On Sun, Mar 26, 2017 at 03:56:18PM +0000, Antti Kantee wrote:
On 26/03/17 09:11, Francesco Lattanzio wrote:
Yes and no. Yes in that it's a bug, but no in that configure.ac is also
wrong. exporting CC in buildrump.sh in the snippet that runs configure
should fix it, at least assuming you're running buildrump.sh with
"CC=clang ./buildrump.sh". Want to try it (and submit a pull req if so)?
Done. See below.
I don't see a pull req in buildrump.sh?
But speaking of your patch to librumpuser, please just add the necessary
things for FreeBSD using the existing conventions. For example, you
cannot add a requirement for -lpthread, since some systems have the same
functionality in libc without supplying a libpthread at all, and they
would now be broken.
Thinking about it, that might be the reason the
configure script uses AC_TRY_COMPILE instead of LINK to check for
pthread stuff -- can't remember for sure anymore.
Furthermore, there'sCorrect. I'll fix this too.
nothing which says that you cannot have pthread_set_name_np() without
pthread_np.h; you might as well use #ifdef FreeBSD and not bother with
autoconf at all.
Autoconf scripts work best when you test the minimum,
and draw absolutely no a=>b type conclusions along the way.
Does -Wimplicit instead of -Wimplicit-function-declaration work? That
flag is already used by buildrump.sh, so reusing it in the configure
script would avoid adding a dependency on a new compiler flag.
It's not necessary to post the autogenerated parts. I'll regen anyway
as part of the merging and pulling and grinding process.
So in summary, thanks, but the patch needs a bit more work to be
conservative enough. Somewhere down the line I got tired of fixing
things back and forth and got careful ...