make -C tools clean failure on older systems

make -C tools clean failure on older systems

[Arnaldo Carvalho de Melo]

Hi,

	In:

 -------------
commit 67d8712dcc70aa16d8e14a52eb73870e3cbddfc2
Author: Shuah Khan <shuahkh@osg.samsung.com>
Date:   Wed Mar 18 11:57:39 2015 -0600

    selftests: Fix build failures when invoked from kselftest target

 -------------

You cleaned two variables using different methods, any reason for that?

I asked because the 'undefine' method causes it to fail in older
systems:

[acme@rhel5 linux]$ make -C tools/ clean
<SNIP>
  CLEAN    python
make[1]: Leaving directory `/home/acme/git/linux/tools/perf'
  DESCEND  testing/selftests
make[1]: Entering directory
`/home/acme/git/linux/tools/testing/selftests'
Makefile:30: *** missing separator.  Stop.
make[1]: Leaving directory
`/home/acme/git/linux/tools/testing/selftests'
make: *** [selftests_clean] Error 2
make: Leaving directory `/home/acme/git/linux/tools'

 ----------------------------------

[acme@rhel5 linux]$ make --version
GNU Make 3.81
Copyright (C) 2006  Free Software Foundation, Inc.
This is free software; see the source for copying conditions.
There is NO warranty; not even for MERCHANTABILITY or FITNESS FOR A
PARTICULAR PURPOSE.

This program built for x86_64-redhat-linux-gnu
[acme@rhel5 linux]$

Wonder if it would be ok to use:

diff --git a/tools/testing/selftests/Makefile b/tools/testing/selftests/Makefile
index 95abddcd7839..f76830643086 100644
--- a/tools/testing/selftests/Makefile
+++ b/tools/testing/selftests/Makefile
@@ -27,7 +27,7 @@ TARGETS_HOTPLUG += memory-hotplug
 # Makefile to avoid test build failures when test
 # Makefile doesn't have explicit build rules.
 ifeq (1,$(MAKELEVEL))
-undefine LDFLAGS
+override LDFLAGS =
 override MAKEFLAGS =
 endif
 
 ------------------------------------------


- Arnaldo

Re: make -C tools clean failure on older systems

[Shuah Khan]

On 05/14/2015 09:12 AM, Arnaldo Carvalho de Melo wrote:
> Hi,
> 
> 	In:
> 
>  -------------
> commit 67d8712dcc70aa16d8e14a52eb73870e3cbddfc2
> Author: Shuah Khan <shuahkh@osg.samsung.com>
> Date:   Wed Mar 18 11:57:39 2015 -0600
> 
>     selftests: Fix build failures when invoked from kselftest target
> 
>  -------------
> 
> You cleaned two variables using different methods, any reason for that?
> 
> I asked because the 'undefine' method causes it to fail in older
> systems:

The reason for this change is some tests fail to build when invoked
from the main Makefile level. The commit log explains the change:

Several tests that rely on implicit build rules fail to build,
when invoked from the main Makefile kselftest target. These
failures are due to --no-builtin-rules and --no-builtin-variables
options set in the inherited MAKEFLAGS.

--no-builtin-rules eliminates the use of built-in implicit rules
and --no-builtin-variables is for not defining built-in variables.
These two options override the use of implicit rules resulting in
build failures. In addition, inherited LDFLAGS result in build
failures and there is no need to define LDFLAGS.  Clear LDFLAGS
and MAKEFLAG when make is invoked from the main Makefile kselftest
target. Fixing this at selftests Makefile avoids changing the main
Makefile and keeps this change self contained at selftests level.

> 
> [acme@rhel5 linux]$ make -C tools/ clean
> <SNIP>
>   CLEAN    python
> make[1]: Leaving directory `/home/acme/git/linux/tools/perf'
>   DESCEND  testing/selftests
> make[1]: Entering directory
> `/home/acme/git/linux/tools/testing/selftests'
> Makefile:30: *** missing separator.  Stop.
> make[1]: Leaving directory
> `/home/acme/git/linux/tools/testing/selftests'
> make: *** [selftests_clean] Error 2
> make: Leaving directory `/home/acme/git/linux/tools'
> 
>  ----------------------------------
> 
> [acme@rhel5 linux]$ make --version
> GNU Make 3.81
> Copyright (C) 2006  Free Software Foundation, Inc.
> This is free software; see the source for copying conditions.
> There is NO warranty; not even for MERCHANTABILITY or FITNESS FOR A
> PARTICULAR PURPOSE.
> 
> This program built for x86_64-redhat-linux-gnu
> [acme@rhel5 linux]$
> 
> Wonder if it would be ok to use:
> 
> diff --git a/tools/testing/selftests/Makefile b/tools/testing/selftests/Makefile
> index 95abddcd7839..f76830643086 100644
> --- a/tools/testing/selftests/Makefile
> +++ b/tools/testing/selftests/Makefile
> @@ -27,7 +27,7 @@ TARGETS_HOTPLUG += memory-hotplug
>  # Makefile to avoid test build failures when test
>  # Makefile doesn't have explicit build rules.
>  ifeq (1,$(MAKELEVEL))
> -undefine LDFLAGS
> +override LDFLAGS =
>  override MAKEFLAGS =
>  endif

I recall testing with override and remember it to not work. If you would
like experiment with it, feel free to send a patch with that change. My
make version is very new:

make --version
GNU Make 4.0
Built for x86_64-pc-linux-gnu

You have to run make kselftest from the main kernel Makefile to see
the build failures that undefine LDFLAGS fixed.

thanks,
-- Shuah



-- 
Shuah Khan
Sr. Linux Kernel Developer
Open Source Innovation Group
Samsung Research America (Silicon Valley)
shuahkh@osg.samsung.com | (970) 217-8978

Re: make -C tools clean failure on older systems

[Arnaldo Carvalho de Melo]

Em Thu, May 14, 2015 at 09:39:43AM -0600, Shuah Khan escreveu:
> On 05/14/2015 09:12 AM, Arnaldo Carvalho de Melo wrote:
> > Hi,
> > 
> > 	In:
> > 
> >  -------------
> > commit 67d8712dcc70aa16d8e14a52eb73870e3cbddfc2
> > Author: Shuah Khan <shuahkh@osg.samsung.com>
> > Date:   Wed Mar 18 11:57:39 2015 -0600
> > 
> >     selftests: Fix build failures when invoked from kselftest target
> > 
> >  -------------
> > 
> > You cleaned two variables using different methods, any reason for that?
> > 
> > I asked because the 'undefine' method causes it to fail in older
> > systems:
> 
> The reason for this change is some tests fail to build when invoked
> from the main Makefile level. The commit log explains the change:

I saw it

> > +++ b/tools/testing/selftests/Makefile
> > @@ -27,7 +27,7 @@ TARGETS_HOTPLUG += memory-hotplug
> >  # Makefile to avoid test build failures when test
> >  # Makefile doesn't have explicit build rules.
> >  ifeq (1,$(MAKELEVEL))
> > -undefine LDFLAGS
> > +override LDFLAGS =
> >  override MAKEFLAGS =
> >  endif
> 
> I recall testing with override and remember it to not work. If you would

bummer, then its not that easy, I thought it was an oversight, but if
you tested it and it didn't work, nevermind, no need, I guess, to
support older systems if it requires more, non trivial, work.

> like experiment with it, feel free to send a patch with that change. My
> make version is very new:
> 
> make --version
> GNU Make 4.0
> Built for x86_64-pc-linux-gnu
> 
> You have to run make kselftest from the main kernel Makefile to see
> the build failures that undefine LDFLAGS fixed.
> 
> thanks,
> -- Shuah
> 
> 
> 
> -- 
> Shuah Khan
> Sr. Linux Kernel Developer
> Open Source Innovation Group
> Samsung Research America (Silicon Valley)
> shuahkh@osg.samsung.com | (970) 217-8978

Re: make -C tools clean failure on older systems

[Ingo Molnar]

* Arnaldo Carvalho de Melo <acme@kernel.org> wrote:

> > > @@ -27,7 +27,7 @@ TARGETS_HOTPLUG += memory-hotplug
> > >  # Makefile to avoid test build failures when test
> > >  # Makefile doesn't have explicit build rules.
> > >  ifeq (1,$(MAKELEVEL))
> > > -undefine LDFLAGS
> > > +override LDFLAGS =
> > >  override MAKEFLAGS =
> > >  endif
> > 
> > I recall testing with override and remember it to not work. If you would
> 
> bummer, then its not that easy, I thought it was an oversight, but 
> if you tested it and it didn't work, nevermind, no need, I guess, to 
> support older systems if it requires more, non trivial, work.

So what's the 'make' version cutoff for this capability?

We should not require insane hacks, the minimum required make version 
for the kernel, as per Documentation/Changes, is v3.80:

 o  GNU make               3.80                    # make --version

Thanks,

	Ingo

Re: make -C tools clean failure on older systems

[Shuah Khan]

On 05/14/2015 09:54 AM, Arnaldo Carvalho de Melo wrote:
> Em Thu, May 14, 2015 at 09:39:43AM -0600, Shuah Khan escreveu:
>> On 05/14/2015 09:12 AM, Arnaldo Carvalho de Melo wrote:
>>> Hi,
>>>
>>> 	In:
>>>
>>>  -------------
>>> commit 67d8712dcc70aa16d8e14a52eb73870e3cbddfc2
>>> Author: Shuah Khan <shuahkh@osg.samsung.com>
>>> Date:   Wed Mar 18 11:57:39 2015 -0600
>>>
>>>     selftests: Fix build failures when invoked from kselftest target
>>>
>>>  -------------
>>>
>>> You cleaned two variables using different methods, any reason for that?
>>>
>>> I asked because the 'undefine' method causes it to fail in older
>>> systems:
>>
>> The reason for this change is some tests fail to build when invoked
>> from the main Makefile level. The commit log explains the change:
> 
> I saw it
> 
>>> +++ b/tools/testing/selftests/Makefile
>>> @@ -27,7 +27,7 @@ TARGETS_HOTPLUG += memory-hotplug
>>>  # Makefile to avoid test build failures when test
>>>  # Makefile doesn't have explicit build rules.
>>>  ifeq (1,$(MAKELEVEL))
>>> -undefine LDFLAGS
>>> +override LDFLAGS =
>>>  override MAKEFLAGS =
>>>  endif
>>
>> I recall testing with override and remember it to not work. If you would
> 
> bummer, then its not that easy, I thought it was an oversight, but if
> you tested it and it didn't work, nevermind, no need, I guess, to
> support older systems if it requires more, non trivial, work.
> 
>> like experiment with it, feel free to send a patch with that change. My
>> make version is very new:
>>
>> make --version
>> GNU Make 4.0
>> Built for x86_64-pc-linux-gnu
>>
>> You have to run make kselftest from the main kernel Makefile to see
>> the build failures that undefine LDFLAGS fixed.
>>

Hmm. I ran a quick test with override and it worked. My recollection
must not be good. :) Here is what I tried:

diff --git a/tools/testing/selftests/Makefile
b/tools/testing/selftests/Makefile
index ebac6b8..56b11cd 100644
--- a/tools/testing/selftests/Makefile
+++ b/tools/testing/selftests/Makefile
@@ -28,7 +28,7 @@ TARGETS_HOTPLUG += memory-hotplug
 # Makefile to avoid test build failures when test
 # Makefile doesn't have explicit build rules.
 ifeq (1,$(MAKELEVEL))
-undefine LDFLAGS
+override LDFLAGS =
 override MAKEFLAGS =
 endif


Could you please try this and let me know if it works and if
you have time, please send me a patch as well.

thanks,
-- Shuah

-- 
Shuah Khan
Sr. Linux Kernel Developer
Open Source Innovation Group
Samsung Research America (Silicon Valley)
shuahkh@osg.samsung.com | (970) 217-8978

Re: make -C tools clean failure on older systems

[Arnaldo Carvalho de Melo]

