Re: versions / suffixes in experimental

[Russ Allbery <rra@debian.org>]

Simon McVittie <smcv@debian.org> writes:

> Approach 1, which is (IMO) better when the changes you are making in
> experimental are truly experimental, like enabling features or patches
> whose medium-term future you're not sure about:

> 2.2.5-5+exp1, ... or -6~exp1, ... or whatever to experimental
> 2.2.5-6 to unstable

> Approach 2, which is (IMO) better when the changes you are making in
> experimental are the main line of development, and you're only not
> uploading to unstable because you're trying to avoid a freeze or getting
> tangled into a transition or something:

> 2.2.5-6, -7, ... to experimental
> 2.2.5-5+deb8u1 to unstable (if needed)

> (i.e. in approach 2 you're treating the unstable branch as
> stable-updates to a stable release that doesn't exist yet).

> Either can work. I've done both in the past.

Yes.  To some extent it's a matter of style, and different people will
have different styles, and that's okay.

My personal feeling on this is that I believe people generally over-think
version numbers and add more complexity than is actually required.  I
therefore have a personal rule that I use the simplest version numbers
that I can get away with in any situation.  I've not seen much practical
reason to prefer the sequence:

    2.2.5-6~exp1  (experimental)
    2.2.5-6~exp2  (experimental)
    2.2.5-7       (unstable)

to:

    2.2.5-7       (experimental)
    2.2.5-8       (experimental)
    2.2.5-9       (unstable)

and the latter is simpler, so I pretty much always use that.

Either way, you have to do something "weird" if you need to upload
something to unstable from a different branch, particularly if you don't
want the unstable version to be newer than the experimental version (which
I almost never do).

The only argument that I've found convincing for putting an "exp" in the
experimental package versions is if they're *really* experimental, as in
"this may all be a horrible idea that I will disclaim in the morning"
sort of experimental, and it's really important to get that information in
front of the user in the version number.

But in general I think people are way too conservative about not just
using the next version number.  Integers are cheap, and you won't ever run
out.  :)  It's akin to the problem of endless releases of software widely
used all over the world that still has a 0.x version number.  Just call it
1.0 already.

-- 
Russ Allbery (rra@debian.org)               <http://www.eyrie.org/~eagle/>