Re: Automated report: NetBSD-current/i386 test failure

[Robert Elz <kre%munnari.OZ.AU@localhost>]

    Date:        Sun,  7 Aug 2016 07:17:49 +0000 (UTC)
    From:        NetBSD Test Fixture <bracket%NetBSD.org@localhost>
    Message-ID:  <20160807071749.1515.63938%babylon5.netbsd.org@localhost>

  | The newly failing test case is:
  | 
  |     lib/libc/sys/t_mlock:mlock_err

I believe this is related to the addr 0 changes.   The test expects
mlock(NULL, ...) to give ENOMEM (and the man page agrees) but now
the error returned is EINVAL - whick according to mlock(2) should
imply that the length is negative (it isn't) or the address is
not aligned (it is - 0 is aligned with everything.)

The test actually expects EINVAL if vm.minaddress > 0 -- which I don't
believe reflects what the man page says will happen, so that is probably
just the test testing what actually happens, rather than testing what
should happen (I have seen tests that are rigged to "just work" by simply
accepting whatever happens is correct - rather than having the test fail
until a bug is fixed before - that's a real nuisance.)

The kernel should probably be changed to return ENOMEM in all these
cases, and the test changed to stop checking vm.minaddress - that would
make both kernel and test match the man page.

Alternatively, the man page could be changed to make EINVAL apply to
invalid (for any reason) addresses, and all mention of ENOMEM removed,
and the test updated (which would simplify it.)

kre