On 12/12/17 10:26, Oliver Schonrock wrote: > It seems that this file has the > answer: > > /usr/src/share/mk/bsd.cpu.mk >
...
>
> So I should be able to put CPUTYPE=sandybridge into /etc/make.conf
> and...happy days? No more nasty surprises?
Actually I am not entirely convinced that will work
eventhough /usr/src/share/mk/bsd.cpu.mk does a lovely job of defining
MACHINE_CPU based on what CPUTYPE we put in /etc/make.conf, it does not
appear that this value actually gets user (ie translated into CFLAGS). I
couldn't find anything in /usr/src or /usr/ports which uses MACHINE_CPU.
From a comment at bottom of /usr/src/share/mk/sys.mk, it seems that this
is the responsibility of the ports themselves:
# Hack for ports compatibility. Historically, ports makefiles have
# assumed they can examine MACHINE_CPU without including anything
# because this was automatically included in sys.mk. For /usr/src,
# this file has moved to being included from bsd.opts.mk. Until all
# the ports files are modernized, and a reasonable transition
# period has passed, include it while we're in a ports tree here
# to preserve historic behavior.
and indeed some ports do that, eg
/usr/ports/security/cryptopp/Makefile
. if ${MACHINE_CPU:Msse2}
USES+= compiler:c++14-lang
CXXFLAGS+= -msse2
yet security/libargon2 does not reference MACHINE_CPU at all.
So it might be all for nought?
Does the port maintainer for security/libargon2 need to add such a
CFLAGS switch based on MACHINE_CPU into the Makefile?
Or is there some "magic" which feeds the "feature info" from MACHINE_CPU
into clang via CFLAGS (or some other way) automatically and my grep'ing
didn't find it?
--
Oliver Schönrock
email : oli...@schonrocks.com
signature.asc
Description: OpenPGP digital signature
