[Buildroot] Reduce toolchain options

[Buildroot] Reduce toolchain options

[Gustavo Zacarias]

Hi.
How about we kill non-lfs and non-ipv6 toolchains?
It's 2015 after all, and the size delta is minor:

tar file delta, arm926 uclibc+busybox
=====================================
Threading (NPTL)        +153600 bytes
Largefile               +20480 bytes
IPv6                    +10240 bytes
RPC                     +61440 bytes
WCHAR                   +30720 bytes

Benefits:
* Less toolchain options to test/care for in packages
* More bang for the buck for autobuilders with less combinations to test

Even though the size delta would be somewhat more (tiny though) when
adding packages/libraries that's generally offset by the added size of
said packages.

The only downside i can think of is badly designed packages that won't
work well on a non-ipv6 linux kernel. These should be very few, if any.

LFS is a no brainer IMO.

Regards.

[Buildroot] Reduce toolchain options

[Peter Korsgaard]

>>>>> "Gustavo" == Gustavo Zacarias <gustavo@zacarias.com.ar> writes:

 > Hi.
 > How about we kill non-lfs and non-ipv6 toolchains?
 > It's 2015 after all, and the size delta is minor:

 > tar file delta, arm926 uclibc+busybox
 > =====================================
 > Threading (NPTL)        +153600 bytes
 > Largefile               +20480 bytes
 > IPv6                    +10240 bytes
 > RPC                     +61440 bytes
 > WCHAR                   +30720 bytes

 > Benefits:
 > * Less toolchain options to test/care for in packages
 > * More bang for the buck for autobuilders with less combinations to test

 > Even though the size delta would be somewhat more (tiny though) when
 > adding packages/libraries that's generally offset by the added size of
 > said packages.

 > The only downside i can think of is badly designed packages that won't
 > work well on a non-ipv6 linux kernel. These should be very few, if any.

 > LFS is a no brainer IMO.

Yes, I agree.

I'm also somewhat scared of forcing IPv6 support but perhaps it will be
safe, people using !uClibc have probably noticed these issues already.

-- 
Bye, Peter Korsgaard

[Buildroot] Reduce toolchain options

[Gustavo Zacarias]

On 03/11/2015 01:28 PM, Peter Korsgaard wrote:

> Yes, I agree.
> 
> I'm also somewhat scared of forcing IPv6 support but perhaps it will be
> safe, people using !uClibc have probably noticed these issues already.

I've found one downside of dropping non-IPv6: the codesourcery SH2A
toolchains lack IPv6 support :-(
We can still kill the LFS option though.
Regards.

[Buildroot] Reduce toolchain options

[Peter Korsgaard]

>>>>> "Gustavo" == Gustavo Zacarias <gustavo@zacarias.com.ar> writes:

 > On 03/11/2015 01:28 PM, Peter Korsgaard wrote:
 >> Yes, I agree.
 >> 
 >> I'm also somewhat scared of forcing IPv6 support but perhaps it will be
 >> safe, people using !uClibc have probably noticed these issues already.

 > I've found one downside of dropping non-IPv6: the codesourcery SH2A
 > toolchains lack IPv6 support :-(

Ahh, that might indeed be a problem for external uClibc based
toolchains.

> We can still kill the LFS option though.

Yeah.

-- 
Bye, Peter Korsgaard 

[Buildroot] Reduce toolchain options

[Thomas Petazzoni]

Dear Gustavo Zacarias,

On Wed, 11 Mar 2015 15:27:27 -0300, Gustavo Zacarias wrote:

> I've found one downside of dropping non-IPv6: the codesourcery SH2A
> toolchains lack IPv6 support :-(

As you say, it's 2015, so maybe we should just drop this toolchain. Or
at least deprecate it now so that we can get rid of it in the future.

Best regards,

Thomas
-- 
Thomas Petazzoni, CTO, Free Electrons
Embedded Linux, Kernel and Android engineering
http://free-electrons.com

[Buildroot] Reduce toolchain options

[Thomas Petazzoni]

Dear Gustavo Zacarias,

On Wed, 11 Mar 2015 13:14:33 -0300, Gustavo Zacarias wrote:

> How about we kill non-lfs and non-ipv6 toolchains?

I fully support this idea.

> It's 2015 after all, and the size delta is minor:
> 
> tar file delta, arm926 uclibc+busybox
> =====================================
> Threading (NPTL)        +153600 bytes

Ah, too bad. I was hoping to also suggest to remove the no-thread
solution. The number of packages that need threads these days is really
enormous, and I'm not really sure in practice how many people are doing
no-thread based systems.

Thomas
-- 
Thomas Petazzoni, CTO, Free Electrons
Embedded Linux, Kernel and Android engineering
http://free-electrons.com

[Buildroot] Reduce toolchain options

[Peter Korsgaard]

>>>>> "Thomas" == Thomas Petazzoni <thomas.petazzoni@free-electrons.com> writes:

Hi,


 >> It's 2015 after all, and the size delta is minor:
 >> 
 >> tar file delta, arm926 uclibc+busybox
 >> =====================================
 >> Threading (NPTL)        +153600 bytes

 > Ah, too bad. I was hoping to also suggest to remove the no-thread
 > solution. The number of packages that need threads these days is really
 > enormous, and I'm not really sure in practice how many people are doing
 > no-thread based systems.

Yeah, it is potentially useful for tiny initramfs'es with only busybox
and similar stuff, but not for anything bigger.

-- 
Bye, Peter Korsgaard

[Buildroot] Reduce toolchain options

[Gustavo Zacarias]

On 03/11/2015 05:45 PM, Thomas Petazzoni wrote:

>> It's 2015 after all, and the size delta is minor:
>>
>> tar file delta, arm926 uclibc+busybox
>> =====================================
>> Threading (NPTL)        +153600 bytes
> 
> Ah, too bad. I was hoping to also suggest to remove the no-thread
> solution. The number of packages that need threads these days is really
> enormous, and I'm not really sure in practice how many people are doing
> no-thread based systems.

I posted the numbers to get some opinions on the others.
Locale is potentially big because of all the message translations so i
didn't even size it.
Regards.

[Buildroot] Reduce toolchain options

[Gustavo Zacarias]

On 03/11/2015 05:44 PM, Thomas Petazzoni wrote:

> As you say, it's 2015, so maybe we should just drop this toolchain. Or
> at least deprecate it now so that we can get rid of it in the future.

I've considered that, but then recalled our internal SH toolchains
aren't up to snuff.
On the other hand it's a nommu sh target, how many people use this?
Regards.

[Buildroot] Reduce toolchain options

[Gustavo Zacarias]

On 03/11/2015 06:40 PM, Peter Korsgaard wrote:
> Yeah, it is potentially useful for tiny initramfs'es with only busybox
> and similar stuff, but not for anything bigger.

I'll cook the non-LFS killer patchset and send it and wait for opinions
on the others.
Regarding toolchains i'll keep the check for external ones to avoid
noise when an improper one is used.
Regards.

[Buildroot] Reduce toolchain options

[Arnout Vandecappelle]

On 11/03/15 17:14, Gustavo Zacarias wrote:
> 
> Hi.
> How about we kill non-lfs and non-ipv6 toolchains?

 Sounds like a good idea!

> It's 2015 after all, and the size delta is minor:
> 
> tar file delta, arm926 uclibc+busybox
> =====================================
> Threading (NPTL)        +153600 bytes
> Largefile               +20480 bytes
> IPv6                    +10240 bytes
> RPC                     +61440 bytes
> WCHAR                   +30720 bytes

 Could you calculate this with STATIC and thumb and also report the baseline?
IIRC a minimal statically linked busybox was only about 200K so the 20K
largefile is borderline significant...


 Regards,
 Arnout

> 
> Benefits:
> * Less toolchain options to test/care for in packages
> * More bang for the buck for autobuilders with less combinations to test
> 
> Even though the size delta would be somewhat more (tiny though) when
> adding packages/libraries that's generally offset by the added size of
> said packages.
> 
> The only downside i can think of is badly designed packages that won't
> work well on a non-ipv6 linux kernel. These should be very few, if any.
> 
> LFS is a no brainer IMO.
> 
> Regards.
> _______________________________________________
> buildroot mailing list
> buildroot at busybox.net
> http://lists.busybox.net/mailman/listinfo/buildroot
> 
> 


-- 
Arnout Vandecappelle                          arnout at mind be
Senior Embedded Software Architect            +32-16-286500
Essensium/Mind                                http://www.mind.be
G.Geenslaan 9, 3001 Leuven, Belgium           BE 872 984 063 RPR Leuven
LinkedIn profile: http://www.linkedin.com/in/arnoutvandecappelle
GPG fingerprint:  7CB5 E4CC 6C2E EFD4 6E3D A754 F963 ECAB 2450 2F1F

[Buildroot] Reduce toolchain options

[Gustavo Zacarias]

On 03/11/2015 07:04 PM, Arnout Vandecappelle wrote:

>  Could you calculate this with STATIC and thumb and also report the baseline?
> IIRC a minimal statically linked busybox was only about 200K so the 20K
> largefile is borderline significant...

Default busybox config arm926 thumb1 static (no threads):
LFS = 576212 bytes
non-LFS = 571156 bytes
Delta = 5056 bytes.

Regards.