[csw-maintainers] [policy] Re: feature patching, and naming

[Peter Bonivart]

On Sat, Feb 5, 2011 at 5:57 PM, Philip Brown <phil at bolthole.com> wrote:
> (side comment: whatever we use, conceptually becomes "rev=" reborn.
> but at least using ",xx" is much less ugly than
>   REV=xxxx,rev=yyy)

The main problem is all the variants that have existed over the years
since the policy is not clear on this. Here's some examples from the
current catalog not counting all ugly ones with "_rev=foo":

autossh 1.4,REV=2009.06.25_b
boincclient 6.7.4,REV=2009.02.05_r17141
mlterm 2.9.3,REV=2007.01.08_i386only
mplayer 1.0.0,REV=2007.09.22.rc1try3
radiance 3.8,REV=2008.02.11,i386only
squid 2.7,REV=2010.10.05_STABLE9

Let's take this opportunity to set this policy clearly and not be up
to interpretation formatting wise. I would like to see some freedom
for the content left of REV but the formatting of all parts should be
strict.

I like "1.2,p,REV=1234.56.78" much better than
"1.2,REV=1234.56.78_rev=p", I agree with you there. A quick glance
through the pkgutil code didn't reveal any negatives but I would like
to set up some tests and get back to you on that.

Looking at what's written now at
http://www.opencsw.org/extend-it/contribute-packages/build-standards/#versioning,
I think it should be allowed (but not to be used unless needed) for
more numeric fields than YYYY.MM.DD since that makes it easy to make a
package the same day that distinguishes itself from the other one by
using, e.g., YYYY.MM.DD.HH.mm. Pkgutil processes any number of these
fields until one package "wins". I don't know if you do but it's a
suggestion.

I also think we should make it clear that the version string is now in
three parts where the middle one is optional, the content of it should
be from a fixed list which from the start should contain "p" for
patched. Something like this:

1.2.3[,x],REV=YYYY.MM.DD[.xx]

What do you think?

/peter