Non-prefix safe ports

[John Hein]

Jan Beich wrote at 17:53 +0100 on Nov 23, 2015:
 > "John Hein" <jh-fbml at snkmail.com> writes:
 > 
 > > Jan Beich wrote at 01:18 +0100 on Nov 22, 2015:
 > >
 > >  > "Mikhail T." <mi+thun at aldan.algebra.com> writes:
 > >  > 
 > >  > > Building out a new machine from scratch using /opt (rather than the
 > >  > > usual /usr/local) as LOCALBASE, I found the following two ports, which
 > >  > > break:
 > >  > 
 > >  > poudriere testport -P would have been easier. That said it would be nice
 > >  > to have them as periodic exp-runs. I'm sure there're way more ports
 > >  > with broken PREFIX != LOCALBASE e.g., anything lua-*.
 > >  > 
 > >  > >
 > >  > >   * multimedia/openh264 - the files built by the port get installed into
 > >  > >     stage/usr/local/... instead of stage/opt -- I had to manually move
 > >  > >     the files around before "make install" succeeded
 > >  > 
 > >  > This affects all ports with USES=webplugin. Fixing isn't trivial, so
 > >  > better file a bug or it will be forgotten.
 > >  > 
 > >  > >   * graphics/gtk-update-icon-cache - at configure-stage, the port
 > >  > >     wrongly complaints, that the cairo installed on the system does not
 > >  > >     support X11. After I created a symlink: /usr/local -> ../opt, the
 > >  > >     problem went away.
 > >  > 
 > >  > make(1) flow control is evaluated at parsing time. Try using sh(1)
 > >  > conditional to postpone it until the target is run.
 > >
 > > I've had this patch in my ports tree forever - one of those patches
 > > for which I just never got around to the last step: actually
 > > submitting a bug.  Instead of hard-coding a path, it uses the
 > > pkg-config tool to detect.  But it postpones the detection until
 > > the target just as Jan did.
 > 
 > The issue is actually a regression unrelated to PREFIX != LOCALBASE. I've
 > submitted your version (slightly improved) on Phabricator and added a comment
 > to avoid repeating it again.

Understood... I was just latching on to that part since I had
a local change.  Thanks for taking it.


 > https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=166279
 > https://reviews.freebsd.org/D4260
 > 
 > --
 > X11 check indicates a deficiency in the framework. There's currently no
 > standard way to require some options enabled in dependencies.

I have a simple improvement to allow pkg-config specification without
full pathname in *DEPENDS in my tree for years.  I'll dust it off
and send it out.  It may not be all that's needed (I'm sure it's
not), but it's a start anyway.