[gentoo-dev] The bindist USE flag and RESTRICT=bindist

[Ulrich Mueller]

Hi,
We currently have two mechanisms to influence building of packages
that cannot be binary redistributed. Some 150 packages have
RESTRICT=bindist (unconditional, or USE conditional), whereas
27 packages have the bindist flag in their IUSE. Until very recently,
the two sets were practically disjoint, with only one single package
(www-client/chromium) in both sets.

As a consequence, neither building with the well-known USE=bindist nor
with ACCEPT_RESTRICT="* -bindist" (presumably less known and Portage
only?) will guarantee that only redistributable binaries will be
built [1].

So a few days ago I filed bug 541408 and dependent bugs, with the goal
that ebuilds with the bindist flag in their IUSE should also add
RESTRICT="!bindist? ( bindist )". However, it turned out that some 10
packages are using the bindist flag only in REQUIRED_USE but nowhere
else [2] (with foo being some feature flag):

   IUSE="bindist foo"
   REQUIRED_USE="bindist? ( !foo )"

IMHO, the bindist USE flag is redundant in these cases. So we should
get rid of the REQUIRED_USE and add a restriction instead:

   IUSE="foo"
   RESTRICT="foo? ( bindist )"

What do you think? Should we proceed in this direction?

While discussing the issue in #-releng, it was also suggested that a
news item should be prepared, to inform users that they cannot rely on
USE=bindist. (They cannot rely on it already now, though.) Does it
make sense to have a news item for this?

Ulrich


[1] We also have some 300 packages that are mirror restricted but not
    bindist restricted. It is likely that at least for part of them
    binaries cannot be redistributed either. That is a separate issue,
    though.
[2] Thanks to aballier for pointing this out in bug 541588.