[OE-core] [discussion] perf: specify SLANG_INC dir for perf
Liang Li
liang.li at windriver.com
Wed Aug 22 08:17:47 UTC 2012
On 2012-08-22 13:41, Bruce Ashfield <bruce.ashfield at windriver.com> wrote:
> On 12-08-21 11:01 PM, Liang Li wrote:
> > On 2012-08-21 21:07, Bruce Ashfield<bruce.ashfield at windriver.com> wrote:
> >> On 12-08-21 01:08 AM, Liang Li wrote:
> >>> On 2012-08-20 22:48, Bruce Ashfield<bruce.ashfield at windriver.com> wrote:
> >>>> On 12-08-17 09:05 AM, Liang Li wrote:
> >>>>> On 2012-08-17 21:01, Liang Li<liang.li at windriver.com> wrote:
> >>>>>> On 2012-08-17 18:53, Richard Purdie<richard.purdie at linuxfoundation.org> wrote:
> >>>>>>> On Fri, 2012-08-17 at 18:00 +0800, Liang Li wrote:
> >>>>>>>> I am totally confused, you mentioned 'general kernel do_install', I
> >>>>>>>> assume it's oe-core kernel.bbclass concept. Then you mentioned 'get
> >>>>>>>> the fix upstream in the mainline kernel', how could that happen?
> >>>>>>>>
> >>>>>>>> We are discussing about the solution to 'fix the compile warning to
> >>>>>>>> error' stuff that triggered by the '-I/usr/include/slang', right?
> >>>>>>>
> >>>>>>> Yes.
> >>>>>>>
> >>>>>>>> We do not necessarily have to change recipe to fix it since the issue
> >>>>>>>> is not introduced by the recipe, the hard coded '-I/usr/include/slang'
> >>>>>>>> in the Makefile cause the issue, we can fix the root cause by kernel
> >>>>>>>> patch(other than just comment the line out). I see your previous patch
> >>>>>>>> to kernel, by comment out the '-I/usr/include/slang' line in the
> >>>>>>>> Makefile, is the same behavior, but we won't have the change(comment
> >>>>>>>> out -I.. in Makefile) upstream to mainline, right?
> >>>>>>>
> >>>>>>> I am suggesting that firstly, someone send a patch to the mainline
> >>>>>>> kernel which changes -I/usr/include/slang to -I=/usr/include/slang in
> >>>>>>> that Makefile.
> >>>>>>>
> >>>>>>> Secondly, I'm suggesting that we add a line to kernel_do_install() in
> >>>>>>> kernel.bbclass which does a sed on the Makefile as installed into
> >>>>>>> $kerneldir which changes -I/usr/include/slang to -I=/usr/include/slang.
> >>>>>>>
> >>>>>>> We can then drop the patch I added to the linux-yocto kernels.
> >>>>>>>
> >>>>>>> This is all that should be needed, it should fix all the issues people
> >>>>>>> have reported in a way that is acceptable to everyone.
> >>>>>>>
> >>>>>>
> >>>>>> Ah, I see what you mean now. But we have push acceptable kernel patch
> >>>>>
> >>>>
> >>>> One final (I hope) follow up on this.
> >>>>
> >>>> Liang: were you going to put together (and test) the 'sed fix' for
> >>>> kernel.bbclass ?
> >>>>
> >>>
> >>> No problem, the patch for kernel.bbclass:
> >>>
> >>> commit 60a0b06
> >>> Author: Liang Li<liang.li at windriver.com>
> >>> Date: Tue Aug 21 11:06:01 2012 +0800
> >>>
> >>> kernel.bbclass: fix INC directory for SLANG
> >>>
> >>> The change is intend to fix the hardcoded '-I/usr/include/slang' in
> >>> the Makefile to be able to aware of SYSROOT if its specified.
> >>>
> >>> A planned kernel patch almost did the same change, but the change here
> >>> won't conflict with it so this change could work for all kernels.
> >>>
> >>> Signed-off-by: Liang Li<liang.li at windriver.com>
> >>>
> >>> diff --git a/meta/classes/kernel.bbclass b/meta/classes/kernel.bbclass
> >>> index 1afb9ab..282194d 100644
> >>> --- a/meta/classes/kernel.bbclass
> >>> +++ b/meta/classes/kernel.bbclass
> >>> @@ -190,6 +190,9 @@ kernel_do_install() {
> >>> for entry in $bin_files; do
> >>> rm -f $kerneldir/$entry
> >>> done
> >>> +
> >>> + # Fix SLNAG_INC for slang.h
> >>
> >> s/SLNAG_INC/SLANG_INC/
> >>
> >>> + sed -i 's#-I/usr/include/slang#-I=/usr/include/slang#g' $kerneldir/tools/perf/Makefile
> >>
> >> It looks like your baseline for this patch is the denzil branch. We'd
> >> want a version for master, which we could backport as required.
