[CF-Devel] About CVS server stability.

[Yann Chachkoff]

>
     
      My personal thought is that if we do start doing branches, it probably makes
     
     >
     
     more sense to make a 'stable' branch, instead of branching for each feature.
     
     >
     
     The main reasons for this are:
     
     >
     
     
     >
     
     1) Most bugfixes tend to make small changes to the code, so tend to be easier 
     
     to
>
     
     backport from the main branch.
     
     >
     
     2) Because the changes are small, much less likely to have conflicts.
     
     >
     
     3) the main CVS branch really is the latest and bugiest, which is probably 
     
     what
>
     
     it should be.
     
     >
     
     
     I agree with that. When a new feature pops out, we put it on the devel CVS. 
Then, if after some tests it looks stable enough, we make a release (either as 
a new package or a commit in the "stable" CVS repository). The best thing 
would be of course to test only one addition at a time, but I do understand it 
could be quite difficult to achieve.
Maybe the sourceforge features (those under the "tracker" subsection) should 
be more intensively used ?

>
     
      I don't know the issues behind the java editor.  but IMO, AV pretty much
     
     >
     
     followed proper protocol on this checkin: He discussed it ahead of time, made
     
     >
     
     changes to fix some peoples issue with the code, and then checked it in.  
     
     Maybe
>
     
     the discussion time wasn't really long, but at the same time, the amount of 
     
     code
>
     
     isn't tremendous.
     
     >
     
     
     That's right. On the other hand, it seems that a general consensus wasn't 
achieved.

>
     
      Since it has been agreed that CVS is for latest and greatest, people running
     
     >
     
     CVS know their is risk of new bugs.  That is nothing new - in any case, AV 
     
     would
>
     
     have had to put this code into CVS at some point - realistically, that is the
     
     >
     
     only way to really work out the bugs and potential balance issues.
     
     
Since there are not a lot of "stable" releases, the CVS is the only realistic 
source for up-to-date stuff. Most people will use it simply because the stable 
packages are 6 months old and thus quite outdated. I agree that when you use 
the CVS, you also understand the risks. But if there are no serious 
alternative to CVS, then we need to ensure that the code in it is stable 
enough under normal circumstances.

Now I also disagree when you say that it is the "only way to really work out 
the bugs" - patches do exist and are commonly used in other projects when 
submitting new stuff (that also explains why sourceforge has a "patches" 
service). Why are we unable to use them more widely ?

I repeat again what I've said a lot: if we do not agree on a clear 
submission/test/update protocol, the problems will get worse and will *not* 
solve by themselves.

Of course, this is just my opinion - maybe I'm as wrong as usual :)
Y. Chachkoff

------------------------------------------------
Visit 
     
     http://www.chachkoff.pronym.org
     
      for a 
journey into a fantastic world !
(But don't expect too much...)
GPG: 
     
     http://www.chachkoff.pronym.org/gros.pub
     
     
------------------------------------------------