build order race after SUBDIR_PARALLEL (264303)

[Andreas Nilsson]

On Fri, May 23, 2014 at 2:26 PM, Ian Lepore <ian at freebsd.org> wrote:

> On Fri, 2014-05-23 at 08:56 +0200, Luigi Rizzo wrote:
> > Hi,
> > I have recently hit a problem when building stable/10 with
> > "make -j 8" which i tracked down to svn 264303, the
> > MFC of the SUBDIR_PARALLEL build feature.
> >
> > With a large number of parallel tasks (and GCC;
> > it is a race condition so timing matters), the build fails
> > during the toolchain target with this:
> >
> > --- sig_party.o ---
> > cc   -O2 -pipe
>  -I/usr/home/luigi/FreeBSD/R10/lib/libngatm/../../sys/contrib/ngatm
> -I/usr/home/luigi/FreeBSD/R10/../usr/obj-pico-amd64/usr/home/luigi/FreeBSD/R10/lib/libngatm
> > -I/usr/home/luigi/FreeBSD/R10/lib/libngatm/../../contrib/ngatm/libngatm
> -std=gnu99  -fstack-protector -Wsystem-headers -Werror -Wall
> -Wno-format-y2k -W -Wno-unused-parameter -
> > Wstrict-prototypes -Wmissing-prototypes -Wpointer-arith -Wreturn-type
> -Wcast-qual -Wwrite-strings -Wswitch -Wshadow -Wunused-parameter
> -Wcast-align -Wchar-subscripts -Winline
> > -Wnested-externs -Wredundant-decls -Wold-style-definition
> -Wno-pointer-sign -c
> /usr/home/luigi/FreeBSD/R10/lib/libngatm/../../sys/contrib/ngatm/netnatm/sig/sig_party.c
> -o sig_
> > party.o
> > --- all_subdir_libproc ---
> >
> /usr/home/luigi/FreeBSD/R10/../usr/obj-pico-amd64/usr/home/luigi/FreeBSD/R10/tmp/usr/bin/ld:
> cannot find -lsupc++
> > *** [libproc.so.2] Error code 1
> >
> > make[5]: stopped in /usr/home/luigi/FreeBSD/R10/lib/libproc
> > 1 error
> > ...
> >
> >
> > Turns out that before SUBDIR_PARALLEL, libsupc++ was built
> > before libproc thus satisfying the dependency; but with
> > -j XXX we cannot guarantee the ordering and there is no
> > explicit constraint or .ORDER in the makefiles to build
> > things in the correct order.
> >
> > This is a race condition so you may or may not see the problem
> > depending on what you are building and where.
> >
> > I am seeing the problem consistently on stable/10 after 264303
> > when building toolchain with GCC (not clang) on an 8-core machine
> > and make -j 8
> >
> > Building the above with -j 4 seem to work fine,
> > as it works building HEAD with -j 8
> >
> > cheers
> > luigi
>
> Aha.  I asked for exactly this info on the commit list (so you can
> ignore that request).  Warner & I kicked around some ideas on how to
> solve this a while back, I'll see if I can make some progress towards
> better dependency controls for SUBDIR_PARALLEL this weekend.
>
> -- Ian
>
>
> Sorry for a perhaps a cheeky question, but is this not all due to not
telling make the dependencies, and/or due to "recursive" make?

When is decided to learn make (although that was mainly gnu make ) I found
the following http://aegis.sourceforge.net/auug97.pdf very enlightening.

I do realise that such a redesign is a tremendous effort.

Best regards
Andreas