Re: GOP-PROP 12: keep master in ready-to-release state (probable decision)

[Graham Percival]

On Fri, Sep 23, 2011 at 03:03:46PM +0200, David Kastrup wrote:
> But we still need to find a solution where I have a chance of getting
> work done.

dev/staging

I just realized that I forgot to add this to the proposal.  I'll
do that later tonight and send an updated version.

Or better yet -- dev/lexer-fixes.  Suppose that (almost) all your
work for the past 2 weeks went onto that branch.  You think you're
done, so ask if it's ok to merge with master.  James tests it,
sends you the output of "make doc" failing; you work a bit more,
maybe revert something, then announce that you think it's working
now.  James tests again -- and since this is a major merge, he
tests a completely blank build dir, doing a full make doc, etc --
and this time it passes.  Merge happens, git master doens't break,
much celebrations by everybody.

> Can we compromise on "make info" instead?

No, because that doesn't compile any @lilypond examples.  Besides,
"make info" is run during a normal "make" !

There are ways of speeding up the whole "make doc" process... we
can probably cut the time in half by using a server, for example,
since that avoids loading the guile libraries for every single
@lilypond.  OTOH, that would only work for hundreds of examples if
we didn't have any memory leaks.  We could probably save another
factor of two by using guile 2.0.

Both of these options would take 20-100+ hours of work, though.


I really think that separate branches are the best compromise for
the less-than-6-months future.

Cheers,
- Graham