Date: Sun, 13 Jun 2010 10:10:06 -0600 (MDT) From: "M. Warner Losh" <imp@bsdimp.com> To: jroberson@jroberson.net Cc: des@des.no, arch@freebsd.org Subject: Re: Change to config(8) for OFED Message-ID: <20100613.101006.549303830244327332.imp@bsdimp.com> In-Reply-To: <alpine.BSF.2.00.1006122328470.1435@desktop> References: <86zkz05bra.fsf@ds4.des.no> <20100613.023223.972525811890973035.imp@bsdimp.com> <alpine.BSF.2.00.1006122328470.1435@desktop>
next in thread | previous in thread | raw e-mail | index | archive | help
In message: <alpine.BSF.2.00.1006122328470.1435@desktop>
Jeff Roberson <jroberson@jroberson.net> writes:
: On Sun, 13 Jun 2010, M. Warner Losh wrote:
:
: > In message: <86zkz05bra.fsf@ds4.des.no>
: > Dag-Erling Sm?rgrav <des@des.no> writes:
: > : "M. Warner Losh" <imp@bsdimp.com> writes:
: > : > but NORMAL_C is
: > : >
: > : > NORMAL_C= ${CC} -c ${CFLAGS} ${WERROR} ${PROF} ${.IMPSRC}
: > : >
: > : > and .IMPSRC is null. That seems like a bug to me, but I'm not
: > sure if
: > : > the bug is that .IMPSRC is computed wrong, or if it really should
: > be
: > : > ${.ALLSRC}
: > :
: > : .IMPSRC is the *implicit* source and is only valid in *implicit*
: > rules.
: > : Consider the following:
: > :
: > : .SUFFIXES: .foo .bar
: > : .foo.bar: foo2bar.sh
: > : /bin/sh foo2bar.sh ${.IMPSRC} ${.TARGET}
: > :
: > : When building hello.bar from hello.foo, .IMPSRC is "hello.foo" but
: > .ALLSRC
: > : is "hello.foo foo2bar.sh".
: >
: > Yes, it appears to be set only when the target matches the object, so
: > the .c.o rule matches, even if the more specific one is used (where
: > there is not a match).
: >
: > : GNU make has a way of defining more complex implicit rules (pattern
: > : rules). If we had something similar, we could do:
: > :
: > : foo_%.o: %.c
: > : ${NORMAL_C}
: >
: > Yes. I'm starting to think that this would be necessary to implement
: > this properly. Or, as Jeff suggested, have config generate a complete
: > compile line with -o foo_bar.o -c bar.c in the line...
:
: Here's what I ended up with:
:
: In kern.pre.mk:
:
: +OFED_C_NOIMP= ${CC} -c ${OFEDCFLAGS} ${WERROR} ${OFEDNOERR} ${PROF}
: +OFED_C= ${OFED_C_NOIMP} ${.IMPSRC}
:
: I don't use NORMAL_C as linux doesn't use some warnings that we
: specify and I need to modify the include path to search the linux
: wrapper paths first.
:
: in files:
: +ofed/drivers/infiniband/hw/mlx4/ah.c optional mlx4 \
: + no-depend obj-prefix "mlx4_" \
: + compile-with "${OFED_C_NOIMP} -I$S/ofed/drivers/infiniband/hw/mlx4/"
:
: Generated Makefile:
: mlx4_ah.o: $S/ofed/drivers/infiniband/hw/mlx4/ah.c
: ${OFED_C_NOIMP} -I$S/ofed/drivers/infiniband/hw/mlx4/
: $S/ofed/drivers/infiniband/hw/mlx4/ah.c
:
:
: And the patch to config:
: http://people.freebsd.org/~jeff/objprefix.diff
:
: The full source path is only emitted when an object prefix is set. If
: you don't like this diff please speak now before I commit. It will go
: to my private branch for ofed first but I'm sure I'll forget about it
: before I merge.
:
: I still need to change config so it will do proper depends for files
: which require extra includes. Before then I just specify no-depend
: for ib.
:
: Is it possible to set make variables within targets? I can export
: shell variables since that's an executable command. I could then
: refer to the shell variables in the scripts but that seems really
: gross.
No. You can't set make variables within targets, unless you are doing
subtargets:
foo.o: bar.c
(cd subdir; make FRED=WILMA BARNEY=BETTY gerbils)
Warner
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20100613.101006.549303830244327332.imp>