Re: toolchain target
Dimitry Andric wrote: > Hmm, is this still not fixed? This was broken by Simon's large > "meta-mode" commit r284345. but I would assume Baptiste's fixes after > that might have fixed it. I can't test this myself at the moment, due to > ENOTIME... I think https://reviews.freebsd.org/D2860 will fix this. It puts include of make.conf etc back where they were and introduces new early include. ___ freebsd-current@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-current To unsubscribe, send any mail to "freebsd-current-unsubscr...@freebsd.org"
Re: toolchain target
Andriy Gapon wrote:
> > do you have anything interesting in /etc/make.conf?
>
> Thank you for the hint -- __MAKE_CONF=/dev/null SRC_CONF=/dev/null fix the
> problem.
>
> Now I am trying to figure out what the problem is.
The problem will be that I shifted the include of make.conf and etc
to earlier in sys.mk so that they can provide pointer to external
toolchains.
>
> My make.conf:
> .if defined(CC)
> .if ${CC} == gcc
which will break all this - see the UPDATING entry.
> CPUTYPE?=k8-sse3
> .else
> CPUTYPE?=amdfam10
> .endif
> .endif
> CFLAGS+= -O2 -fno-strict-aliasing -pipe
> CFLAGS+= -fno-omit-frame-pointer
> CXXFLAGS+= -O2 -fno-strict-aliasing -pipe
The root problem is that a global file like /etc/make.conf is a very
fragile concept - makes it almost impossible to evolve the the build.
Especially if you put anything in there more complex that setting global
knobs.
Any use of += etc, makes it sentitive to the point of inclusion.
It is useful to include local.sys.mk (hence src.sys.mk) very early.
So that they can do CC?= blah etc. before sys.mk does
But the current semantic is that make.conf should be included first
which means it needs to be included early .. which breaks everyone who
has "interesting" stuff there.
> And src.conf:
> WITH_DEBUG_FILES=yes
> WITH_CTF=yes
What is perhaps really needed is yet another include that can happen
early and another later during sys.mk processing
We could then move include of make.conf back
Will investigate...
___
freebsd-current@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-current
To unsubscribe, send any mail to "freebsd-current-unsubscr...@freebsd.org"
Re: toolchain target
On 18/06/2015 09:22, Andriy Gapon wrote:
> My make.conf:
> .if defined(CC)
> .if ${CC} == gcc
> CPUTYPE?=k8-sse3
> .else
> CPUTYPE?=amdfam10
> .endif
> .endif
> CFLAGS+= -O2 -fno-strict-aliasing -pipe
> CFLAGS+= -fno-omit-frame-pointer
> CXXFLAGS+= -O2 -fno-strict-aliasing -pipe
>
[snip]
> Looks like my rather innocent manipulations of CFLAGS could be causing the
> problem.
> Without my make.conf:
> mkdep -f .depend -a
> -I/usr/devel/svn2/head/lib/clang/libllvmsupport/../../../contrib/llvm/include
> -I/usr/devel/svn2/head/lib/clang/libllvmsupport/../../../contrib/llvm/tools/clang/include
> -I/usr/devel/svn2/head/lib/clang/libllvmsupport/../../../contrib/llvm/lib/Support
> -I.
> -I/usr/devel/svn2/head/lib/clang/libllvmsupport/../../../contrib/llvm/../../lib/clang/include
> -DLLVM_ON_UNIX -DLLVM_ON_FREEBSD -D__STDC_LIMIT_MACROS
> -D__STDC_CONSTANT_MACROS
> -DLLVM_DEFAULT_TARGET_TRIPLE=\"x86_64-unknown-freebsd11.0\"
> -DLLVM_HOST_TRIPLE=\"x86_64-unknown-freebsd11.0\" -DDEFAULT_SYSROOT=\"\"
> -I/usr/obj/usr/devel/svn2/head/tmp/legacy/usr/include -std=c++11
> -stdlib=libc++
> /usr/devel/svn2/head/lib/clang/libllvmsupport/../../../contrib/llvm/lib/Support/APFloat.cpp
> ...
