Re: new pkg_add library check breaks pkg_add -u

[Greg Troxel <gdt%lexort.com@localhost>]

Thomas Klausner <wiz%gatalith.at@localhost> writes:

> Ok, you win.
>
> If people want to keep using broken packages: here's a patch to try.
> It adds support for -f in in check_requires.
>
> It's untested, and you get to keep all the pieces if anything breaks.

I would like to see something like an environnment variable, so that I
can leave it set, and still get a warning, but not have failures.

I will try the patch; that is an improvement.

I think that what the check is trying to do is fundamentally a good
thing.  But what's not ok -- and why I said this was too much too soon
-- is two things:

  - "make package" with PKG_DEVELOPER=yes, can produce packages that
    pkg_add will then object to.   So I think that before pkg_add
    enforces this with more than a warning, make package has to not
    allow generation of such packages.   Essentially, if we have an
    integrity rule, it needs to be enforced at creation time.

  - "pkg_add -u", when this check fails, should leave the old package
    installed.  I know that was never defined as technically
    transactional, but it was mostly, and now it's really not.    This
    is very bad if make package creates defective packages and probably
    tolerable if it does not.

Thus, I think this should be downgraded to a warning until post-branch
and until after the "make package" problem is resolved.