[Buildroot] [arc-buildroot] [autobuild.buildroot.net] arc build results for 2016-06-19

[Alexey Brodkin]

Hi Arnout,

On Sat, 2016-06-25 at 00:46 +0200, Arnout Vandecappelle wrote:
> On 24-06-16 09:18, Alexey Brodkin wrote:
> > 
> > Hi Thomas,
> > 
> > On Fri, 2016-06-24 at 09:14 +0200, Thomas Petazzoni wrote:
> > > 
> > > Hello,
> > > 
> > > On Fri, 24 Jun 2016 04:34:55 +0000, Alexey Brodkin wrote:
> > > 
> > > > 
> > > > 
> > > > With arc-2016.03 toolchain we're seeing quite a lot of failures here and there.
> > > > That's sort of expected because we switched to binutils rewritten from scratch.
> > > > So we're still ramping up with these new binutils. What is also important these
> > > > new rewritten binutils are in upstream already. I.e. upcoming binutils 2.27 will
> > > > have everything ARC-specific from arc-2016.03 plus some more fixes and
> > > > enhancements that we made since March.
> > > OK, thanks for explaining the situation.
> > > 
> > > > 
> > > > 
> > > > And what we may do in Buildroot we may either wait for 2.27 binutils to be released
> > > > and then apply [backported from mainline master] patches on top of it to fix still
> > > > existing issues or alternatively we may start using so-called "engineering builds"
> > > > of binutils for ARC.
> > > > 
> > > > These "engineering builds" are basically snapshots [that pass internal review and
> > > > testing] made from our dev branch (arc-2016.09) on GitHub like the most recent
> > > > "arc-2016.09-eng004": https://github.com/foss-for-synopsys-dwc-arc-processors/binutils-gdb/releases/tag/arc-2016
> > > > .09-
> > > > eng0
> > > > 04
> > > > 
> > > > We'd prefer to go with "engineering builds" simply because we foresee still a lot
> > > > of changes in ARC port of binutils (remember our port is in its childhood currently)
> > > > while upstream binutils see release about once a year. Which means adding fixes and
> > > > enhancements on top of 2.27 release at some point will become a support nightmare.
> > > > 
> > > > What do you guys think about all that?
> > > I think at some point we will want to use the upstream version of
> > > binutils if there is ARC support upstream. However, I definitely
> > > understand that the upstream support may not be fully ready overnight,
> > > so I'm fine with using those engineering builds for now, and then move
> > > to using the upstream binutils version for binutils 2.28 for example
> > > (or 2.29 if 2.28 is still not good enough).
> > Ok cool!
> > 
> > Then let's get the party started.
> > 
> > Vlad will send a patch which bump ARC tools to arc-2016.09-eng004 shortly.
> > Since other components like gcc and gdb are also not fully upstream we'll be
> > bumping all 3 components at once:
> >  1. Binutils
> >  2. Gcc
> >  3. Gdb
>  I guess all of these will be updated regularly, right? That means Thomas also
> has to respin the autobuilder toolchains with every update. If Synopsys anyway
> produces toolchain binaries already, it could be interesting to add these as
> external toolchains.

Good point.

Actually we didn't plan to put prebuilt tools in the same location as released versions
https://github.com/foss-for-synopsys-dwc-arc-processors/toolchain/releases simply to not confuse
real users (otherwise people may think it's ok to use those engineering builds for production).

Moreover I remember we discussed that topic of ARC prebuilt tools in Buildroot and conclusion was
to disable our prebuilt tools from autobuilder tests because our uClibc configuration had some missing
options if compared to Buildroot.

As the simplest solution (but probably not the best from performance standpoint) could be
to disable use of prebuilt tools for ARC before we have something more or less stable.

But let's discuss all options what we have first.

-Alexey