[RFC} Disable multilib in gcc recipes

[RFC} Disable multilib in gcc recipes

[Khem Raj]

Hi

I would like to propose to remove building multilib with gcc recipes.
It creates problems
and we do not package correct bits on some platforms. As we tend to
build toolchains keeping in mind the machine it is targetting I think
multilib is not that significant

Concerns ? comments?

Thx
-Khem

Re: [RFC} Disable multilib in gcc recipes

[Roman I Khimov]

В сообщении от Четверг 01 июля 2010 11:14:26 автор Khem Raj написал:
> I would like to propose to remove building multilib with gcc recipes.
> It creates problems
> and we do not package correct bits on some platforms. As we tend to
> build toolchains keeping in mind the machine it is targetting I think
> multilib is not that significant

One of the things I've thought of when working on amd64 multilib series. IMO 
it makes sense to make "--disable-multilib" a default and then only enable it 
for "multilib" override. I've just couldn't accurately estimate the potential 
impact for other platforms to propose it.

BTW, I plan to revive it and bring into oe.dev, we're regulary doing multilib 
amd64 builds there.

Re: [RFC} Disable multilib in gcc recipes

[Richard Purdie]

On Thu, 2010-07-01 at 00:14 -0700, Khem Raj wrote:
> I would like to propose to remove building multilib with gcc recipes.
> It creates problems
> and we do not package correct bits on some platforms. As we tend to
> build toolchains keeping in mind the machine it is targetting I think
> multilib is not that significant
> 
> Concerns ? comments?

I'm backing this. Currently multilib is broken and for a good
implementation we really needs a ground up proposal for how to handle it
well. I'm starting to look into the problem FWIW so I'd appreciate being
in the loop in any discussions.

Cheers,

Richard

Re: [RFC} Disable multilib in gcc recipes

[Roman I Khimov]

[-- Attachment #1: Type: Text/Plain, Size: 988 bytes --]

В сообщении от Пятница 02 июля 2010 02:01:34 автор Richard Purdie написал:
> On Thu, 2010-07-01 at 00:14 -0700, Khem Raj wrote:
> > I would like to propose to remove building multilib with gcc recipes.
> > It creates problems
> > and we do not package correct bits on some platforms. As we tend to
> > build toolchains keeping in mind the machine it is targetting I think
> > multilib is not that significant
> >
> > Concerns ? comments?
> 
> I'm backing this. Currently multilib is broken and for a good
> implementation we really needs a ground up proposal for how to handle it
> well. I'm starting to look into the problem FWIW so I'd appreciate being
> in the loop in any discussions.

Please take a look at "rik/amd64" branch. It is outdated, but shows the 
general approach. We're using it for quite some time.

-- 
 http://roman.khimov.ru
mailto: roman@khimov.ru
gpg --keyserver hkp://subkeys.pgp.net --recv-keys 0xE5E055C3

[-- Attachment #2: This is a digitally signed message part. --]
[-- Type: application/pgp-signature, Size: 205 bytes --]

Re: [RFC} Disable multilib in gcc recipes

[Detlef Vollmann]

On 07/02/10 00:01, Richard Purdie wrote:
> On Thu, 2010-07-01 at 00:14 -0700, Khem Raj wrote:
>> I would like to propose to remove building multilib with gcc recipes.
>> It creates problems
>> and we do not package correct bits on some platforms. As we tend to
>> build toolchains keeping in mind the machine it is targetting I think
>> multilib is not that significant
>>
>> Concerns ? comments?
> 
> I'm backing this. Currently multilib is broken and for a good
> implementation we really needs a ground up proposal for how to handle it
> well. I'm starting to look into the problem FWIW so I'd appreciate being
> in the loop in any discussions.
I've run into multilib issues as well, but could always solve it on a
case by case base (e.g. on powerpc).  But I agree it's currently a mess.
But what you need to keep in mind if you actually drop multilib support
is that you need to rename the prefix for the SDK toolchains.
It would be bad if PowerPC 32-bit and 64-bit toolchains were both named
powerpc-angstrom-linux-

   Detlef

Re: [RFC} Disable multilib in gcc recipes

[Khem Raj]

On Sun, Jul 4, 2010 at 6:59 AM, Detlef Vollmann <dv@vollmann.ch> wrote:
> On 07/02/10 00:01, Richard Purdie wrote:
>>
>> On Thu, 2010-07-01 at 00:14 -0700, Khem Raj wrote:
>>>
>>> I would like to propose to remove building multilib with gcc recipes.
>>> It creates problems
>>> and we do not package correct bits on some platforms. As we tend to
>>> build toolchains keeping in mind the machine it is targetting I think
>>> multilib is not that significant
>>>
>>> Concerns ? comments?
>>
>> I'm backing this. Currently multilib is broken and for a good
>> implementation we really needs a ground up proposal for how to handle it
>> well. I'm starting to look into the problem FWIW so I'd appreciate being
>> in the loop in any discussions.
>
> I've run into multilib issues as well, but could always solve it on a
> case by case base (e.g. on powerpc).  But I agree it's currently a mess.
> But what you need to keep in mind if you actually drop multilib support
> is that you need to rename the prefix for the SDK toolchains.
> It would be bad if PowerPC 32-bit and 64-bit toolchains were both named
> powerpc-angstrom-linux-

I think that does not happen. toolchain prefix is generally ${TARGET_ARCH}-x-x
for 64 bit powerpc TARGET_ARCH = "powerpc64"

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

Re: [RFC} Disable multilib in gcc recipes

[Detlef Vollmann]

On 07/04/10 19:52, Khem Raj wrote:
> On Sun, Jul 4, 2010 at 6:59 AM, Detlef Vollmann <dv@vollmann.ch> wrote:
>> But what you need to keep in mind if you actually drop multilib support
>> is that you need to rename the prefix for the SDK toolchains.
>> It would be bad if PowerPC 32-bit and 64-bit toolchains were both named
>> powerpc-angstrom-linux-
> 
> I think that does not happen. toolchain prefix is generally ${TARGET_ARCH}-x-x
> for 64 bit powerpc TARGET_ARCH = "powerpc64"
But TARGET_ARCH typically doesn't include what typ of floating point
you have.

   Detlef

Re: [RFC} Disable multilib in gcc recipes

[Tom Rini]

Detlef Vollmann wrote:
> On 07/02/10 00:01, Richard Purdie wrote:
>> On Thu, 2010-07-01 at 00:14 -0700, Khem Raj wrote:
>>> I would like to propose to remove building multilib with gcc recipes.
>>> It creates problems
>>> and we do not package correct bits on some platforms. As we tend to
>>> build toolchains keeping in mind the machine it is targetting I think
>>> multilib is not that significant
>>>
>>> Concerns ? comments?
>>
>> I'm backing this. Currently multilib is broken and for a good
>> implementation we really needs a ground up proposal for how to handle it
>> well. I'm starting to look into the problem FWIW so I'd appreciate being
>> in the loop in any discussions.
> I've run into multilib issues as well, but could always solve it on a
> case by case base (e.g. on powerpc).  But I agree it's currently a mess.
> But what you need to keep in mind if you actually drop multilib support
> is that you need to rename the prefix for the SDK toolchains.
> It would be bad if PowerPC 32-bit and 64-bit toolchains were both named
> powerpc-angstrom-linux-

Keep in mind that what's being suggested isn't dropping multilib support 
but being clear that it's not something that works right today while we 
re-evaluate doing it properly.  As you note, yes, the SDK case needs 
some thought today as the default paths conflict very easily (and can 
just as easily be worked around, stick FEED_ARCH or similar in there).

-- 
Tom Rini
Mentor Graphics Corporation