[OE-core] [PATCH 3/4] insane.bbclass: fix package_qa_check_buildpaths

Robert Yang liezhi.yang at windriver.com
Wed Sep 16 13:56:41 UTC 2015


On 09/16/2015 08:38 PM, Richard Purdie wrote:
> On Tue, 2015-09-15 at 19:28 -0700, Robert Yang wrote:
>> * Ignore elf files because they usually contain build path:
>>    - The path of the source file such as .c, these are usually happen
>>      when separate B and S since we use absolute path to run configure
>>      script, and then VPATH in Makefile will be an absolute path and
>>      contains build path, we can use relative path in autotools.bbclass
>>      to fix the problem, but we don't have to since they are harmless.
>>    - The configure options such as "configure --with-libtool-sysroot"
>>    - The compile options such as "gcc --sysroot"
>>    These are harmless usually, so ignore elf files.
>
> I'm not sure about this. Yes, elf files have large issues in the debug
> symbols but the main elf files really shouldn't have build paths in
> them. I understand they can creep in through a variety of sources
> including B!=S but we should really be identifying them and fixing them,
> not pretending they don't exist. The whole point of this check
> originally was to get to the bottom of this!

Thanks, the fix will change a lot to the build behaviour, I'd like to fix
them in 2.1.

// Robert

>
>> * Ignore "-dbg" and "-staticdev" package since symbols and .a files
>>    usually contain buildpath.
>
> Ideally, we should be fixing the debug code so its using target paths
> rather than build ones too...
>
> Cheers,
>
> Richard
>
>> * Ignore .a files too since they contain buildpath, this mainly for
>>    ignoring:
>>    glibc-2.21: glibc-dev/usr/lib/libc_nonshared.a
>>    libgcc-4.9.3: libgcc-dev/usr/lib/x86_64-poky-linux/4.9.3/libgcc_eh.a
>>    gcc-runtime-4.9.3: libssp-dev/usr/lib/libssp_nonshared.a
>>
>> * Use full path rather than package_qa_clean_path() when report issue,
>>    this makes it easier to find the file and fix the problem.
>>
>> Then we will verify other warings and fix one be one.
>
>
>
>



More information about the Openembedded-core mailing list