Re: Release procedures improvements.

Greg Hudson <ghudson@MIT.EDU> writes:
> Why not just name the prerelease tarballs something irrelevant, than?
> Version numbers are cheap, but they're not *that* cheap; it's a little
> confusing for 1.1.3 to be succeeded by 1.1.7. If we have a way to avoid
> spending them on prerelease testing (and it seems like it should be
> really, really simple to avoid it), then we should avoid it.

Well, the problem is that, in this proposal, the prerelease unpacks
into a directory named *exactly* what the release will be named. That
is, we may call the prerelease

   subversion-20050405-PRERELEASE.tar.gz

but when unpacked, it would still create

   subversion-1.2.0-rc1/

According to Justin's proposal, we have to treat that unpackment name
as unique forever. If we find a problem with the prerelease, then we
cannot simply release

   subversion-20050411-PRERELEASE.tar.gz

but still have it unpack into

   subversion-1.2.0-rc1/

It must unpack into something new, probably

   subversion-1.2.0-rc2/

(I'm just using these as examples; I'm not proposing that we change
how we're doing the actual 1.2.0-rc* releases.)

So, I don't see how keeping the version number out of the prerelease
tarball's name saves us anything. We still have to toss the version
number if problem is discovered.

-Karl

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Received on Wed Apr 6 19:39:40 2005