that the files could be jpeg encoded or similar. For such files I have
a gut feeling that whenever you open such a file in a paintshop kind
of program and work on it and then save it as JPG, the whole file will
be altered due to the 2D compression algorithms used. Therefore it is
very hard to find any constant parts of the file and thus compute a
binary diff worth anything.

But I have not tried this myself yet so I cannot say for sure. Would
be good to hear from a real life test. You seem to have the
possibility to do this by extracting a number of already checked in
revisions and then using these files to create a test file with kB
storage. Please report your findings comparing this with the original.

Concerning Tortoise, is there not a way to point Tortoise to an
alternate CVSNT binary or does it have to reside in the Tortoise
install directory? If there is you should be able to use the CVSNT
version you have on your server.


/Bo
(Bo Berglund, developer in Sweden)