>
> With my make.conf:
> mkdep -f .depend -a -std=c++11 -stdlib=libc++
> /usr/devel/svn2/head/lib/clang/libllvmsupport/../../../contrib/llvm/lib/Support/APFloat.cpp
> ...
>
> All the preprocessor flags (-I, -D) are gone.
>
Or rather it's CXXFLAGS that is affected.
Looks like now in my environment the initial value of CXXFLAGS is set by
make.conf instead of being derived from CFLAGS in mk/sys.mk. This breaks
expectation of many makefiles that flags added to CFLAGS will also be appear in
CXXFLAGS.
Apparently, this problem is documented in UPDATING 20150616, but it does not
cover all possible variables affected by the change...
So, I guess that I could just remove CXXFLAGS from my make.conf, because it is
quite redundant, but in some cases people may still want to set it.
Or I could copy the CXXFLAGS assignment line from mk/sys.mk to make.conf, but
that's a slippery slope towards having to copy all of mk/sys.mk to make.conf.
Or maybe the historic behavior could be restored and whatever prompted the
change could be done in a different fashion?
--
Andriy Gapon
___
freebsd-current@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-current
To unsubscribe, send any mail to "freebsd-current-unsubscr...@freebsd.org"
Re: toolchain target
On 18 Jun 2015, at 08:24, Garrett Cooper wrote: > > On Jun 17, 2015, at 23:22, Andriy Gapon wrote: ... >> Looks like my rather innocent manipulations of CFLAGS could be causing the >> problem. >> Without my make.conf: >> mkdep -f .depend -a >> -I/usr/devel/svn2/head/lib/clang/libllvmsupport/../../../contrib/llvm/include >> -I/usr/devel/svn2/head/lib/clang/libllvmsupport/../../../contrib/llvm/tools/clang/include >> -I/usr/devel/svn2/head/lib/clang/libllvmsupport/../../../contrib/llvm/lib/Support >> -I. >> -I/usr/devel/svn2/head/lib/clang/libllvmsupport/../../../contrib/llvm/../../lib/clang/include >> -DLLVM_ON_UNIX -DLLVM_ON_FREEBSD -D__STDC_LIMIT_MACROS >> -D__STDC_CONSTANT_MACROS >> -DLLVM_DEFAULT_TARGET_TRIPLE=\"x86_64-unknown-freebsd11.0\" >> -DLLVM_HOST_TRIPLE=\"x86_64-unknown-freebsd11.0\" -DDEFAULT_SYSROOT=\"\" >> -I/usr/obj/usr/devel/svn2/head/tmp/legacy/usr/include -std=c++11 >> -stdlib=libc++ >> /usr/devel/svn2/head/lib/clang/libllvmsupport/../../../contrib/llvm/lib/Support/APFloat.cpp >> ... >> >> With my make.conf: >> mkdep -f .depend -a -std=c++11 -stdlib=libc++ >> /usr/devel/svn2/head/lib/clang/libllvmsupport/../../../contrib/llvm/lib/Support/APFloat.cpp >> ... >> >> All the preprocessor flags (-I, -D) are gone. > > Oh really…? This is going to break a _lot_ of peoples’ custom make.confs :(... Hmm, is this still not fixed? This was broken by Simon's large "meta-mode" commit r284345. but I would assume Baptiste's fixes after that might have fixed it. I can't test this myself at the moment, due to ENOTIME... -Dimitry signature.asc Description: Message signed with OpenPGP using GPGMail
Re: toolchain target
On Jun 17, 2015, at 23:22, Andriy Gapon wrote:
> On 18/06/2015 02:26, Simon J. Gerraty wrote:
>> Andriy Gapon wrote:
>>>> Seems like there is some problem with 'toolchain' target in the latest
>>>> head.
>>>> Output of running `make toolchain TARGET=i386` on amd64 system can be found
>>>> here: http://dpaste.com/3RD3C4V
>>>>
>>>
>>> AFAICS, it still worked as of r283188.
