On Tuesday, August 23, 2011 13:53:26 Michael wrote:
> Mike Frysinger wrote:
> > On Monday, August 22, 2011 15:21:23 Michael wrote:
> >> I wrote a script to search for discrepancies between linked libraries
> >> and what's defined in (R)DEPEND, with the intention of improving QA for
> >> minimal package installs.
> >> 
> >> The devmanual seems to suggest that it's a bad thing for packages to
> >> explicitly depend on packages in the system set, while many packages in
> >> the tree do indeed explicitly depend upon packages such as zlib and
> >> ncurses.
> > 
> > file bugs for missing zlib/ncurses deps.  if there are other packages
> > you're concerned with, ask now and we'll tell you.
> 
> What about packages such as bzip2, xz-utils, or really, app-arch/*?

assuming you're referring to them because the SRC_URI is compressed by the 
relevant formats, atm only bzip2 and gzip is allowed to be assumed.  
everything else has to be in DEPEND.  there is work/discussion to automate 
this (implicit unpack depends based on SRC_URI suffixes).

if your package is actually linking against any of the libs these guys 
provided (like libbz2.so), then you need to treat it like any other link-time 
dependency.

> Also, on a related note, would it be appropriate to report the following
> undeclared deps, given that the package in question already has DEPEND for
> gtk+?

like Nathan said, only if the package directly uses/links against the X11 
API/libs in question.  if the package only uses gtk+ funcs, then it should not 
be declaring a dependency on the X11 libs.

keep in mind that an app that only uses the gtk api can run on a system where 
gtk is not using X11 as its backend.  think raw framebuffer, win32, or 
something equally ridiculous.
-mike