On Monday 31 March 2008 02:29:10 Brian Harring wrote:
> Going to reiterate this one more time; the proposal is simple enough;
> if it's an implicit 0 via cpv parsing, it should *not* be explicitly
> specified on disk.  'diffball-1.0_alpha0.ebuild' can just as easily be
> specified as 'diffball-1.0_alpha.ebuild', 'diffball-1.0-r0.ebuild' can
> just as easily be specified as 'diffball-1.0.ebuild'.
>
> Reiterating the gain: consistancy on disk, consistancy in dealing with
> PV/PVR.  As you keep point out, PV does vary- having the results of
> ebuild execution change depending on whether or not you name your
> ebuild 'diffball-1.0_alpha0.ebuild' or 'diffball-1.0_alpha.ebuild' is
> *not* a feature, it is what you would classically call a "design
> misfeature".  PVR for 'diffball-1.0-r0.ebuild' being '1.0' instead of
> '1.0-r0' is yet another argument for punting explicit -r0 on disk-
> it's a gotcha, design flaw in the format.
>
> It's a simple tweak, with no real loss of functionality (lots of
> claims, no single hard proof thus far).  As spanky has stated, there
> *is* a loss of ease of use in a small subset of ebuilds- worst case,
> .06% ebuilds.  Personally, I don't consider that minority worth
> preserving since preserving that means leaving open several gotchas in
> ebuild writing, and complicates interactions (both pm and dev).
>
> So... there it is.  Would be rather nice to have ebuild devs weight in
> on this one, rather then just package manager monkeys also (they're
> the ones bound most by the change after all).  Laid out the advantages
> to this- kindly lay out the disadvantages, where this makes things
> worse if you're looking for a response.

I struggle to see the technical gain of banning -r0 while allowing _alpha0 and 
002 which have the exact same technical issue. Forcing people to hard code 
versions or use sucky substitutions such as ${PV/_alpha/_alpha0} or whatever 
you'd use to convert cpufrequtils-2 to cpufrequtils-002 is clearly a loss of 
feature.

I agree that explicit -r0 on disk isn't really useful given the current Gentoo 
policy on revision bumps. But I think we established on bug #166522 [1] that 
we don't want to make restrictions based on what is useful. Otherwise we 
should ban _alpha_beta_pre_p too...

[1] https://bugs.gentoo.org/show_bug.cgi?id=166522

-- 
Bo Andresen
Gentoo KDE Dev

Attachment: signature.asc
Description: This is a digitally signed message part.

Reply via email to