From mboxrd@z Thu Jan  1 00:00:00 1970
From: Romain Naour <romain.naour@openwide.fr>
Date: Sat, 25 Jul 2015 11:24:40 +0200
Subject: [Buildroot] [PATCH] ARC: update tools to arc-2015.06 release
In-Reply-To: <20150718031758.GA21884@fox>
References: <1436964678-24352-1-git-send-email-abrodkin@synopsys.com>
 <55A8EAF7.5060506@openwide.fr> <20150717222354.GE3637@free.fr>
 <20150717225415.GF3637@free.fr> <20150718031758.GA21884@fox>
Message-ID: <55B355D8.4070900@openwide.fr>
List-Id: <buildroot.busybox.net>
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-Transfer-Encoding: 7bit
To: buildroot@busybox.net

Hi Guido,

Le 18/07/2015 05:17, Guido Mart?nez a ?crit :
> Hi Yann, all,
> 
> On Sat, Jul 18, 2015 at 12:54:15AM +0200, Yann E. MORIN wrote:
>> Romain, All,
>>
>> On 2015-07-18 00:23 +0200, Yann E. MORIN spake thusly:
>>> On 2015-07-17 13:45 +0200, Romain Naour spake thusly:
>>> [--SNIP--]
>>>> Your patch looks good, but I have a build failure with uclibc:
>>>>
>>>> make O=test/arc/ toolchain 2>&1 | tee build-arc-toolchain.log
>> [--SNIP--]
>>> OK, I was able to reproduce the error. [...]
>>
>> Here's a patch that does fix the issue.
>>
>> Tested on-top of master, with:
>>   - in-tree build
>>   - out-of-tree build without makefile wrapper (Romain's use-case)
>>   - out-of-tree build with makefile wrapper (my use-case)
>>
>> They all worked at least so far as to build uclibc (make uclibc);
>> nothing more was attempted).
>>
>> The change is not unlike what we already had to test where O was coming
>> from, but we just reverse the logic to keep only what we're interested
>> in, and we forcibly unexport O.
>>
>> I'm not really happy with the patch to start with (hence I'm not
>> officially pushing it):
>>
>>     diff --git a/Makefile b/Makefile
>>     index 531ac5d..5abca65 100644
>>     --- a/Makefile
>>     +++ b/Makefile
>>     @@ -29,11 +29,16 @@ UMASK=0022
>>      ifneq ($(shell umask),$(UMASK))
>>      .PHONY: all $(MAKECMDGOALS)
>>      
>>     +ifeq ($(origin O),command line)
>>     +EXTRAMAKEARGS = O=$(O)
>>     +endif
>>     +unexport O
>>     +
>>      all:
>>     -	@umask $(UMASK) && $(MAKE  ) --no-print-directory
>>     +	@umask $(UMASK) && $(MAKE) $(EXTRAMAKEARGS) -- no-print-directory
>>      
>>      $(MAKECMDGOALS):
>>     -	@umask $(UMASK) && $(MAKE) --n o-print-directory $@
>>     +	@umask $(UMASK) && $(MAKE) $(EXTRAMAKEARGS) --no   -print-directory $@
>>      
>>      else # umask
> 
> This works OK. But I think something like:
> 
> 	diff --git a/Makefile b/Makefile
> 	index afdf554..df52c85 100644
> 	--- a/Makefile
> 	+++ b/Makefile
> 	@@ -329,6 +329,7 @@ unexport CONFIG_SITE
> 	 unexport QMAKESPEC
> 	 unexport TERMINFO
> 	 unexport MACHINE
> 	+unexport O
> 
> 	 GNU_HOST_NAME := $(shell support/gnuconfig/config.guess)
> 
> does too. Maybe I'm missing something, but I don't think we would ever
> want to export $(O) to any of the package Makefiles, the same way we
> don't export $(CC).
> 
> Note that $(origin O) will still be "command line" in the sub-make after
> setting the umask, since "O=..." will be set in $(MAKEFLAGS), thus there
> shouldn't be a difference in the run.
> 
> I tested those 3 cases you mentioned and they work OK, but I'll run full
> builds in a minute.
> 

Can you send a proper patch to the list please?
I'm using this change and it fixes the uClibc-ng build without regression so far.

Best regards,
Romain Naour