Summary pretty much describes it; you're flipping off the sandbox during DEPENDS phase which *should* not occur. Move it to the phase that requires it.
Concerns openafs-1.2.10-r1, it seems to contain SANDBOX_DISABLED="1".
Bleh, nice of me to tell you which pkg is at fault :) Sorry about that. Yeah, openafs; adjust sandbox_disable w/in the phase itself that's needed, and only there.
Just as I suspected, I got no sandbox violations at all when just removing the line. But it wasn't a full test, I adapted the ebuild to not compile the kernel module, because I only have 2.6 on my x86's (this ebuild requires x86 AND linux-2.4). It'd be great if someone could test this on a x86 with a 2.4 kernel, so he could use the unmodified ebuild (except for the SANDBOX_DISABLED line). It seems clean, but as a novice gentoo developer I feel some hesitation towards altering a stable ebuild.
Will be fixed by bumping when openafs-1.2.13-r1 proves stable.
Nudge nudge What's going on with this? :)
Waiting to stabilize 1.2.13-r2, so we can move openafs-1.2.10-r1 out of the tree and remove the problem altogether. We've almost passed the nominal 30 days. The only thing that keeps me from stabilizing this is the fact that the documentation is currently inadequate to support users who try this new ebuild. And I currently haven't got much time to fix this :( If this is urgent, I could try to find some spare time to write a text file errata wrt the current documentation. If not I'd rather give it some more time and fix the documentation decently.
SANDBOX_DISABLED is actually something of a hack which I'll be pushing for removal soon (like, probably tonight if a grep of the tree turns up clean). You can do export SANDBOX_ON=0 in the phases that need to explicitly disable the sandbox; SANDBOX_DISABLED effectively functions as RESTRICT="sandbox" in it's current implementation, all it does is toggle SANDBOX_ON=0 when set. No huge rush for stabilization; the trick above, when shoved into appropriate phase hooks (src_compile fex) should address it.
In that case, we're back at comment #3. Seemant has been so kind to find someone willing and able (x86, 2.4-kernel, gentoo stable) to test just removing that SANDBOX_DISABLE line. We should hear from him tomorrow.
I thank Martin Adler for testing this. He reported a clean build, so I just removed the SANDBOX_DISABLE from the ebuild. Changing status to fixed.
Cool, thank you :)