Skip site navigation (1)Skip section navigation (2)
Date:      Thu, 24 Apr 2014 12:14:06 -0600
From:      Ian Lepore <ian@FreeBSD.org>
To:        "Justin T. Gibbs" <gibbs@scsiguy.com>
Cc:        src-committers@FreeBSD.org, Jilles Tjoelker <jilles@stack.nl>, svn-src-all@FreeBSD.org, Dimitry Andric <dim@FreeBSD.org>, svn-src-head@FreeBSD.org, Warner Losh <imp@bsdimp.com>
Subject:   Re: svn commit: r263778 - in head: bin lib lib/clang sbin share/mk usr.bin usr.sbin
Message-ID:  <1398363246.61646.73.camel@revolution.hippie.lan>
In-Reply-To: <EA0A7D4D-2F55-4372-8825-BE907B127CBA@scsiguy.com>
References:  <201403262230.s2QMUdH6021943@svn.freebsd.org> <AA90F6B0-3A7A-473D-82C2-CFDFD263E9AC@gmail.com> <20140327181245.GA69977@stack.nl> <7A86F5E9-DBE9-4D3F-B166-C02F8386B722@FreeBSD.org> <1398086131.1124.371.camel@revolution.hippie.lan> <A3C5D651-D1A4-46AF-900A-6D019AD2D39C@bsdimp.com> <EA0A7D4D-2F55-4372-8825-BE907B127CBA@scsiguy.com>

next in thread | previous in thread | raw e-mail | index | archive | help
I couldn't think of a way to do that, because the directory itself isn't
the dependency.  That is, whether one directory has a newer/older
timestamp than the other isn't a function of the build process.  The
fact that someone else mentioned inserting .WAIT into the list is
something I took as a confirmation that the regular dependency mechanism
wouldn't work (so maybe I didn't think about it hard enough).

