Experimental Physics and
Industrial Control System

Gedare Bloom via Core-talk <core-talk at aps.anl.gov> · Sat, 13 Feb 2021 13:45:48 -0700

On Fri, Feb 12, 2021 at 2:01 PM Michael Davidsaver
<mdavidsaver at gmail.com> wrote:
>
> On 2/12/21 10:47 AM, Gedare Bloom wrote:
> > On Thu, Feb 11, 2021 at 3:12 PM Michael Davidsaver
> > <mdavidsaver at gmail.com> wrote:
> >>
> >> On 2/11/21 1:23 PM, Johnson, Andrew N. via Core-talk wrote:
> >>> Now speaking as an EPICS Core developer the EPICS CI process on GitHub Actions current runs our built-in unit tests against both 4.9 and 4.10 on i386 under qemu, and for RTEMS-5 we will test against i686 under qemu. We could add other architectures too if they can be made to run under qemu. The RTEMS builds that we use are compiled by Michael Davidsaver – we don’t try to use the latest RTEMS git version and I’m not sure exactly which tags his images were built against.
> >>
> >> For the past few years I've been publishing pre-built toolchains
> >> for CI builders.  Practically this has meant building for some
> >> version of Ubuntu.  So far this has been a mix of manual and
> >> automatic processes due CI build time limits, changing CI environments,
> >> as well as my own time constraints.
> >>
> >> The result has been infrequent builds for a small number of BSPs.
> >>
> >> https://github.com/mdavidsaver/rsb/releases
> >>
> > Can you give a little info about what version combinations you test,
> > with 4.9.x and 4.10.x especially? This will help inform the
> > need/desire for longer term support on 4.10.  From RTEMS perspective,
> > 4.9 is EOL. Some patches were put on after the last release, but I
> > don't think anyone has expressed interest in basing new development on
> > 4.9 or in updating their current 4.9 projects at all.
>
> I understand, and didn't expect anything different.  We're still testing
> with 4.9.6 largely because the cost of supporting it has so far been low.
> Our major pain points in terms of compatibility have been with other
> platforms.  This will change with time though, as GCC 4.3 (and 4.4)
> continue to age.
>
> >>
> >> While I don't have a huge amount of time to spend on this, I am looking
> >> at expanding this to the full list of BSPs with configuration in EPICS Base.
> >>
> >> https://github.com/mdavidsaver/rsb/blob/ci/.github/workflows/ci-scripts-build.yml
> >>
> >> I'd be thrilled to have some help with this.
> >
> > This is something that I can try to get my University team to
> > contribute to as well. I also have interest in automating some static
> > analysis tools (e.g., trigger Coverity Scan from travis-ci plugin or
> > similar), but the challenge there is how to integrate the reports with
> > bug tracking (launchpad) without swamping maintainers with false
> > positives. We've struggled some with that in RTEMS, but if EPICS
> > maintainers are interested in at least seeing what the Coverity output
> > may be like for Base, that is something I can put resources into
> > standing up. Let me know, and we can figure out what version(s) to
> > focus on.
>
> Yes, static analysis is useful if the SNR is high enough.
>
> > The management of 4.10 build tools is one of the RTEMS Project key
> > concerns if we go toward LTS. Since the RSB support for 4.10 was not
> > quite mature, we envision archiving reasonably stable sources and
> > hopefully having a minimal patching and build scripting
> > infrastructure. Anything beyond that would likely be prohibitive from
> > a volunteer/maintenance perspective.
>
> Actually.  One data point which might be worth considering.
> RTEMS 4.10 currently ships with GCC 4.4.7.  A big milestone
> for me is C++11 support, which isn't really present until
> GCC 4.8 or 4.9.
>

If we went to an LTS model for 4.10, bumping up the GCC toolchain
might be possible. We could probably go up as high as gcc-7. The
biggest challenge though is that we have to avoid touching newlib too
much, since the RTEMS-newlib headers are tightly integrated. It would
require a lot of testing to make sure things don't break, and to
triage new warnings, ICE faults, or other problems that arise. Hard to
say if it would happen (on volunteer time), and only a possibility if
we (RTEMS Project) do end up committing to an LTS branch on 4.10. So
far only two of us (Joel and I) have really expressed an interest in
seeing 4.10.LTS.

> > Sorry to be a little confusing--as with you guys, I also have to wear
> > two hats (University and RTEMS Project).
> >
> > -Gedare
> >
>

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System