RFC: SDK/external-toolchain fixes

RFC: SDK/external-toolchain fixes

[Denys Dmytriyenko]

Hi,

I've been playing with OE's SDK and external-toolchain features lately and
stumbled upon couple of issues. Thanks to Philip Balister for commiting
trivial fixes in, now I have 2 more, which require wider review. Please let
me know of any problems or comments. Thanks.


#1 gcc-cross-sdk 4.3.1 missing include-fixed
http://bugs.openembedded.net/show_bug.cgi?id=4536

gcc 4.3.x has some GCC-provided headers in include-fixed (namely limits.h),
which are missing in the gcc-cross-sdk version, rendering external toolchain
unusable.

Provided patch copies them in the include directory as a temporary solution.
A similar temporary solution is present in gcc-cross_4.3.1, see
do_stage_append()


#2 external-toolchain misses providing runtime libgcc
http://bugs.openembedded.net/show_bug.cgi?id=4542

Since external-toolchain misses specifying libgcc as provided runtime library,
dependency tree picks up either gcc-cross or even gcc-noemu...

Attached patch fixes the problem.


-- 
Denys

Re: RFC: SDK/external-toolchain fixes

[Khem Raj]

On (11/09/08 15:52), Denys Dmytriyenko wrote:
> Hi,
> 
> I've been playing with OE's SDK and external-toolchain features lately and
> stumbled upon couple of issues. Thanks to Philip Balister for commiting
> trivial fixes in, now I have 2 more, which require wider review. Please let
> me know of any problems or comments. Thanks.
> 
> 
> #1 gcc-cross-sdk 4.3.1 missing include-fixed
> http://bugs.openembedded.net/show_bug.cgi?id=4536
> 
> gcc 4.3.x has some GCC-provided headers in include-fixed (namely limits.h),
> which are missing in the gcc-cross-sdk version, rendering external toolchain
> unusable.

I do not like this fix. IMO Instead of copying them into include directory they should be let alone in include-fixed directory and packaged. You should add knowledge of this new directory to the failing package instead. What packages are broken because of this. 

> 
> Provided patch copies them in the include directory as a temporary solution.
> A similar temporary solution is present in gcc-cross_4.3.1, see
> do_stage_append()
> 
> 
> #2 external-toolchain misses providing runtime libgcc
> http://bugs.openembedded.net/show_bug.cgi?id=4542
> 
> Since external-toolchain misses specifying libgcc as provided runtime library,
> dependency tree picks up either gcc-cross or even gcc-noemu...

This one seems ok to me. 

> 
> Attached patch fixes the problem.
> 
> 
> -- 
> Denys
> 
> _______________________________________________
> Openembedded-devel mailing list
> Openembedded-devel@lists.openembedded.org
> http://lists.linuxtogo.org/cgi-bin/mailman/listinfo/openembedded-devel

Re: RFC: SDK/external-toolchain fixes

[Denys Dmytriyenko]

First of all, thanks for your review.

On Thu, Sep 11, 2008 at 05:02:08PM -0700, Khem Raj wrote:
> On (11/09/08 15:52), Denys Dmytriyenko wrote:
> > 
> > #1 gcc-cross-sdk 4.3.1 missing include-fixed
> > http://bugs.openembedded.net/show_bug.cgi?id=4536
> > 
> > gcc 4.3.x has some GCC-provided headers in include-fixed (namely limits.h),
> > which are missing in the gcc-cross-sdk version, rendering external toolchain
> > unusable.
> 
> I do not like this fix. IMO Instead of copying them into include directory 
> they should be let alone in include-fixed directory and packaged. You should 
> add knowledge of this new directory to the failing package instead. What 
> packages are broken because of this.

