[m-dev.] Continious integration testing

[Paul Bone]

On Tue, Jun 30, 2015 at 03:09:11PM +1000, Julien Fischer wrote:
> 
> Hi,
> 
> On Tue, 30 Jun 2015, Paul Bone wrote:
> 
> >On Fri, Jun 26, 2015 at 08:57:26PM +0000, Charles Shuller wrote:
> >>Hello All,
> >>
> >>I've received the following error on the 3 most recent ROTDs when trying to
> >>build them.
> >>
> >>There is no mmc on my path at configure time.
> >>
> >>mmc --compile-to-c --grade asm_fast.gc     --mercury-linkage shared --flags
> >>LIB_FLAGS   --flags INTER_FLAGS
> >>-R/home/charles/local/stow/mercury-rotd-2015-06-04/lib/mercury/lib/asm_fast.gc
> >>-R/home/charles/local/stow/mercury-rotd-2015-06-04/lib/mercury/lib
> >>--no-warn-insts-without-matching-type   array > array.err 2>&1
> >>/tmp/mmake.MuuPTs:67610: recipe for target 'array.c_date' failed
> >>
> >
> >I think we need to bring something like the old nightly test scripts back.
> 
> That would require having the hardware (or equivalent) to run them on.
> 
> >How did we mis this issue for three months?
> 
> We didn't, IIRC it was reported some time ago, but nobody got around to
> investigating it until the other day.  The old nightly test scripts
> wouldn't have caught it in any case since they didn't test the source
> distribution they generated.  (Indeed, even if they had tested it, they
> might not have caught it, since the problem only occurs when you install
> a 32-bit source distribution on to a 64-bit machine.)

What do you mean a 32-bit source distribution?  The only bit-ness in the
source distributions should be how many tag bits are used (two).

> >I have been installing the Mercury I need and only installing a new
> >one if I need it for some reason (eg: to build the git version).  Have
> >we all been doing this and therefore none of the developers have been
> >regularly testing the ROTD builds?
> 
> The ROTD builds are tested in the same way they always have been,
> bootcheck, install from stage 2, run tests etc.

Yes, but I guess this isn't on multiple systems with varing configurations.

> >It would be good to bring back some kind of testing rigor.  I would also
> >like to use something more like continuous integration (CI) rather than the
> >nightly build scripts.  CI will also give us more granular feedback (one
> >test run per change) and will not run if nothing changed.  I started to
> >investigate this in 2013 but stopped as I (then) lacked the disk space
> >required for so many builds.
> >
> >Thoughts?
> 
> Do you have the resources (incl. time) to do that?  If so, great!

Time is what I don't have, I have a couple of computers that I could have do
this as required.

If I had time & energy this is one of many things that I'd like to work
on.  It's probably about half way down my "Mercury things" list.


-- 
Paul Bone
http://www.bone.id.au