On Mon, Aug 27, 2007 at 01:39:39AM +0200, Carsten Schlote wrote:
> > However, it'll be interesting to see how things turn out. If your
> > experiments look promising, it might be worth to re-activate our old
> > ColdFire boards.
>
> I hope to get an eval board soon. The next days we will know, if we go
> for mcf548x - chances are good :-) I already spoke with the hardware
> developer of the eval board, and he already got a linux terminal
> running on it. So first steps seem to be done already for Linux, and
> it sounds like a good starting point.

If you are interested I can attach a phyCORE-MCF548x board to our remote
lab.

What is your boot strategy? Last time I looked the Freescale method was
prety brain damaged, with a two stage booting process, involving dbug.
I'd go the u-boot-v2 way if you start something new.

> Yes, that unfortunetely might be true. So this is -maybe- one of my
> last chances to work on something different than i386 derivates or
> ARM/MIPS stuff. But a good and performant Linux available for these
> targets might change things.

Things become "good" and "performant" if people spend significant
ammounts of time/money on a platform. I doubt that this will happen for
m68k.

> I already made some thoughts about some kind of "nightly build/refresh
> toplevel makefile", which checks deps on the ptxconfigs in the
> toolchain - whenever a config updates, the makefile should rebuild the
> chain.

It's not only the configs but also the files in the repository itself.

> All status info (compiled/compile-breakage) and logfiles should be
> accessible via some html page created by this makefile. Of course,
> something similiar can be done with a bash script as well.

I'd write the script in bash - keep it simple and stupid. The result
should be some easily parsable ASCII table; one can create a nice HTML
page from it with a few lines of scripting. No need for active
components, it can all be cron triggered.

> Such toplevel makefile could 'svn export' the gcc trunk to some dummy
> archive (e.g. gcc-svn-trunk.tar.bz2), whenever the revision changes (ok,
> seems to happen every day anyway).

I'd start with the officially released upstream snapshots. You don't
want to follow each check-in, it takes too much time.

ftp://ftp.fu-berlin.de/unix/languages/gcc/snapshots

> A matching ptxconfig could be
> executed on this snapshot gcc. Same could be done to capture glibc
> changes. As soon as some patch starts to fail, we would be informed by
> the script.

I'd monitor the snapshots directory and trigger a script when something
new appears. Then run the scripts and report the result via mail.

> I never spent much time on these tests. I always got the impression,
> that these tests work best on the target system. That's no easy task,
> if you work on an embedded target with just a minimal root fs
> installed.

Nobody said that it's easy :) But running the regression test suites on
a regular base would definitely be a good thing for PTXdist.

> So we probably need some working eval platforms or emulators with some
> NFS mounted root FS.

Can be done here, at least for serveral ARM and PowerPC boards.

Robert
-- 
Pengutronix - Linux Solutions for Science and Industry
Entwicklungszentrum Nord     http://www.pengutronix.de

-- 
ptxdist mailing list
[email protected]

Reply via email to