Em Thu, May 14, 2015 at 12:29:37PM -0600, Shuah Khan escreveu:
> On 05/14/2015 09:54 AM, Arnaldo Carvalho de Melo wrote:
> > Em Thu, May 14, 2015 at 09:39:43AM -0600, Shuah Khan escreveu:
> >> On 05/14/2015 09:12 AM, Arnaldo Carvalho de Melo wrote:
> >>> Hi,
> >>>
> >>> 	In:
> >>>
> >>>  -------------
> >>> commit 67d8712dcc70aa16d8e14a52eb73870e3cbddfc2
> >>> Author: Shuah Khan <shuahkh@osg.samsung.com>
> >>> Date:   Wed Mar 18 11:57:39 2015 -0600
> >>>
> >>>     selftests: Fix build failures when invoked from kselftest target
> >>>
> >>>  -------------
> >>>
> >>> You cleaned two variables using different methods, any reason for that?
> >>>
> >>> I asked because the 'undefine' method causes it to fail in older
> >>> systems:
> >>
> >> The reason for this change is some tests fail to build when invoked
> >> from the main Makefile level. The commit log explains the change:
> > 
> > I saw it
> > 
> >>> +++ b/tools/testing/selftests/Makefile
> >>> @@ -27,7 +27,7 @@ TARGETS_HOTPLUG += memory-hotplug
> >>>  # Makefile to avoid test build failures when test
> >>>  # Makefile doesn't have explicit build rules.
> >>>  ifeq (1,$(MAKELEVEL))
> >>> -undefine LDFLAGS
> >>> +override LDFLAGS =
> >>>  override MAKEFLAGS =
> >>>  endif
> >>
> >> I recall testing with override and remember it to not work. If you would
> > 
> > bummer, then its not that easy, I thought it was an oversight, but if
> > you tested it and it didn't work, nevermind, no need, I guess, to
> > support older systems if it requires more, non trivial, work.
> > 
> >> like experiment with it, feel free to send a patch with that change. My
> >> make version is very new:
> >>
> >> make --version
> >> GNU Make 4.0
> >> Built for x86_64-pc-linux-gnu
> >>
> >> You have to run make kselftest from the main kernel Makefile to see
> >> the build failures that undefine LDFLAGS fixed.
> >>
> 
> Hmm. I ran a quick test with override and it worked. My recollection
> must not be good. :) Here is what I tried:
> 
> diff --git a/tools/testing/selftests/Makefile
> b/tools/testing/selftests/Makefile
> index ebac6b8..56b11cd 100644
> --- a/tools/testing/selftests/Makefile
> +++ b/tools/testing/selftests/Makefile
> @@ -28,7 +28,7 @@ TARGETS_HOTPLUG += memory-hotplug
>  # Makefile to avoid test build failures when test
>  # Makefile doesn't have explicit build rules.
>  ifeq (1,$(MAKELEVEL))
> -undefine LDFLAGS
> +override LDFLAGS =
>  override MAKEFLAGS =
>  endif

Yeah, that is perfect! I.e. it is the patch I had proposed :o)

Ok, will send a patch to you, works well for me for:

Red Hat Enterprise Linux 5 (Fedora Core 6 based, ides of 2.6.18 kernels)

[acme@rhel5 tools]$ make -v
GNU Make 3.81

And:

Fedora 21:

[acme@zoo tools]$ make --version
GNU Make 4.0

- Arnaldo

Re: make -C tools clean failure on older systems

[Arnaldo Carvalho de Melo]

Em Thu, May 14, 2015 at 07:42:22PM +0200, Ingo Molnar escreveu:
> * Arnaldo Carvalho de Melo <acme@kernel.org> wrote:
> > > >  # Makefile doesn't have explicit build rules.
> > > >  ifeq (1,$(MAKELEVEL))
> > > > -undefine LDFLAGS
> > > > +override LDFLAGS =
> > > >  override MAKEFLAGS =

> > > I recall testing with override and remember it to not work. If you would

> > bummer, then its not that easy, I thought it was an oversight, but 
> > if you tested it and it didn't work, nevermind, no need, I guess, to 
> > support older systems if it requires more, non trivial, work.
 
> So what's the 'make' version cutoff for this capability?
 
> We should not require insane hacks, the minimum required make version 
> for the kernel, as per Documentation/Changes, is v3.80:
 
>  o  GNU make               3.80                    # make --version

3.81 is what is used where I found the problem.

With his confirmation that one of the two variants that he used on the
same patch to do what seemingly has the same effect for the problem at
hand, the one that is supported in make 3.81 (RHEL5), all is well :-)

- Arnaldo