I agree that this is not a proper fix, but rather a hack. glibc has its own 
limits.h, which wants to include gcc's limits.h like this:

 /* Get the compiler's limits.h, which defines almost all the ISO constants.
    We put this #include_next outside the double inclusion check because
    it should be possible to include this file more than once and still get
    the definitions from gcc's header.  */
#if defined __GNUC__ && !defined _GCC_LIMITS_H_
/* `_GCC_LIMITS_H_' is what GCC's file defines.  */
# include_next <limits.h>
#endif

This issue was brought up before, but as far as I can tell, it wasn't fixed 
properly.
This is initial Koen's post about the issue:
http://projects.linuxtogo.org/pipermail/openembedded-devel/2008-April/005108.html
And here is your own reply with one of the possible solutions:
http://projects.linuxtogo.org/pipermail/openembedded-devel/2008-April/005111.html

Here is how it's "solved" (i.e. worked around) in gcc-cross_4.3.1.bb (4.3.2):

# Hack till we fix *libc properly
do_stage_append() {
        ln -sf ${CROSS_DIR}/lib/gcc/${TARGET_SYS}/${BINV}/include-fixed/* ${CROSS_DIR}/lib/gcc/${TARGET_SYS}/${BINV}/include/
}

I'd love to see it fixed properly, but for now I'm trying to bring 
gcc-cross-sdk on par with gcc-cross.

> > Provided patch copies them in the include directory as a temporary solution.
> > A similar temporary solution is present in gcc-cross_4.3.1, see
> > do_stage_append()
> > 
> > 
> > #2 external-toolchain misses providing runtime libgcc
> > http://bugs.openembedded.net/show_bug.cgi?id=4542
> > 
> > Since external-toolchain misses specifying libgcc as provided runtime library,
> > dependency tree picks up either gcc-cross or even gcc-noemu...
> 
> This one seems ok to me. 

Thanks! Whom should I ping to get it commited?

-- 
Denys

Re: RFC: SDK/external-toolchain fixes

[Khem Raj]

On (11/09/08 21:53), Denys Dmytriyenko wrote:
> First of all, thanks for your review.
> 
> On Thu, Sep 11, 2008 at 05:02:08PM -0700, Khem Raj wrote:
> > On (11/09/08 15:52), Denys Dmytriyenko wrote:
> > > 
> > > #1 gcc-cross-sdk 4.3.1 missing include-fixed
> > > http://bugs.openembedded.net/show_bug.cgi?id=4536
> > > 
> > > gcc 4.3.x has some GCC-provided headers in include-fixed (namely limits.h),
> > > which are missing in the gcc-cross-sdk version, rendering external toolchain
> > > unusable.
> > 
> > I do not like this fix. IMO Instead of copying them into include directory 
> > they should be let alone in include-fixed directory and packaged. You should 
> > add knowledge of this new directory to the failing package instead. What 
> > packages are broken because of this.
> 
> I agree that this is not a proper fix, but rather a hack. glibc has its own 
> limits.h, which wants to include gcc's limits.h like this:
> 
>  /* Get the compiler's limits.h, which defines almost all the ISO constants.
>     We put this #include_next outside the double inclusion check because
>     it should be possible to include this file more than once and still get
>     the definitions from gcc's header.  */
> #if defined __GNUC__ && !defined _GCC_LIMITS_H_
> /* `_GCC_LIMITS_H_' is what GCC's file defines.  */
> # include_next <limits.h>
> #endif
> 
> This issue was brought up before, but as far as I can tell, it wasn't fixed 
> properly.
> This is initial Koen's post about the issue:
> http://projects.linuxtogo.org/pipermail/openembedded-devel/2008-April/005108.html
> And here is your own reply with one of the possible solutions:
> http://projects.linuxtogo.org/pipermail/openembedded-devel/2008-April/005111.html
> 

right. So can you try this patch and see if this fixes the problem ?

> Here is how it's "solved" (i.e. worked around) in gcc-cross_4.3.1.bb (4.3.2):
> 
> # Hack till we fix *libc properly
> do_stage_append() {
>         ln -sf ${CROSS_DIR}/lib/gcc/${TARGET_SYS}/${BINV}/include-fixed/* ${CROSS_DIR}/lib/gcc/${TARGET_SYS}/${BINV}/include/
> }
> 
> I'd love to see it fixed properly, but for now I'm trying to bring 
> gcc-cross-sdk on par with gcc-cross.
> 
> > > Provided patch copies them in the include directory as a temporary solution.
> > > A similar temporary solution is present in gcc-cross_4.3.1, see
> > > do_stage_append()
> > > 
> > > 
> > > #2 external-toolchain misses providing runtime libgcc
> > > http://bugs.openembedded.net/show_bug.cgi?id=4542
> > > 
> > > Since external-toolchain misses specifying libgcc as provided runtime library,
> > > dependency tree picks up either gcc-cross or even gcc-noemu...
> > 
> > This one seems ok to me. 
> 
> Thanks! Whom should I ping to get it commited?

