Re: patch make struct protosw pr_input non-variadic

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

    Date:        Sun, 15 May 2016 19:50:19 -0400
    From:        Tyler Retzlaff <rtr%netbsd.org@localhost>
    Message-ID:  <de1ac713-5693-2aca-2c7a-0341630c1b29%netbsd.org@localhost>

Just a comment off the side ...

  | Have we addressed all the concerns that have been raised?

And not about that (from what I can see the change looks OK, but I have
not looked in detail at the code) but...

  | Do we need more discussion or shall we proceed?

Back in the good old days - I mean 4BSD, and certainly 4.1 .. 4.3 BSD
(I wasn't involved as much with 4.4, so can't really say for that one
way or the other) no-one would have even dared suggest a change anything
like this without backing it up with performance measurements, gprof
data showing that (at least) it was not worse then before, and before and
after throughput measurements.

We seem to have lost that here, and comments are made (both side) based upon
speculation about what seems likely might hurt, improve, or not affect
performamce, rather than by any actual testing.

It is different for an obvious bug fix (code that did not work before has
0 performance after all, anything is better) but when the change is intended
in some way to improve things, it ought be shown that it does.   The
"cleanliness" improvement seems obvious here, but if there were to be a
performance downside (and I am not saying that there is, I have not measured
it either) then there would need to be a decision on whether or not the code
improvement is worth the cost (of course, what sometimes happens, is that
cleaner code is also faster, and everyone is happy.)

kre