malta missed out from snapshot and release builds

[Daniel Golle]

On Sun, Feb 26, 2017 at 08:44:23PM +0800, Yousong Zhou wrote:
> On 26 February 2017 at 20:08, Felix Fietkau <nbd at nbd.name> wrote:
> > On 2017-02-26 12:11, Daniel Golle wrote:
> >> On Sun, Feb 26, 2017 at 05:55:02PM +0800, Yousong Zhou wrote:
> >>> On 26 February 2017 at 17:26, Matthias Schiffer
> >>> <mschiffer at universe-factory.net> wrote:
> >>> > On 02/26/2017 07:22 AM, Yousong Zhou wrote:
> >>> >> Timestamp of the last snapshot build is "Tue Jan 24 13:01:09 2017"
> >>> >> [1].  It's missing from all 17.01.0 release builds.
> >>> >>
> >>> >> Not sure how this happened, but we need to fix this.
> >>> >>
> >>> >>  [1] http://downloads.lede-project.org/snapshots/targets/malta/
> >>> >>
> >>> >> Regards,
> >>> >>                 yousong
> >>> >
> >>> > This is the effect of the source-only flag in target/linux/malta/Makefile,
> >>> > which deliberately excludes the target from buildbot builds. I think the
> >>> > reasoning was that the malta target is only useful for development and
> >>> > pre-built images just use up buildbot resources and mirror space.
> >>> >
> >>> > Matthias
> >>> >
> >>>
> >>> I see now, thanks.
> >>>
> >>> What do you guys think about bringing back the be and le subtarget?
> >>> They share the same arch as ramips and ar71xx and having such a target
> >>> readily available will be handy for testing, debugging, trying both
> >>> snapshot and release builds.
> >>
> >> +1 malta binaries *are* useful in practise, I use them with
> >> ImageBuilder for debugging MIPS software with tools too large to fit
> >> into the actual target RAM. Having release binaries for malta is just
> >> as crucial as it is for armvirt, this can be used to verify that
> >> software package X behaves well on MIPS(el) in a *reproducible and
> >> (host-)platform agnostic* way.
> >> And they are also useful for KVM-enabled targets (such as mt7621).
> >> If anything is missing, I'd happily help out here.
> > Do we need both le and be? Isn't one of those two enough?
> >
> > - Felix
> >
> 
> If I a choice has to be made, I will pick "be" ;)  I prefer it over le
> ever since hnyman reported an issue [1] which could only be triggered
> on a big-endian platform...  For the very same reason, it would be
> better if we can have them both.

To me also 'be' makes most sense to be part of the binary release as
armvirt and all the x86 targets are 'le' and having at least on
virtual 'be' target allows us to reproduce endianess-related bugs more
easily.

> 
> [1] https://github.com/openwrt/packages/commit/1e29676a8ac74f797f8ca799364681cec575ae6f#commitcomment-12901931
> 
>                 yousong
> 
> _______________________________________________
> lede-adm mailing list
> lede-adm at lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/lede-adm