Re: epoch fix?

[Thorsten Glaser <tg@debian.org>]

Bart Martens <bartm <at> debian.org> writes:

> Michael Biebl wrote :
> > The usage of really (...) that you don't have to fix all r-deps to
include
> > the the epoch in the Build-Depends.
> 
> Why would adding an epoch cause the need for adding the epoch in the
> build-dependent packages ?

Interestingly enough, I was just thinking about writing a message that
said something to the point of, using “really” for a once-off botched
upload would be okay.

Then I considered versioning on the r-deps.

Now imagine the following:

• foo 1.0-1 uploaded
• bar 1.0-1 uploaded, depends on foo-dev (>= 1.0)
• foo 1.1-1 uploaded
• bar 1.1-1 uploaded, depends on foo-dev (>= 1.1)
• foo 1.1-1really1.0-1 uploaded

That’s a massive “oops” in both cases. Funnily enough, using the
epoch will, yes, force an update of all r-deps, BUT it’s the only
sane way for the r-deps because otherwise the saga continues:

• bar 1.1-2 uploaded, depends on foo-dev (>= 1.1),
   foo-dev (<< 1.1-1really1.0-1~) | foo-dev (>= 1.1-2~)
• foo 1.1-2 uploaded
• foo 1.1-2really1.1-1 uploaded………

For r-deps it’s much clearer if you use epochs. It’ll disallow
some versions, but it’s easier to see what’s really depended on.
Also, once a r-dep includes the “really” it gets visually uglier
too. (It’s bad enough we have build-depends with Ubuntu version
numbers in them, but I can understand that too.)

> I agree with that.  I think version ordering should be preserved forever.

This is also an interesting point, yes.

bye,
//mirabilos