Re: CVS commit: src/sys/arch

[Klaus Klein <kleink%kleink.org@localhost>]

On Mon, May 19, 2014 at 06:16:06PM -0400, Christos Zoulas wrote:
> On May 19, 11:28pm, kleink%kleink.org@localhost (Klaus Klein) wrote:
> -- Subject: Re: CVS commit: src/sys/arch
> 
> | What's the purpose of this?  ssize_t is meant to be "a count of
> | bytes or an error indication", so having SSIZE_MIN doesn't make
> | sense.  Nor does its only in-tree, broken use in libevent; searching
> | outside our tree I wasn't able to come up with a use of EV_SSIZE_MIN
> | except for the aforementioned regression test.
> 
> The purpose of this is to make it easier to query the range of the type.
> (ssize_t)-1 would work too.

But having SSIZE_MIN at all is at odds with the scope of ssize_t
(I've quoted the definition above).


> | (I'll be the first to admit I'm somewhat late in voicing my concerns,
> | but I was ... busy.  *sigh*)
> 
> Heh. It would have been easier if that was a bit sooner, so I could
> remember the exact situation. From what I can remember, I searched for
> it, I saw that others had it too (linux, etc.) so I decided to add it.

I now see this change coincides with
external/bsd/libevent/dist/include/event2/util.h rev. 1.2.

Actually no: glibc as no such thing.  Googling for it I've found several
applications jumping through hoops to define some kind of SSIZE_MIN without
ever using it afterwards.  libevent's regression test might be its
lone user.

Since the netbsd-6 branch precedes this change I propose the timely
removal of SSIZE_MIN and resurrection of a suitable version of the
kludge libevent used before.  (I doubt the removal of EV_SSIZE_MIN
would cause much grief except for the regression test; I won't mind
asking the libevent maintainers about that constant.)


- Klaus