I haven't checked the dev list, but I'm sure this topic must have been
discussed before. So, apologies ... but, maybe it's time to re-discuss anyway?
The variant +universal seems to have a special status among variants: If
+universal is specified for building then 'port' will check all dependencies to
make sure they are already installed as +universal (or cannot be) and if not
try to upgrade them to be so; 'port' will print an error message if it cannot
accomplish this task. This check / upgrade works quite well for this specific
variant, and so I'm wondering if it can also be applied to other variants such
as +debug, +x11, +quartz, and so forth.
For example: If I have nothing already installed (except 'port') and do "sudo
port install gtk2 +quartz" then 'port' will go off and try to apply the +quartz
variant to all dependencies -- those that can use it will do so; those that
cannot will ignore it. But after all of the dependencies are installed, if I
do "sudo port install gtk2 +x11" then 'port' does not go back and try to
reinstall the dependencies with the +x11 variant. Instead it either (1)
requires the Portfile to manually check to make sure the installed variant is
correct & if not then tell the user; and/or (2) tries to build the port and
will in general eventually generate an error since the correct dependency
variant isn't installed. Portfiles that come to mind that do this to some
degree include libglade2, gtk2, pango, and cairo.
For Qt4, in the PortGroup I put in a manual check for the +debug variant, which
covers just Qt4 itself. So, for example, amarok would want to not just check
Qt4 (which it does automatically via the qt4 PortGroup), but ideally it would
also check kdelibs4, phonon, and a bunch of others. One could come up with a
PortGroup for doing such checking (for example: create a list of variants to
check for, with each one containing a list of {the dependency port name and a
file specific to this port and variant} to check) -- but that's what
"depends_FOO" provides info for to a first order & hence it seems silly to
duplicate the information that's already there (IIUC sort of like what
"archcheck" did, but then deprecated because it was incorporated into
"universal").
It seems to me that changing all variants to use the same behavior as is done
by +universal makes sense (there could also be an opt-out keyword for ports
that just want a unique variant that does not propagate to other ports), even
if it would require "standardizing" what a variant means. I'm wondering if
this behavior is desirable to other developers ... - MLD
_______________________________________________
macports-dev mailing list
[email protected]
http://lists.macosforge.org/mailman/listinfo.cgi/macports-dev