I can do it.

> 
> -- 
> Denys
> 
> _______________________________________________
> Openembedded-devel mailing list
> Openembedded-devel@lists.openembedded.org
> http://lists.linuxtogo.org/cgi-bin/mailman/listinfo/openembedded-devel

Re: RFC: SDK/external-toolchain fixes

[Denys Dmytriyenko]

On Thu, Sep 11, 2008 at 07:11:36PM -0700, Khem Raj wrote:
> > > > #1 gcc-cross-sdk 4.3.1 missing include-fixed
> > > > http://bugs.openembedded.net/show_bug.cgi?id=4536
> > > > 
> > > > gcc 4.3.x has some GCC-provided headers in include-fixed (namely limits.h),
> > > > which are missing in the gcc-cross-sdk version, rendering external toolchain
> > > > unusable.
> > > 
> > > I do not like this fix. IMO Instead of copying them into include directory 
> > > they should be let alone in include-fixed directory and packaged. You should 
> > > add knowledge of this new directory to the failing package instead. What 
> > > packages are broken because of this.
> > 
> > I agree that this is not a proper fix, but rather a hack. glibc has its own 
> > limits.h, which wants to include gcc's limits.h like this:
> > 
> > This issue was brought up before, but as far as I can tell, it wasn't fixed 
> > properly.
> > This is initial Koen's post about the issue:
> > http://projects.linuxtogo.org/pipermail/openembedded-devel/2008-April/005108.html
> > And here is your own reply with one of the possible solutions:
> > http://projects.linuxtogo.org/pipermail/openembedded-devel/2008-April/005111.html
> 
> right. So can you try this patch and see if this fixes the problem ?

Let me try it. The thing is it's for glibc only, I guess.

> > > > #2 external-toolchain misses providing runtime libgcc
> > > > http://bugs.openembedded.net/show_bug.cgi?id=4542
> > > > 
> > > > Since external-toolchain misses specifying libgcc as provided runtime library,
> > > > dependency tree picks up either gcc-cross or even gcc-noemu...
> > > 
> > > This one seems ok to me. 
> > 
> > Thanks! Whom should I ping to get it commited?
> 
> I can do it.

That would be great. Thanks.

-- 
Denys

Re: RFC: SDK/external-toolchain fixes

[Koen Kooi]

Khem Raj wrote:
> On (11/09/08 21:53), Denys Dmytriyenko wrote:
>> First of all, thanks for your review.
>>
>> On Thu, Sep 11, 2008 at 05:02:08PM -0700, Khem Raj wrote:
>>> On (11/09/08 15:52), Denys Dmytriyenko wrote:
>>>> #1 gcc-cross-sdk 4.3.1 missing include-fixed
>>>> http://bugs.openembedded.net/show_bug.cgi?id=4536
>>>>
>>>> gcc 4.3.x has some GCC-provided headers in include-fixed (namely limits.h),
>>>> which are missing in the gcc-cross-sdk version, rendering external toolchain
>>>> unusable.
>>> I do not like this fix. IMO Instead of copying them into include directory
>>> they should be let alone in include-fixed directory and packaged. You should
>>> add knowledge of this new directory to the failing package instead. What
>>> packages are broken because of this.
>> I agree that this is not a proper fix, but rather a hack. glibc has its own
>> limits.h, which wants to include gcc's limits.h like this:
>>
>>   /* Get the compiler's limits.h, which defines almost all the ISO constants.
>>      We put this #include_next outside the double inclusion check because
>>      it should be possible to include this file more than once and still get
>>      the definitions from gcc's header.  */
>> #if defined __GNUC__&&  !defined _GCC_LIMITS_H_
>> /* `_GCC_LIMITS_H_' is what GCC's file defines.  */
>> # include_next<limits.h>
>> #endif
>>
>> This issue was brought up before, but as far as I can tell, it wasn't fixed
>> properly.
>> This is initial Koen's post about the issue:
>> http://projects.linuxtogo.org/pipermail/openembedded-devel/2008-April/005108.html
>> And here is your own reply with one of the possible solutions:
>> http://projects.linuxtogo.org/pipermail/openembedded-devel/2008-April/005111.html
>>
>
> right. So can you try this patch and see if this fixes the problem ?
>
>> Here is how it's "solved" (i.e. worked around) in gcc-cross_4.3.1.bb (4.3.2):
>>
>> # Hack till we fix *libc properly
>> do_stage_append() {
>>          ln -sf ${CROSS_DIR}/lib/gcc/${TARGET_SYS}/${BINV}/include-fixed/* ${CROSS_DIR}/lib/gcc/${TARGET_SYS}/${BINV}/include/
>> }
>>
>> I'd love to see it fixed properly, but for now I'm trying to bring
>> gcc-cross-sdk on par with gcc-cross.
>>
>>>> Provided patch copies them in the include directory as a temporary solution.
>>>> A similar temporary solution is present in gcc-cross_4.3.1, see
>>>> do_stage_append()
>>>>
>>>>
>>>> #2 external-toolchain misses providing runtime libgcc
>>>> http://bugs.openembedded.net/show_bug.cgi?id=4542
>>>>
>>>> Since external-toolchain misses specifying libgcc as provided runtime library,
>>>> dependency tree picks up either gcc-cross or even gcc-noemu...
>>> This one seems ok to me.
>> Thanks! Whom should I ping to get it commited?
>
> I can do it.

