[haiku-development] Re: Haiku R1A5 release timeline
- From: Oliver Tappe <zooey@xxxxxxxxxxxxxxx>
- To: haiku-development@xxxxxxxxxxxxx
- Date: Fri, 30 May 2014 00:37:00 +0200
On 2014-05-29 at 23:41:08 [+0200], Urias McCullough <umccullough@xxxxxxxxx>
wrote:
[ ... ]
>
> It's not always easy for someone to say: "we must not release because
> of X and Y" - but those discussions have to happen or we'll end up
> with disaster. This happens all the time, sometimes the list is
> longer, and requires more careful planning. Lately (the last year) the
> list has been overwhelming - and that is upsetting people who are
> generally not involved in the process anyway.
>
> As we strive for higher quality with each alpha release, we're taking
> longer and longer to release them. That issue needs to be fixed
> somehow. Unfortunately it will probably require some amount of
> consensus rather than a handful of people who will try to do it and be
> shunned for their attempts.
Well said.
To give another example of what the quality concerns are: today, I've spent
more than 8 hours trying to fix a problem in Pe. Doing the actual fix in code
took about 15 mins. The rest of the time was spent trying to build the three
packages of Pe for x86_gcc2, x86 and x86_64. During that process, I've
noticed and fixed a problem in the binutils recipe, which caused binutils to
require flex without declaring that (so binutils would work on a full Haiku
installation but not in the chroots haikuporter uses for building. When that
was fixed and Pe had been built for all three architectures, I noticed that
it was using libpcre-8.33, which is the newest version in the haikuports
tree, but isn't contained in the actual HaikuPorts repositories used by
Haiku. So, the built Pe packages wouldn't work when I'd upload them ...
The same problems are to be expected (on a much larger scale) when preparing
a release, as all packages would have to be rebuilt. The current state of the
HaikuPorts tree simply isn't in a releasable state (neither the master nor
the testing_x86_64 branch apparently). Unless we fix this, any release
process is going to be a *huge* pain and the result is likely to push away
any developers trying to build something with haikporter. Having spent so
many hours of work on PM and the ports, I'd really dislike to release it in
this state.
What leaves me puzzled is that I keep saying this and people just seem to
ignore it. Why?
In order to solve the mess, someone needs to create a testing_x86_gcc2 branch
and put a consistent set of packages into it, which can then be used to build
and populate r1a5.
The actual building process (bootstrap) is rather automatic once a consistent
set of packages has been put together, but to get there is going to take a
while ...
cheers,
Oliver (drama queen)
Other related posts:
