Maverick toolchain

[Robbie Williamson]

On Tue, 2010-07-20 at 21:45 +0200, Matthias Klose wrote: 
> On 20.07.2010 21:03, Mark Shuttleworth wrote:
> > On 20/07/10 16:53, Robbie Williamson wrote:
> >> On 20/07/10 12:25, Matthias Klose wrote:
> >>
> >>> I suggest to enable the Linaro GCC on amd64, i386 and powerpc as well,
> >>> given the  comittment of the Linaro toolchain working group to
> >>> addressing potential regressions on amd64 and i386 too.
> >>>
> >>
> >> Doko, thank you for the feedback. Given that we are midway through the
> >> release cycle, and that we have not done *any* system testing with the
> >> amd64 and i386 packages/kernels built using this toolchain, I'm
> >> concerned with making this change now.  As acting release manager, I
> >> propose we move to Linaro's GCC now for the armel flavor of Ubuntu
> >> 10.10, and make the change for amd64 and i386 in 11.04.
> >>
> >
> > I don't want to rock the RM boat, especially since I'm responsible for
> > the tighter timelines of 10.10, but this is a change which has been in
> > play since UDS, has had extensive analysis and a full rebuild test with
> > detailed examination. It's certainly not a new thing being raised at
> > this stage, and we agreed not to press the issue earlier only because
> > the offer was made to do the testing and decide then. It would be wrong
> > to have made that offer and then decline to make the change even after
> > the testing comes back as favourably as it did - next time, folks won't
> > be as willing to leave the door open for extended analysis.
> >
> > It's also quite material to the collaboration of Linaro and the rest of
> > the platform.
> 
> Afaiu the biggest concern to the Linaro GCC is it's use for the kernel build. 
> Apparently nobody including myself did explicitly communicate to the kernel team 
> about the decision in the arm toolchain session (arm-m-tool-chain-selection).
In addition, nothing has ever been posted to ubuntu-devel or
ubuntu-devel-discuss about this, so I feel it may catch many Ubuntu
contributors and our derivatives off guard.  I know this was discussed
at UDS, but I don't know if most thought a session titled "ARM Toolchain
Selection" [1] encompassed toolchain selection for non-ARM architectures
as well.  

[1] https://wiki.ubuntu.com/Specs/M/ARMToolChainSelection

> 
> If this is the only major concern I'd like to propose to use the gcc-4.4-fsf 
> packages (which still need landing in maverick) to build the kernel (shipping 
> the gcc-4.4-fsf binary with the name kgcc on the CDs).
Please talk to the kernel team and see what they say.  

> 
> Another concern raised is the amount of rebuilds which may hit the archive for 
> the rebuild of package to make use of the new arm optimizations.  According to 
> James T.  a rebuild of all packages found on the CD images would be fine, and 
> maybe as well a rebuild of main if this is stretched over some time.
> 
>    Matthias
As far as I know, the Linaro team is fine with only having support for
armel in Ubuntu 10.10, if I'm wrong I will assume Loic/Steve will
correct me.  

With all this said, as someone who's pushed in major boot changes late
in a release cycle, I personally don't mind if the toolchain is
switched. ;)  However, as release manager I feel it's my responsibility
to proxy for those who are more risk adverse than myself.  If the tech
board is fine with this move, I will support it.

-- 
Robbie Williamson                                      robbie at canonical.com 
Canonical, Ltd.                                   robbiew[irc.freenode.net]                               

"You can't be lucky all the time, but you can be smart everyday" 
 -Mos Def

"Arrogance is thinking you are better than everyone else, while
Confidence is knowing no one else is better than you." -Me ;)