gcc_4.3.x.bb also needs a fix since I get all kinds of limits.h related 
errors when trying to do native compilation on the beagle.

regards,

Koen

Re: RFC: SDK/external-toolchain fixes

[Khem Raj]

On (13/09/08 15:49), Koen Kooi wrote:
> Khem Raj wrote:
>> On (11/09/08 21:53), Denys Dmytriyenko wrote:
>>> First of all, thanks for your review.
>>>
>>> On Thu, Sep 11, 2008 at 05:02:08PM -0700, Khem Raj wrote:
>>>> On (11/09/08 15:52), Denys Dmytriyenko wrote:
>>>>> #1 gcc-cross-sdk 4.3.1 missing include-fixed
>>>>> http://bugs.openembedded.net/show_bug.cgi?id=4536
>>>>>
>>>>> gcc 4.3.x has some GCC-provided headers in include-fixed (namely limits.h),
>>>>> which are missing in the gcc-cross-sdk version, rendering external toolchain
>>>>> unusable.
>>>> I do not like this fix. IMO Instead of copying them into include directory
>>>> they should be let alone in include-fixed directory and packaged. You should
>>>> add knowledge of this new directory to the failing package instead. What
>>>> packages are broken because of this.
>>> I agree that this is not a proper fix, but rather a hack. glibc has its own
>>> limits.h, which wants to include gcc's limits.h like this:
>>>
>>>   /* Get the compiler's limits.h, which defines almost all the ISO constants.
>>>      We put this #include_next outside the double inclusion check because
>>>      it should be possible to include this file more than once and still get
>>>      the definitions from gcc's header.  */
>>> #if defined __GNUC__&&  !defined _GCC_LIMITS_H_
>>> /* `_GCC_LIMITS_H_' is what GCC's file defines.  */
>>> # include_next<limits.h>
>>> #endif
>>>
>>> This issue was brought up before, but as far as I can tell, it wasn't fixed
>>> properly.
>>> This is initial Koen's post about the issue:
>>> http://projects.linuxtogo.org/pipermail/openembedded-devel/2008-April/005108.html
>>> And here is your own reply with one of the possible solutions:
>>> http://projects.linuxtogo.org/pipermail/openembedded-devel/2008-April/005111.html
>>>
>>
>> right. So can you try this patch and see if this fixes the problem ?
>>
>>> Here is how it's "solved" (i.e. worked around) in gcc-cross_4.3.1.bb (4.3.2):
>>>
>>> # Hack till we fix *libc properly
>>> do_stage_append() {
>>>          ln -sf ${CROSS_DIR}/lib/gcc/${TARGET_SYS}/${BINV}/include-fixed/* ${CROSS_DIR}/lib/gcc/${TARGET_SYS}/${BINV}/include/
>>> }
>>>
>>> I'd love to see it fixed properly, but for now I'm trying to bring
>>> gcc-cross-sdk on par with gcc-cross.
>>>
>>>>> Provided patch copies them in the include directory as a temporary solution.
>>>>> A similar temporary solution is present in gcc-cross_4.3.1, see
>>>>> do_stage_append()
>>>>>
>>>>>
>>>>> #2 external-toolchain misses providing runtime libgcc
>>>>> http://bugs.openembedded.net/show_bug.cgi?id=4542
>>>>>
>>>>> Since external-toolchain misses specifying libgcc as provided runtime library,
>>>>> dependency tree picks up either gcc-cross or even gcc-noemu...
>>>> This one seems ok to me.
>>> Thanks! Whom should I ping to get it commited?
>>
>> I can do it.
>
> gcc_4.3.x.bb also needs a fix since I get all kinds of limits.h related  
>, errors when trying to do native compilation on the beagle.

