On 03 Jun 2015 00:28, Michael Palimaka wrote:
> On 02/06/15 23:52, Mike Frysinger wrote:
> > On 02 Jun 2015 23:07, Michael Palimaka wrote:
> >> On 02/06/15 21:38, Mike Frysinger wrote:
> >>> On 02 Jun 2015 20:47, Michael Palimaka wrote:
> >>>> On 02/06/15 17:04, Michał Górny wrote:
> >>>>> Dnia 2015-06-02, o godz. 03:58:35
> >>>>> "Michael Sterrett (mr_bones_)" <mr_bones_@gentoo.org> napisał(a):
> >>>>>> -DEPEND="readline? ( sys-libs/readline )
> >>>>>> +DEPEND="readline? ( sys-libs/readline:0 )
> >>>>>
> >>>>> This should be actually := (or :0=) for both deps since gnugo links to
> >>>>> them. This also applies to your remaining 'warning silencing' commits.
> >>>>
> >>>> Why? Blindly adding the subslot dep is a bad idea.
> >>>
> >>> in this particular case, the subslot usage is what we want since we're 
> >>> compiling+linking against it.  using readline:0 vs readline is still an 
> >>> improvement though.
> >>>
> >>> we also want a subslot on ncurses since we compile+link against it.
> >>>
> >>> i think it's pretty uncommon to use readline in a package and not want a 
> >>> subslot.  your package would have to be doing something uncommon like
> >>> dlopening it since the only thing readline provides is a library ...
> >>
> >> Neither readline nor ncurses define an explicit subslot, so I don't know
> >> what their future meaning might be.
> > 
> > their meaning would be the reasonable one -- to track the SONAME.  while it 
> > hasn't been deployed yet (due to those packages being on EAPI=4), i don't know 
> > what other value you'd expect it to be.  they've both broken their SONAMEs in 
> > the past.  readline in particular has been every major version (4.x, 5.x, 6.x).
> 
> Since you've clarified the future meaning for ncurses/readline, it's not
> a problem.
> 
> The point was it's not a good idea to use the operator unless without
> knowing what it means for the package in question (since a subslot can
> be used to handle a number of different situations, as I wrote previously).

you make a reasonable point, but i'd consider some of your examples as 
(ab|mis)use of subslots.  their purpose is to track ABI changes so as
to signal rebuilds.  if they've been appropriated for other uses, then
perhaps those libraries are doing it wrong ?  i expect subslots to do
one thing (what PMS describes/intends) and would be surprised to find
out that some packages are doing something else.
-mike