Re: [HACKERS] Current sources?

The Hermit Hacker <scrappy(at)hub(dot)org> writes:
> On Mon, 25 May 1998, David Gould wrote:
>> - if the build and regression is good, then a snapshot is made into a
>> "last known good" location.

> Actually, ummm...I've been considering removing the snapshot's
> altogether, now that anoncvs works.

It may be worth pointing out that cvs allows anyone to retrieve *any*
prior state of the code. This opens up a great number of options that
a simple periodic snapshot does not. I think it's worth continuing the
snapshot series for those who don't want to install cvs for some reason,
but that probably won't be the primary access method anymore.

The thing that I thought was worth adopting from David's list was the
nightly automatic regression test run. Assuming that there are cycles
to spare on the server, posting the results of a build and regression
test attempt would help provide a reality check for everyone. (It'd
be too bulky to send to the mailing lists, and not worth archiving
anyway; perhaps the output could be put up as a web page at
postgresql.org?)

This sort of fiasco could be minimized if everyone got in the habit of
running regression tests before submitting their patches. Here I have
to disagree with Marc's opinion that it's not really important whether
pre-alpha code works. If the tree is currently broken, that prevents
everyone else from running regression tests on what *they* are doing,
and consequently encourages the submission of even more code that hasn't
been adequately tested. I would like to see a policy that you don't
check in code until it passes regression test for you locally. We will
still have failures because of (a) portability problems --- ie it works
at your site, but not for someone else; and (b) unforeseen interactions
between patches submitted at about the same time. But hopefully those
will be relatively easy to track down if the normal state is that things
mostly work.

We might also consider making more use of cvs' ability to track multiple
development branches. If several people need to cooperate on a large
change, they could work together in a cvs branch until their mods are
finished, allowing them to share development files without breaking the
main branch for others.

regards, tom lane