> >>
> >
> > Yes, the change for the master branch is almost the same, except line
> > number context:
> >
> > diff --git a/meta/classes/kernel.bbclass b/meta/classes/kernel.bbclass
> > index f34e632..fdef1be 100644
> > --- a/meta/classes/kernel.bbclass
> > +++ b/meta/classes/kernel.bbclass
> > @@ -204,6 +204,9 @@ kernel_do_install() {
> > for entry in $bin_files; do
> > rm -f $kerneldir/$entry
> > done
> > +
> > + # Fix SLANG_INC for slang.h
> > + sed -i 's#-I/usr/include/slang#-I=/usr/include/slang#g' $kerneldir/tools/perf/Makefile
> > }
> >
> > sysroot_stage_all_append() {
> >
> > ---
> >
> >>> }
> >>>
> >>> PACKAGE_PREPROCESS_FUNCS += "kernel_package_preprocess"
> >>>
> >>> ---
> >>>
> >>> The patch for kernel tree:
> >>>
> >>> commit 6b72896
> >>> Author: Liang Li<liang.li at windriver.com>
> >>> Date: Wed Aug 1 14:31:24 2012 +0800
> >>>
> >>> perf: add SLANG_INC for slang.h
> >>>
> >>> Previously we hard code '-I/usr/include/slang' to CFLAGS to works with
> >>> some hosts that has /usr/include/slang/slang.h other than
> >>> /usr/include/slang.h like Fedora. This will cause compiling warnings
> >>> in some cases.
> >>
> >> I'd rephrase this slightly as:
> >>
> >> CFLAGS was previously hard coded to contain "-I/usr/include/slang" to
> >> work with hosts that have "/usr/include/slang/slang.h" as well as hosts
> >> that have "/usr/include/slang.h". This path can cause compile warnings
> >> in some cases:
> >>
> >> <put the warnings here>
> >>
> >> .. and indicate that these warnings can actually cause build errors if
> >> WERROR is enabled.
> >>
> >>>
> >>> We could downgrade the priority of the default hard coded path, and
> >>> provide user a chance to specify correct location of slang.h then user
> >>> could specify SLANG_INC to avoid compile warnings like the
> >>> '/usr/include/slang' is not exists etc.
> >>
> >> Another minor rephrase:
> >>
> >> To fix this issue, we can use -idirafter to downgrade the priority of the
> >> default hard coded path. We can also make the slang include directory
> >> a variable, to allow the user to specify SLANG_INC and set their own
> >> include location.
> >>
> >
> > No problem, rephrased commit message:
> >
> > Subject: [PATCH] perf: add SLANG_INC for slang.h
> >
> > CFLAGS was previously hard coded to contain "-I/usr/include/slang" to
> > work with hosts that have "/usr/include/slang/slang.h" as well as hosts
> > that have "/usr/include/slang.h". This path can cause compile warnings
> > like:
> >
> > cc1: warning: '/usr/include/slang' doesn't exists.
> >
> > or
> >
> > cc1: warning: include location "/usr/include/slang" is unsafe for
> > cross-compilation [-Wpoison-system-directories]
> >
> > Then in some cases warnings become errors if WERROR is enabled hence
> > build errors.
> >
> > To fix this issue, we can use -idirafter to downgrade the priority of the
> > default hard coded path. We can also make the slang include directory
> > a variable, to allow the user to specify SLANG_INC and set their own
> > include location. And add a '=' prefix to indicate better
> > compatibility with sysroot/cross compile cases.
> >
> > Signed-off-by: Liang Li<liang.li at windriver.com>
> >
> >>>
> >>> Signed-off-by: Liang Li<liang.li at windriver.com>
> >>>
> >>> diff --git a/tools/perf/Makefile b/tools/perf/Makefile
> >>> index b7a7a87..e403c36 100644
> >>> --- a/tools/perf/Makefile
> >>> +++ b/tools/perf/Makefile
> >>> @@ -496,8 +496,10 @@ else
> >>> msg := $(warning newt not found, disables TUI support. Please install newt-devel or libnewt-dev);
> >>> BASIC_CFLAGS += -DNO_NEWT_SUPPORT
> >>> else
> >>> - # Fedora has /usr/include/slang/slang.h, but ubuntu /usr/include/slang.h
> >>> - BASIC_CFLAGS += -I/usr/include/slang
> >>> + # Some releases like Fedora has /usr/include/slang/slang.h other than /usr/include/slang.h
> >>> + SLANG_INC ?= -idirafter =/usr/include/slang
> >>
> >> One more question, have you confirmed that gcc is fine with this being
> >> in sysroot notation ? (I assume it is .. but I need to ask.
> >>
> >
> > Confirmed.
>
> Both look fine to me, if you send me a pull request for the kernel.bbclass
> part, and a separate one for the kernel patch, I'll take care
> of dropping the old slang patch, and getting all the changes
> to Richard in a single pull request.
>
Done. Thanks.
Regards,
Liang Li
More information about the Openembedded-core
mailing list