On Sat, 2019-09-21 at 08:43 +0200, Fabian Groffen wrote:
> On 20-09-2019 22:53:53 +0200, Michał Górny wrote:
> > On Fri, 2019-09-20 at 13:46 -0700, Zac Medico wrote:
> > > If we take this underscore rule to its logical extreme, then we should
> > > rename python_targets_python3_7 to python_targets_python3-7, yes?
> > 
> > Believe me, I would have done that already if not the fact that with all
> > the dependency logic around here it would be totally destructive to all
> > Gentoo systems.
> 
> Honestly, with this reasoning, why force other packages to go through
> USE-flag renaming in that case?  A major consumer of USE_EXPAND isn't
> sticking to the rule, which makes any benefit of it moot.  Tools cannot
> assume the last underscore separates the USE_EXPAND var from its value,
> users cannot see what is the value either, without knowledge.

The major consumer is fixable.  Sure, it will take years but that's
better than leaving things wrong forever and saying wrong is good.

> Why not teach our tools (equery, quse, etc.) to print these USE-flags
> like Portage does?  (looking them up to be valid expands)
> Then users have nothing to be confused about (no distinction between
> foo_bar and FOO="bar"), and new USE_EXPANDS cannot be
> silently/accidentially introduced.

I don't see how that solves the problem.  More tools having distinct
output don't change the fact that anyone with a bit of ebuild knowledge
will say 'this looks like USE_EXPAND' while looking at it, independently
of what some tools would say.

-- 
Best regards,
Michał Górny