>>
>> There has been a clang update since then.
>>
>> I just did make -j12 toolchain TARGET=i386 ok
>> do you have anything interesting in /etc/make.conf?
>
> Thank you for the hint -- __MAKE_CONF=/dev/null SRC_CONF=/dev/null fix the
> problem.
>
> Now I am trying to figure out what the problem is.
>
> My make.conf:
> .if defined(CC)
> .if ${CC} == gcc
> CPUTYPE?=k8-sse3
> .else
> CPUTYPE?=amdfam10
> .endif
> .endif
> CFLAGS+= -O2 -fno-strict-aliasing -pipe
> CFLAGS+= -fno-omit-frame-pointer
> CXXFLAGS+= -O2 -fno-strict-aliasing -pipe
>
> And src.conf:
> WITH_DEBUG_FILES=yes
> WITH_CTF=yes
> WITHOUT_INET6=YES
> WITHOUT_PROFILE=YES
> WITHOUT_FORTRAN=YES
> WITHOUT_I4B=YES
> WITHOUT_IPFILTER=YES
> WITHOUT_ATM=YES
> WITHOUT_IPX=YES
> WITHOUT_LPR=yes
> WITHOUT_ZONEINFO=yes
> MALLOC_PRODUCTION=yes
> LOADER_BZIP2_SUPPORT=yes
> LOADER_FIREWIRE_SUPPORT=yes
>
>
> Looks like my rather innocent manipulations of CFLAGS could be causing the
> problem.
> Without my make.conf:
> mkdep -f .depend -a
> -I/usr/devel/svn2/head/lib/clang/libllvmsupport/../../../contrib/llvm/include
> -I/usr/devel/svn2/head/lib/clang/libllvmsupport/../../../contrib/llvm/tools/clang/include
> -I/usr/devel/svn2/head/lib/clang/libllvmsupport/../../../contrib/llvm/lib/Support
> -I.
> -I/usr/devel/svn2/head/lib/clang/libllvmsupport/../../../contrib/llvm/../../lib/clang/include
> -DLLVM_ON_UNIX -DLLVM_ON_FREEBSD -D__STDC_LIMIT_MACROS
> -D__STDC_CONSTANT_MACROS
> -DLLVM_DEFAULT_TARGET_TRIPLE=\"x86_64-unknown-freebsd11.0\"
> -DLLVM_HOST_TRIPLE=\"x86_64-unknown-freebsd11.0\" -DDEFAULT_SYSROOT=\"\"
> -I/usr/obj/usr/devel/svn2/head/tmp/legacy/usr/include -std=c++11
> -stdlib=libc++
> /usr/devel/svn2/head/lib/clang/libllvmsupport/../../../contrib/llvm/lib/Support/APFloat.cpp
> ...
>
> With my make.conf:
> mkdep -f .depend -a -std=c++11 -stdlib=libc++
> /usr/devel/svn2/head/lib/clang/libllvmsupport/../../../contrib/llvm/lib/Support/APFloat.cpp
> ...
>
> All the preprocessor flags (-I, -D) are gone.
Oh really…? This is going to break a _lot_ of peoples’ custom make.confs :(...
signature.asc
Description: Message signed with OpenPGP using GPGMail
Re: toolchain target
On 18/06/2015 02:26, Simon J. Gerraty wrote:
> Andriy Gapon wrote:
>>> Seems like there is some problem with 'toolchain' target in the latest head.
>>> Output of running `make toolchain TARGET=i386` on amd64 system can be found
>>> here: http://dpaste.com/3RD3C4V
>>>
>>
>> AFAICS, it still worked as of r283188.
>
> There has been a clang update since then.
>
> I just did make -j12 toolchain TARGET=i386 ok
> do you have anything interesting in /etc/make.conf?
Thank you for the hint -- __MAKE_CONF=/dev/null SRC_CONF=/dev/null fix the
problem.
Now I am trying to figure out what the problem is.
