Feature test bikeshed summary

[Rich Felker <dalias@aerifal.cx>]

Here's the summary of email results so far. I know I'm probably
omitting responses from a few who've only spoken on IRC and not posted
here, but it looks like we have a fairly even split, especially if I
add myself to the mildly pro-change side.

Status quo camp:
nsz
Luca
Bobby Bingham
nwmcsween

Pro-change camp:
Gregor (XSI+BSD_specific)
orc (kitchen sink)
Isaac Dunham (XSI+BSD)
philomath (unspecified featureset, option 3)

Would anyone who wants to see the behavior changed be willing to try a
couple different potential profiles with the pkgsrc build and see how
many packages work with just _XOPEN_SOURCE=600 or just _BSD_SOURCE or
similar? In principle, I think the nicest variant of option 3 would be
the minimal set of features that makes the majority of programs just
work, without exposing nasty stuff that pollutes the namespace (like
sys/types.h and sys/sysmacros.h getting included via stdlib.h...). If
we do make the change, it would be nice if programs depending on
stupid things like that still break so we can get them fixed upstream;
I think our "it breaks with musl" bug reports would carry a lot more
weight if we were providing a robust, clean yet compatibility-oriented
featureset by default.

Anyway, no decisions yet, but it'd be nice to do some tests..

Rich