-- Ian

On Thu, 2014-04-24 at 11:17 -0600, Justin T. Gibbs wrote:
> Why wouldn=92t we fix these with explicit subdir dependencies (e.g: sub=
dira: subdirb)?  I=92m pretty sure I had this working with a hacked up bs=
d.subdir.mk at a previous job, but my memory is hazy.
>=20
> =97
> Justin
>  =20
> On Apr 21, 2014, at 7:35 AM, Warner Losh <imp@bsdimp.com> wrote:
>=20
> > (sorry for the top post)
> >=20
> > This looks good to my eye. I=92d be tempted to toss in a comment abou=
t the
> > __wait=3D.WAIT construct is due to the primitive nature of bmake=92s =
parser
> > so it runs afoul of the .for/.if construction rules and this is neede=
d to expand
> > the for variable. It tripped me up when I looked at it, until I recal=
led a comment
> > from similar code in NetBSD.
> >=20
> > Warner
> >=20
> > On Apr 21, 2014, at 7:15 AM, Ian Lepore <ian@FreeBSD.org> wrote:
> >=20
> >> On Thu, 2014-03-27 at 20:44 +0100, Dimitry Andric wrote:
> >>> On 27 Mar 2014, at 19:12, Jilles Tjoelker <jilles@stack.nl> wrote:
> >>>> On Thu, Mar 27, 2014 at 11:05:00AM -0600, Warner Losh wrote:
> >>>>> On Mar 26, 2014, at 4:30 PM, Dimitry Andric <dim@freebsd.org> wro=
te:
> >>>>>> Author: dim
> >>>>>> Date: Wed Mar 26 22:30:38 2014
> >>>>>> New Revision: 263778
> >>>>>> URL: http://svnweb.freebsd.org/changeset/base/263778
> >>>>=20
> >>>>>> Log:
> >>>>>> Add a SUBDIR_PARALLEL option to bsd.subdir.mk, to allow make to =
process
> >>>>>> all the SUBDIR entries in parallel, instead of serially.  Apply =
this
> >>>>>> option to a selected number of Makefiles, which can greatly spee=
d up the
> >>>>>> build on multi-core machines, when using make -j.
> >>>>=20
> >>>>>> This can be extended to more Makefiles later on, whenever they a=
re
> >>>>>> verified to work correctly with parallel building.
> >>>>=20
> >>>>> Why not have this =91opt out=92 rather than =91opt in=92 like it =
is now? Are
> >>>>> there any known bad dependencies this introduces?
> >>>>=20
> >>>> I'm paranoid about build systems ;) It is easy to add dependencies
> >>>> across directories and as long as directories are built in sequenc=
e,
> >>>> nothing goes wrong.
> >>>>=20
> >>>> In fact, I had enabled SUBDIR_PARALLEL in sys/modules/Makefile as =
well,
> >>>> but this caused mysterious failures with some kernels such as mips
> >>>> ADM5120.
> >>>=20
> >>> There are a bunch of other parts that don't really like parallel bu=
ilds
> >>> at the moment.  For example, gnu/usr.bin/binutils needs its librari=
es
> >>> (libbfd.a, etc) built first, before it can link the programs.  Simi=
lar
> >>> for gnu/usr.bin/cc, which needs libiberty, libcpp, etc before being=
 able
> >>> to build the rest of gcc.
> >>>=20
> >>> Most of these cases can hopefully be solved by adding .WAIT targets=
 at
> >>> strategic points in the SUBDIR lists, but this also needs a bit of =
extra
> >>> logic in bsd.subdir.mk.
> >>>=20
> >>> -Dimitry
> >>>=20
> >>=20
> >> It turns out I needed the .WAIT functionality to use SUBDIR_PARALLEL=
 for
> >> $work, so I came up with the attached, does this look okay to commit=
?
> >>=20
> >> -- Ian
> >>=20
> >> diff -r 67802e319fc6 share/mk/bsd.subdir.mk
> >> --- a/share/mk/bsd.subdir.mk	Sun Apr 20 21:01:07 2014 -0600
> >> +++ b/share/mk/bsd.subdir.mk	Mon Apr 21 06:59:37 2014 -0600
> >> @@ -4,10 +4,10 @@
> >> # The include file <bsd.subdir.mk> contains the default targets
> >> # for building subdirectories.
> >> #
> >> -# For all of the directories listed in the variable SUBDIRS, the
> >> +# For all of the directories listed in the variable SUBDIR, the
> >> # specified directory will be visited and the target made. There is
> >> # also a default target which allows the command "make subdir" where
> >> -# subdir is any directory listed in the variable SUBDIRS.
> >> +# subdir is any directory listed in the variable SUBDIR.
> >> #
> >> #
> >> # +++ variables +++
> >> @@ -42,7 +42,7 @@ distribute:
> >>=20
> >> _SUBDIR: .USE
> >> .if defined(SUBDIR) && !empty(SUBDIR) && !defined(NO_SUBDIR)
> >> -	@${_+_}for entry in ${SUBDIR}; do \
> >> +	@${_+_}for entry in ${SUBDIR:N.WAIT}; do \
> >> 		if test -d ${.CURDIR}/$${entry}.${MACHINE_ARCH}; then \
> >> 			${ECHODIR} "=3D=3D=3D> ${DIRPRFX}$${entry}.${MACHINE_ARCH} (${.TA=
RGET:realinstall=3Dinstall})"; \
> >> 			edir=3D$${entry}.${MACHINE_ARCH}; \
> >> @@ -57,7 +57,7 @@ distribute:
> >> 	done
> >> .endif
> >>=20
> >> -${SUBDIR}: .PHONY
> >> +${SUBDIR:N.WAIT}: .PHONY
> >> 	${_+_}@if test -d ${.TARGET}.${MACHINE_ARCH}; then \
> >> 		cd ${.CURDIR}/${.TARGET}.${MACHINE_ARCH}; \
> >> 	else \
> >> @@ -65,13 +65,18 @@ distribute:
> >> 	fi; \
> >> 	${MAKE} all
> >>=20
> >> +__wait=3D.WAIT
> >> .for __target in all all-man checkdpadd clean cleandepend cleandir \
> >>    depend distribute lint maninstall manlint \
> >>    obj objlink realinstall regress tags \
> >>    ${SUBDIR_TARGETS}
> >> .ifdef SUBDIR_PARALLEL
> >> +__subdir_targets=3D
> >> .for __dir in ${SUBDIR}
> >> -${__target}: ${__target}_subdir_${__dir}
> >> +.if ${__wait} =3D=3D ${__dir}
> >> +__subdir_targets+=3D .WAIT
> >> +.else
> >> +__subdir_targets+=3D ${__target}_subdir_${__dir}
> >> ${__target}_subdir_${__dir}: .MAKE
> >> 	@${_+_}set -e; \
> >> 		if test -d ${.CURDIR}/${__dir}.${MACHINE_ARCH}; then \
> >> @@ -85,7 +90,9 @@ distribute:
> >> 		fi; \
> >> 		${MAKE} ${__target:realinstall=3Dinstall} \
> >> 		    DIRPRFX=3D${DIRPRFX}$$edir/
> >> +.endif
> >> .endfor
> >> +${__target}: ${__subdir_targets}
> >> .else
> >> ${__target}: _SUBDIR
> >> .endif
> >=20
> >=20
> >=20
>=20
>=20





Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?1398363246.61646.73.camel>