My make.conf:
.if defined(CC)
.if ${CC} == gcc
CPUTYPE?=k8-sse3
.else
CPUTYPE?=amdfam10
.endif
.endif
CFLAGS+= -O2 -fno-strict-aliasing -pipe
CFLAGS+= -fno-omit-frame-pointer
CXXFLAGS+= -O2 -fno-strict-aliasing -pipe
And src.conf:
WITH_DEBUG_FILES=yes
WITH_CTF=yes
WITHOUT_INET6=YES
WITHOUT_PROFILE=YES
WITHOUT_FORTRAN=YES
WITHOUT_I4B=YES
WITHOUT_IPFILTER=YES
WITHOUT_ATM=YES
WITHOUT_IPX=YES
WITHOUT_LPR=yes
WITHOUT_ZONEINFO=yes
MALLOC_PRODUCTION=yes
LOADER_BZIP2_SUPPORT=yes
LOADER_FIREWIRE_SUPPORT=yes
Looks like my rather innocent manipulations of CFLAGS could be causing the
problem.
Without my make.conf:
mkdep -f .depend -a
-I/usr/devel/svn2/head/lib/clang/libllvmsupport/../../../contrib/llvm/include
-I/usr/devel/svn2/head/lib/clang/libllvmsupport/../../../contrib/llvm/tools/clang/include
-I/usr/devel/svn2/head/lib/clang/libllvmsupport/../../../contrib/llvm/lib/Support
-I.
-I/usr/devel/svn2/head/lib/clang/libllvmsupport/../../../contrib/llvm/../../lib/clang/include
-DLLVM_ON_UNIX -DLLVM_ON_FREEBSD -D__STDC_LIMIT_MACROS -D__STDC_CONSTANT_MACROS
-DLLVM_DEFAULT_TARGET_TRIPLE=\"x86_64-unknown-freebsd11.0\"
-DLLVM_HOST_TRIPLE=\"x86_64-unknown-freebsd11.0\" -DDEFAULT_SYSROOT=\"\"
-I/usr/obj/usr/devel/svn2/head/tmp/legacy/usr/include -std=c++11
-stdlib=libc++
/usr/devel/svn2/head/lib/clang/libllvmsupport/../../../contrib/llvm/lib/Support/APFloat.cpp
...
With my make.conf:
mkdep -f .depend -a -std=c++11 -stdlib=libc++
/usr/devel/svn2/head/lib/clang/libllvmsupport/../../../contrib/llvm/lib/Support/APFloat.cpp
...
All the preprocessor flags (-I, -D) are gone.
--
Andriy Gapon
___
freebsd-current@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-current
To unsubscribe, send any mail to "freebsd-current-unsubscr...@freebsd.org"
Re: toolchain target
Andriy Gapon wrote: > > Seems like there is some problem with 'toolchain' target in the latest head. > > Output of running `make toolchain TARGET=i386` on amd64 system can be found > > here: http://dpaste.com/3RD3C4V > > > > AFAICS, it still worked as of r283188. There has been a clang update since then. I just did make -j12 toolchain TARGET=i386 ok do you have anything interesting in /etc/make.conf? ___ freebsd-current@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-current To unsubscribe, send any mail to "freebsd-current-unsubscr...@freebsd.org"
Re: toolchain target
On 17/06/2015 23:42, Andriy Gapon wrote: > > Seems like there is some problem with 'toolchain' target in the latest head. > Output of running `make toolchain TARGET=i386` on amd64 system can be found > here: http://dpaste.com/3RD3C4V > AFAICS, it still worked as of r283188. -- Andriy Gapon ___ freebsd-current@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-current To unsubscribe, send any mail to "freebsd-current-unsubscr...@freebsd.org"
toolchain target
Seems like there is some problem with 'toolchain' target in the latest head. Output of running `make toolchain TARGET=i386` on amd64 system can be found here: http://dpaste.com/3RD3C4V -- Andriy Gapon ___ freebsd-current@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-current To unsubscribe, send any mail to "freebsd-current-unsubscr...@freebsd.org"
