svn commit: r276052 - head

[Dimitry Andric]

On 01 Jan 2015, at 17:41, Ian Lepore <ian at freebsd.org> wrote:
> 
> On Thu, 2015-01-01 at 16:50 +0100, Dimitry Andric wrote:
>> On 22 Dec 2014, at 05:52, Garrett Cooper <ngie at FreeBSD.org> wrote:
>>> 
>>> Author: ngie
>>> Date: Mon Dec 22 04:52:24 2014
>>> New Revision: 276052
>>> URL: https://svnweb.freebsd.org/changeset/base/276052
>>> 
>>> Log:
>>> Build selective portions of gnu/usr.bin/texinfo as part of build-tools to
>>> ensure that building on a host without makeinfo (i.e. a host where
>>> make delete-old -DWITHOUT_INFO was run), then building with MK_INFO == yes
>>> doesn't manifest in build errors when building info pages
>> ..
>>> @@ -1398,6 +1403,16 @@ build-tools: .MAKE
>>> 		${MAKE} DIRPRFX=${_tool}/ depend && \
>>> 		${MAKE} DIRPRFX=${_tool}/ all
>>> .endfor
>>> +.for _tool in \
>>> +    ${_texinfo}
>>> +	${_+_}@${ECHODIR} "===> ${_tool} (obj,depend,all)"; \
>>> +		cd ${.CURDIR}/${_tool} && \
>>> +		${MAKE} DIRPRFX=${_tool}/ obj && \
>>> +		${MAKE} DIRPRFX=${_tool}/ depend && \
>>> +		${MAKE} DIRPRFX=${_tool}/ all && \
>>> +		${MAKE} DIRPRFX=${_tool}/ install DESTDIR=${WORLDTMP}
>>> +.endfor
>> 
>> Strangely, this hunk seems to work incorrectly for non-native builds.
>> For example, I tried a TARGET=arm buildworld just now, and that dies
>> with the following error:
>> 
>> [...]
>>>>> stage 2.3: build tools
>> [...]
>> ===> gnu/usr.bin/texinfo/makeinfo (obj,depend,all,install)
>> [...]
>> gzip -cn /usr/src/gnu/usr.bin/texinfo/makeinfo/../../../../contrib/texinfo/doc/makeinfo.1 > makeinfo.1.gz
>> sh /usr/src/tools/install.sh -s -o root -g wheel -m 555   makeinfo /usr/obj/arm.arm/arm.arm/usr/src/tmp/usr/bin/makeinfo
>> install: /usr/obj/arm.arm/arm.arm/usr/src/tmp/usr/bin/makeinfo: No such file or directory
>> *** Error code 71
>> 
>> Specifically, the /usr/obj/arm.arm/arm.arm directory is incorrect, there
>> should be only one "arm.arm" in that path.  I don't really understand
>> how that value comes to pass, though.  When I put an echo statement just
>> before the make install, which shows the values of ${_tool} and
>> ${WORLDTMP}, it prints:
>> 
>> DEBUG: _tool=gnu/usr.bin/texinfo/makeinfo, WORLDTMP=/usr/obj/arm.arm/arm.arm/usr/src/tmp
>> 
>> So for some strange reason, ${WORLDTMP} is incorrect at that point?  I
>> think something is appending one path component too many...
>> 
>> -Dimitry
>> 
> 
> try TARGET_ARCH=arm

No, that does not help, unfortunately.  The problem seems to be that in
Makefile.inc1, TMAKE is passing MAKEOBJDIRPREFIX=${OBJTREE} via its
environment, and this changes the value of WORLDTMP to an incorrect
value.

At first it looked like I could just force the correct WORLDTMP value
via the command line instead, e.g. like this:

Index: Makefile.inc1
===================================================================
--- Makefile.inc1       (revision 276480)
+++ Makefile.inc1       (working copy)
@@ -270,6 +270,7 @@
 TMAKE=         MAKEOBJDIRPREFIX=${OBJTREE} \
                ${BMAKEENV} ${MAKE} ${WORLD_FLAGS} -f Makefile.inc1 \
                TARGET=${TARGET} TARGET_ARCH=${TARGET_ARCH} \
+               WORLDTMP=${WORLDTMP} \
                DESTDIR= \
                BOOTSTRAPPING=${OSRELDATE} \
                SSP_CFLAGS= \

However, this leads to other problems further down the road, where it
gets errors installing parts of libllvmsupport and tblgen.

So at this point, non-native builds with MK_INFO=yes are broken.

-Dimitry

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 194 bytes
Desc: Message signed with OpenPGP using GPGMail
URL: <http://lists.freebsd.org/pipermail/svn-src-all/attachments/20150101/144f30e0/attachment.sig>