I only did cross build/testing no target gcc stuff so I did not see this problem thus far.
I thought gcc driver should have added include-fixed to default search paths for includes unless you disable it to do so like glibc does in such a case the application should add it to include search paths manually. 

If you can give more details about this failure on native build. I could try to dig more.

Thanks
-Khem

>
> regards,
>
> Koen
>
>
> _______________________________________________
> Openembedded-devel mailing list
> Openembedded-devel@lists.openembedded.org
> http://lists.linuxtogo.org/cgi-bin/mailman/listinfo/openembedded-devel

Re: RFC: SDK/external-toolchain fixes

[Koen Kooi]

Khem Raj wrote:
> On (13/09/08 15:49), Koen Kooi wrote:

>> gcc_4.3.x.bb also needs a fix since I get all kinds of limits.h related
>> , errors when trying to do native compilation on the beagle.
>
> I only did cross build/testing no target gcc stuff so I did not see this problem thus far.
> I thought gcc driver should have added include-fixed to default search paths for includes unless you disable it to do so like glibc does in such a case the application should add it to include search paths manually.
>
> If you can give more details about this failure on native build. I could try to dig more.

I'll post some logs when IBC is over.

regards,

Koen

Re: RFC: SDK/external-toolchain fixes

[Denys Dmytriyenko]

On Sat, Sep 13, 2008 at 12:06:43PM -0700, Khem Raj wrote:
> >>>>> gcc 4.3.x has some GCC-provided headers in include-fixed (namely limits.h),
> >>>>> which are missing in the gcc-cross-sdk version, rendering external toolchain
> >>>>> unusable.
> >>>> I do not like this fix. IMO Instead of copying them into include directory
> >>>> they should be let alone in include-fixed directory and packaged. You should
> >>>> add knowledge of this new directory to the failing package instead. What
> >>>> packages are broken because of this.
> >>> I agree that this is not a proper fix, but rather a hack. glibc has its own
> >>> limits.h, which wants to include gcc's limits.h like this:
> >>>
> >>> This issue was brought up before, but as far as I can tell, it wasn't fixed
> >>> properly.
> >>> This is initial Koen's post about the issue:
> >>> http://projects.linuxtogo.org/pipermail/openembedded-devel/2008-April/005108.html
> >>> And here is your own reply with one of the possible solutions:
> >>> http://projects.linuxtogo.org/pipermail/openembedded-devel/2008-April/005111.html
> >>>
> >>
> >> right. So can you try this patch and see if this fixes the problem ?
> 
> I only did cross build/testing no target gcc stuff so I did not see this problem thus far.
> I thought gcc driver should have added include-fixed to default search paths for includes unless you disable it to do so like glibc does in such a case the application should add it to include search paths manually. 
> 
> If you can give more details about this failure on native build. I could try to dig more.

The proper fix seems to work fine. Thanks for applying the patch.

-- 
Denys

Re: RFC: SDK/external-toolchain fixes

[Denys Dmytriyenko]

Khem,

On Mon, Sep 15, 2008 at 03:18:49AM -0400, Denys Dmytriyenko wrote:
> On Sat, Sep 13, 2008 at 12:06:43PM -0700, Khem Raj wrote:
> > >>> I agree that this is not a proper fix, but rather a hack. glibc has its own
> > >>> limits.h, which wants to include gcc's limits.h like this:
> > >>>
> > >>> This issue was brought up before, but as far as I can tell, it wasn't fixed
> > >>> properly.
> > >>> This is initial Koen's post about the issue:
> > >>> http://projects.linuxtogo.org/pipermail/openembedded-devel/2008-April/005108.html
> > >>> And here is your own reply with one of the possible solutions:
> > >>> http://projects.linuxtogo.org/pipermail/openembedded-devel/2008-April/005111.html
> > >>
> > >> right. So can you try this patch and see if this fixes the problem ?
> 
> The proper fix seems to work fine. Thanks for applying the patch.

