On Sun, 29 Jul 2007 00:16:15 +0200
"Jaroslaw Swierczynski" <[EMAIL PROTECTED]> wrote:

> 2007/7/29, Jan de Groot <[EMAIL PROTECTED]>:
> >
> > When it does things other than it's documented, it's buggy.
> 
> Ever heard of "known bugs"? ;)
> 

That's different than "behaving the we way designed it."  

> > long long time ago we had this thing called staging and testing,
> > where we uploaded versions as bind-9.4.1-1t1, which is the test
> > version of bind-9.4.1-1. This logic also works for rc/beta
> > packages: gcc-4.2.1rc1-1 will be lower versioned than gcc-4.2.1-1.
> > There's logic behind it, so it's not a bug ;)
> 
> Behind rc being lower version - yes, there is. Let's say if a number
> is followed directly by a letter, it's higher version. If a number is
> followed directly by a letter which is followed directly by a number,
> it's lower version. At this point openssh came to my mind but I
> checked and there hasn't been a single version without p[1-9] in it.
> This makes sense to me.
> 

The problem is this is an arbitrary decision, and I can just as easily
come up with package names that break it:

Say 1.2.3pre is a prerelease.  By your rule above, it should be a
higher version than 1.2.3, but it's really a lower version.

I suppose we could impose some arbitrary new package versioning scheme
on packagers, but it would just be a hack.

"Packagers, ensure if your package version contains letters, AND the
version comes before a similar numbered version without letters, that
you sick an underscore before the letters."  seems kinda silly.

Although, I would be interested to know how gentoo's code handles
versioning, since it's apparently better than pacman for this case.

--
Travis

_______________________________________________
arch mailing list
[email protected]
http://archlinux.org/mailman/listinfo/arch

Reply via email to