From: AlisdairM (alisdair.meredith_at_[hidden])
Date: 2004-01-09 12:49:22
Robert Ramey <ramey_at_[hidden]> wrote in
> The first warning is when I hear people wanting to get
> in "one more feature"
> In my view, I see the CVS development tree as a
> "release candidate" and would contain code that that
> as far as the author knows is ready for release. Of
> course, this is never quite true due to the fact that
> we test on more platforms than any one person has.
> I know this might sound ridiculous but
> how about a "surpise release" strategy?
OTOH, where else will development occur? I agree that alpha-code should
not be appearing in the main tree, but it is often the case that a library
developer does not have access to the full range of compilers used in the
regression tests, and it is the push for release that shakes out the final
solutions from those that do.
Likewise, much of that development that was being held back until ready
suddenly is faced with missing a release, or coming out NOW! If the
library author believes that code ready for this cycle they are going to
commit, and that is when those other compilers really start shaking out.
That is why I put up my alternate suggestion, announce at least one branch-
date ahead of time so authors know when they need to get their updates
ready by. If they know the branch date in advance, they are better able to
guess if their latest developments can make the release with that extra
push, or if they should focus on the following release instead.
When the branch is announced late, naturally anyone who is working on
something panics and tries to push it into the current release, or else all
that work is lost for at least 3 months, and maybe 6 or more.
Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk