On Sun 2019-05-05 07:49:16 -0300, David Bremner wrote:
> My take away from that is that I should probably squash these two
> commits into one.

I've read this series and it seems reasonable to me.  i agree that we
should be testing messages smaller than 4096 octets, either by squashing
the two commits into one, or by only conditionally marking the tests as
known-broken.  it's not too bad to mark it known-broken conditionally if
you want to do the two-stage commit:

 [ size -lt 4096 ] || test_subtest_known_broken

but it's a little weird, because it's only *actually* "known broken"
depending on your platform's buffer size -- i could test this series in
2030 on debian GNU/Hurd 15 and find that because our buffer size is now
65536 none of the tests are actually broken yet :P

I also learned some extra bash from Tomi's nit-picking review, for which
i'm always grateful.  thanks, Tomi! :)  And thanks Rob for finding this!

Please merge this series.

       --dkg