On Mon, Oct 07, 2019 at 04:31:29PM +0200, Michal Privoznik wrote:
On 10/7/19 3:43 PM, Daniel P. Berrangé wrote:
> On Mon, Oct 07, 2019 at 03:03:51PM +0200, Michal Privoznik wrote:
> > On 10/3/19 12:51 PM, Daniel P. Berrangé wrote:
> > > Our syntax-check rule relies in various files imported during the gnulib
> > > bootstrap process. As we switch to meson & try to eliminate gnulib,
we
> > > need to take ownership of syntax-check.
> > >
> > > This patch series follows the approach taken for libosinfo & other
> > > projects wrt syntax-check rules, but goes a step further and eliminates
> > > alot of baggage we don't care about & merges all logic into one
file.
> > >
> > > Daniel P. Berrangé (4):
> > > build: import gnulib's syntax-check make rules
> > > build: remove all logic unrelated to syntax-check
> > > build: delete all syntax check rules we're skipping
> > > build: merge all syntax-check logic into one file
> > >
> > > Makefile.am | 7 +-
> > > bootstrap.conf | 4 -
> > > cfg.mk => build-aux/syntax-check.mk | 1141
+++++++++++++++++++++++++--
> > > build-aux/useless-if-before-free | 226 ++++++
> > > build-aux/vc-list-files | 113 +++
> > > 5 files changed, 1411 insertions(+), 80 deletions(-)
> > > rename cfg.mk => build-aux/syntax-check.mk (52%)
> > > create mode 100755 build-aux/useless-if-before-free
> > > create mode 100755 build-aux/vc-list-files
> > >
> >
> > Problem I'm seeing with these patches is that when I "make -j10 all
> > syntax-check check" (which is how I usually compile libvirt) then make
does
> > not order targets properly and:
>
> That's odd. There's no dependancy between 'syntax-check' and
'all' that
> I know of even in current git master. We can run a syntax-check without
> forcing an 'all' build. 'check' has a dep on 'all' obviously
so those
> two are ordered.
>
> IOW, I would have expect make to parallelize 'syntax-check' with execution
> of 'all' and 'check' already.
>
> I wonder what black magic is preventing this....
>
> > 1) I can see syntax-check output interleaved with 'all' output (or
even
> > 'check'), but which is worse:
> >
> > 2) there must be a dependency missing somewhere, because after 'git clean
> > -fxd; ./autogen.sh --system' I see the following error:
> >
> >
> > make[4]: Entering directory
> > '/home/zippy/work/libvirt/libvirt.git/gnulib/lib'
> > GEN sys/ioctl.h
> > GEN arpa/inet.h
> > GEN sys/select.h
> > GEN pthread.h
> > GEN sys/socket.h
> > GEN sched.h
> > GEN stdio.h
> > GEN sys/stat.h
> > GEN string.h
> > GEN stdlib.h
> > GEN sys/time.h
> > GEN sys/socket.h
> > GEN sys/stat.h
> > GEN sys/types.h
> > mv: cannot stat 'string.h-t': No such file or directory
> > GEN sys/uio.h
> > make[4]: *** [Makefile:4311: string.h] Error 1
> > make[4]: *** Waiting for unfinished jobs....
>
> What distro do you get this on ? I can't reproduce that problem yet
This is gentoo, but I'm able to reproduce on rawhide too. The important part
is to enable parallel build (make -j10 or so).
I think I might see the cause. The GNUmakefile has some black magic
in it to block parallelisation
ALL_RECURSIVE_TARGETS += $(AM_RECURSIVE_TARGETS)
ifneq ($(word 2, $(MAKECMDGOALS)), )
ifneq ($(filter $(ALL_RECURSIVE_TARGETS), $(MAKECMDGOALS)), )
.NOTPARALLEL:
endif
endif
This means that the three targets you list are built strictly in
order.
My change looses this rule and I thin that probably casues both
the failures you report. I'll repost a new series with a fix
for this
Regards,
Daniel
--
|:
https://berrange.com -o-
https://www.flickr.com/photos/dberrange :|
|:
https://libvirt.org -o-
https://fstop138.berrange.com :|
|:
https://entangle-photo.org -o-
https://www.instagram.com/dberrange :|