I see most of gcc-cross* recipes been cleaned removing the hack mentioned 
above, but there are few more, mostly CodeSourcery ones. The patch for those 
is in the ticket #4560:

http://bugs.openembedded.net/show_bug.cgi?id=4560

-- 
Denys

Re: RFC: SDK/external-toolchain fixes

[Denys Dmytriyenko]

On Fri, Sep 19, 2008 at 05:01:18PM -0400, Denys Dmytriyenko wrote:
> Khem,
> 
> On Mon, Sep 15, 2008 at 03:18:49AM -0400, Denys Dmytriyenko wrote:
> > On Sat, Sep 13, 2008 at 12:06:43PM -0700, Khem Raj wrote:
> > > >>> I agree that this is not a proper fix, but rather a hack. glibc has its own
> > > >>> limits.h, which wants to include gcc's limits.h like this:
> > > >>>
> > > >>> This issue was brought up before, but as far as I can tell, it wasn't fixed
> > > >>> properly.
> > > >>> This is initial Koen's post about the issue:
> > > >>> http://projects.linuxtogo.org/pipermail/openembedded-devel/2008-April/005108.html
> > > >>> And here is your own reply with one of the possible solutions:
> > > >>> http://projects.linuxtogo.org/pipermail/openembedded-devel/2008-April/005111.html
> > > >>
> > > >> right. So can you try this patch and see if this fixes the problem ?
> > 
> > The proper fix seems to work fine. Thanks for applying the patch.
> 
> I see most of gcc-cross* recipes been cleaned removing the hack mentioned 
> above, but there are few more, mostly CodeSourcery ones. The patch for those 
> is in the ticket #4560:
> 
> http://bugs.openembedded.net/show_bug.cgi?id=4560

Any updates for this item? Please let me know if the patch needs more work. 
Thanks.

-- 
Denys

Re: RFC: SDK/external-toolchain fixes

[Khem Raj]

I have no problems with this patch.

-Khem

On Tue, Sep 23, 2008 at 10:16 AM, Denys Dmytriyenko <denis@denix.org> wrote:
> On Fri, Sep 19, 2008 at 05:01:18PM -0400, Denys Dmytriyenko wrote:
>> Khem,
>>
>> On Mon, Sep 15, 2008 at 03:18:49AM -0400, Denys Dmytriyenko wrote:
>> > On Sat, Sep 13, 2008 at 12:06:43PM -0700, Khem Raj wrote:
>> > > >>> I agree that this is not a proper fix, but rather a hack. glibc has its own
>> > > >>> limits.h, which wants to include gcc's limits.h like this:
>> > > >>>
>> > > >>> This issue was brought up before, but as far as I can tell, it wasn't fixed
>> > > >>> properly.
>> > > >>> This is initial Koen's post about the issue:
>> > > >>> http://projects.linuxtogo.org/pipermail/openembedded-devel/2008-April/005108.html
>> > > >>> And here is your own reply with one of the possible solutions:
>> > > >>> http://projects.linuxtogo.org/pipermail/openembedded-devel/2008-April/005111.html
>> > > >>
>> > > >> right. So can you try this patch and see if this fixes the problem ?
>> >
>> > The proper fix seems to work fine. Thanks for applying the patch.
>>
>> I see most of gcc-cross* recipes been cleaned removing the hack mentioned
>> above, but there are few more, mostly CodeSourcery ones. The patch for those
>> is in the ticket #4560:
>>
>> http://bugs.openembedded.net/show_bug.cgi?id=4560
>
> Any updates for this item? Please let me know if the patch needs more work.
> Thanks.
>
> --
> Denys
>
> _______________________________________________
> Openembedded-devel mailing list
> Openembedded-devel@lists.openembedded.org
> http://lists.linuxtogo.org/cgi-bin/mailman/listinfo/openembedded-devel
>