Jan Wieck wrote:

> The following is not about this particular commit, but more general
> about what is going on.
>
> Yesterday I got an "RC1" tarball from Chris, only to discover that the
> ducttape test_x_pgbench scripts don't finish any more. The database
> comparision at the end has been moved out into a script part that is
> sourced in, but the source syntax doesn't take parameters, at least
> not in a standard Bourne shell. The change responsible for that was
> done nearly a month ago and not discovered. This makes me wonder on
> how many platforms this change was tested.

I know that particular part runs successfully on Linux and AIX; those
have been usefully enough different as to provide some "parallax."

> Now we have changes to the build system AFTER the tree has been tagged
> as Release Candidate.
>
> This is really not the way to ensure the quality of the upcoming
> release. I call this modus operandi a sure recipe for a release with a
> lot of trouble.

I have expressed the same irritation on IRC...

Some changes have gone in today with regards to the whole "--with-docs"
thing, and been tested by various of us on several (if not all
imaginable) systems. 

It appears that a future release of Red Hat's distributions will address
the problems users have been having with the documentation; in the
interim, we have a (finally confidence-boosting) way to avoid the errors.

> Chris, since the RC1 tarball was not uploaded thus far I suggest we
> fix the scripts, make sure they and the changes to the build system
> pass a decent number of different platforms and then "retag" (move the
> rc1 tag) for a new tarball.

In view of things, I think we need to step back and take a breath.

This week has 'bitten' us with two discoveries:

1.  There are quite a few platforms where building the SGML-based docs
is seriously inconvenient.  This led to some running around on configure
and makefile options to try to drop that step out "by default."  Quite a
bit has been resolved; Red Hat Software acknowledged a bug report and
has indicated they'll be applying some fixes to some subsequent
distribution release.

This led to a bunch of changes going to and fro with RPM packaging, too,
which led to a lot of patches, albeit localized to RPM-specific files.

2.  There were some platform dependancies still breaking; threads
detection on AIX, notably.

Discussion on IRC has pointed to trying to get to the RC in two phases
based on two dates...

Phase I:  Trying to finalize the nagging little details - Monday June 6th
------------------------------------------------------------------------------------------------------------------------
Everyone who is interested in testing things to get it released should
try out a CVS checkout between now and Monday on their favorite
platform, try to get a compile, and see what they get.

If you discover problems, please send an email to the list, and we'll
try to collect on IRC in order to respond to the problems.

Changes shouldn't just go into CVS; they need to be discussed on
list/IRC before going in once people agree.

Hopefully some small (ideally, nonexistent) set of changes will be found
Monday, which will lead to a BETA FINAL being generated.

Phase II: Testing for Release - Friday June 10th
-----------------------------------------------------------------------------------
People should then retest the BETA FINAL on their various favorite
platforms.

Changes shouldn't go in directly; patches should be submitted to the
list for discussion.  Once people agree, changes go in, but hopefully
there will be very little of that.

Anything that can wait until after 1.1 should wait...