svn commit: r276052 - head

[Garrett Cooper]

On Jan 1, 2015, at 10:58, Garrett Cooper <yaneurabeya at gmail.com> wrote:

> On Jan 1, 2015, at 10:51, Dimitry Andric <dim at FreeBSD.org> wrote:
> 
>> On 01 Jan 2015, at 19:42, Garrett Cooper <yaneurabeya at gmail.com> wrote:
>>> 
>>> On Jan 1, 2015, at 10:38, Garrett Cooper <yaneurabeya at gmail.com> wrote:
>>>> On Jan 1, 2015, at 9:50, Dimitry Andric <dim at FreeBSD.org> wrote:
>> ...
>>>>> No, just MK_INFO=yes, as I said.  That is what Garrett tried to fix in
>>>>> this commit, but which only works for native builds.
>>>>> 
>>>>> I'll be coming up with a fix shortly.
>>>> 
>>>> OBJTREE is being populated with TARGET.TARGET_ARCH twice… That’s the bug.
>>>> 
>>>> Unfortunately this commit only made this issue apparent. Sorry for not testing with a non-native build :(...
>>> 
>>> I suspect this will fix it. Going to do a tinderbox run then I’ll put it out for review:
>>> 
>>> Index: Makefile.inc1
>>> ===================================================================
>>> --- Makefile.inc1	(revision 276492)
>>> +++ Makefile.inc1	(working copy)
>>> @@ -267,8 +267,7 @@
>>> 		MK_LLDB=no MK_TESTS=no
>>> 
>>> # build-tools stage
>>> -TMAKE=		MAKEOBJDIRPREFIX=${OBJTREE} \
>>> -		${BMAKEENV} ${MAKE} ${WORLD_FLAGS} -f Makefile.inc1 \
>>> +TMAKE=		${BMAKEENV} ${MAKE} ${WORLD_FLAGS} -f Makefile.inc1 \
>>> 		TARGET=${TARGET} TARGET_ARCH=${TARGET_ARCH} \
>>> 		DESTDIR= \
>>> 		BOOTSTRAPPING=${OSRELDATE} \
>> 
>> Hm, wasn't MAKEOBJDIRPREFIX set to that value on purpose, originally?
>> And if so, won't removing it break some assumption elsewhere?  That
>> said, this was added in r54684, 15 years ago, so maybe this is slightly
>> out of date. :-)
>> 
>> What I've tried instead, is this:
>> 
>> Index: Makefile.inc1
>> ===================================================================
>> --- Makefile.inc1       (revision 276480)
>> +++ Makefile.inc1       (working copy)
>> @@ -270,6 +271,7 @@
>> TMAKE=         MAKEOBJDIRPREFIX=${OBJTREE} \
>>               ${BMAKEENV} ${MAKE} ${WORLD_FLAGS} -f Makefile.inc1 \
>>               TARGET=${TARGET} TARGET_ARCH=${TARGET_ARCH} \
>> +               WORLDTMP=${WORLDTMP} \
>>               DESTDIR= \
>>               BOOTSTRAPPING=${OSRELDATE} \
>>               SSP_CFLAGS= \
>> 
>> I'm currently testing this with a buildworld, and while it isn't
>> finished yet, it has died either, until now. :)
> 
> Removing MAKEOBJDIRPREFIX got me through build-tools with TARGET/TARGET_ARCH=powerpc and a native build as well.
> 
> MAKEOBJDIRPREFIX gets set above it:
> 
> 190 .if ${MACHINE} == ${TARGET} && ${MACHINE_ARCH} == ${TARGET_ARCH} && !defined(CROSS_BUILD_TESTING)
> 191 OBJTREE=        ${MAKEOBJDIRPREFIX}
> 192 .else
> 193 OBJTREE=        ${MAKEOBJDIRPREFIX}/${TARGET}.${TARGET_ARCH}
> 194 .endif
> ...
> 270 TMAKE=          ${BMAKEENV} ${MAKE} ${WORLD_FLAGS} -f Makefile.inc1 \
> 
> That being said, my commit was probably broken in a cross-build environment anyhow, as it was installing to ${WORLDTMP}/usr/bin, not ${WORLDTMP}/legacy/usr/bin (the latter is where all of the build tools go).

Oh cool. More make tinderbox broken-ness… https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=196425 .
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 496 bytes
Desc: Message signed with OpenPGP using GPGMail
URL: <http://lists.freebsd.org/pipermail/svn-src-head/attachments/20150101/adbe6b53/attachment.sig>