These are chat archives for machinekit/machinekit

25th
Nov 2015
Michael Haberler
@mhaberler
Nov 25 2015 00:27
looks like avahi-daemon works fine (so far)
GP Orcullo
@kinsamanka
Nov 25 2015 00:46
PR triggers are solved then
Let me finish consolidating all the containers in one place before we start integrating with machinekit.
We also need to solve the issue about the version numbers
Michael Haberler
@mhaberler
Nov 25 2015 00:57
yeah, John is in the claws of the family, will take a day or two for him to reappear
state of siege if I understand it correctly
GP Orcullo
@kinsamanka
Nov 25 2015 11:01
The pkg build for armhf (3 flavors) just barely made it at 47 mins
Michael Haberler
@mhaberler
Nov 25 2015 11:02
was that all in one go?
maybe we should do travis just for regressions and do pkg builds on mah.priv.at for now? we can always move it/clone it on travis post-split
GP Orcullo
@kinsamanka
Nov 25 2015 11:03
Yes
I haven't yet checked how long it took to build on your system
I think splitting it to 3 jobs would reduce the time further
Michael Haberler
@mhaberler
Nov 25 2015 11:06
well if we go that route, then the goal for the "travis regression test container" becomes: amd64/wheezy(I guess), run RIP build for now, run regressions, integrate with github - I guess this one triggers on all PR's
re splitting: sure, but I would prefer to split once cmake gets traction.
btw thinking of it, the mk-protobuf package will be required by both mk-cnc and mk-rt; halcmd/rtapi_app/cython bindings user protobuf messaging
GP Orcullo
@kinsamanka
Nov 25 2015 11:08
I think @bobvanderlinden can start integrating the travis tests on machinekit. :smiley:
Michael Haberler
@mhaberler
Nov 25 2015 11:08
gotta run, will be on reduced presence until saturday
fair enough
GP Orcullo
@kinsamanka
Nov 25 2015 11:09
What i meant was separate the jobs into individual flavors
Michael Haberler
@mhaberler
Nov 25 2015 11:09
I see
well broken merges can be patched up quickly, and no breakage potential, so not much danger in trying something out
bbl
GP Orcullo
@kinsamanka
Nov 25 2015 11:11
Tc
Michael Haberler
@mhaberler
Nov 25 2015 12:10
#794 : John reported about problems with avahi and chroot
but if we are doing regressions on amd64/posix flavor there would be no requirement to do a chroot or proot
GP Orcullo
@kinsamanka
Nov 25 2015 13:50
I haven't had issues with x86, only arm. I'll check it later
Bob van der Linden
@bobvanderlinden
Nov 25 2015 21:10
Michael Haberler
@mhaberler
Nov 25 2015 21:11
yay!
Bob van der Linden
@bobvanderlinden
Nov 25 2015 21:17
@mhaberler If you want, you can do a PR on https://github.com/bobvanderlinden/machinekit to check whether PRs work well
Michael Haberler
@mhaberler
Nov 25 2015 21:17
those are pkg builds, not regressions right?
Bob van der Linden
@bobvanderlinden
Nov 25 2015 21:19
the first one is doing the tests (runtests tests)
Michael Haberler
@mhaberler
Nov 25 2015 21:19
aja
Bob van der Linden
@bobvanderlinden
Nov 25 2015 21:19
19.1
ah that's one is still failing
Michael Haberler
@mhaberler
Nov 25 2015 21:21
well but the fail seems to be just a wrong path issue or so
Bob van der Linden
@bobvanderlinden
Nov 25 2015 21:21
yes exactly, that should be fixed in build 20.1
Michael Haberler
@mhaberler
Nov 25 2015 21:21
boy, this is zipping along!
Bob van der Linden
@bobvanderlinden
Nov 25 2015 21:24
i've now basically copied @kinsamanka 's scripts and they need to adhere to the machinekit directory structure (before it would clone machinekit, now machinekit is cloned by travis)
we'll see how it goes
i'm off for now
Michael Haberler
@mhaberler
Nov 25 2015 21:24
cu!
Bob van der Linden
@bobvanderlinden
Nov 25 2015 22:42
yay, build passed
Michael Haberler
@mhaberler
Nov 25 2015 22:42
wow!
Bob van der Linden
@bobvanderlinden
Nov 25 2015 22:42
what i'm wondering though is how the deployment should work when the tests are run in parallel
or rather, the .deb packages are built in parallel to the tests
Michael Haberler
@mhaberler
Nov 25 2015 22:43
good point, I think the package build should only commence once runtests pass
not sure how that can be chained
maybe tests just on PR's and pkg build only on merge
not the same thing though
Bob van der Linden
@bobvanderlinden
Nov 25 2015 22:44
on master it should build, test, package and deploy
on PR it should build and test
Michael Haberler
@mhaberler
Nov 25 2015 22:45
fair enough
Bob van der Linden
@bobvanderlinden
Nov 25 2015 22:45
(i think)
Michael Haberler
@mhaberler
Nov 25 2015 22:45
sounds right
Bob van der Linden
@bobvanderlinden
Nov 25 2015 22:45
but the flavors and all... hmm
Michael Haberler
@mhaberler
Nov 25 2015 22:45
in practice there is low value in testing anything else than amd64/posix
Bob van der Linden
@bobvanderlinden
Nov 25 2015 22:45
ah ok
Michael Haberler
@mhaberler
Nov 25 2015 22:46
80/20 .. not really worth the difference, and very few people touch xenomai-related code, those usually know how to runtests locally
so not worth the effort
maybe there is a simple way to record a certain sha passed, and make the build contional on that
Bob van der Linden
@bobvanderlinden
Nov 25 2015 22:49
sha?
oh like git sha
Michael Haberler
@mhaberler
Nov 25 2015 22:50
is there a way to swindle a private key into a travis run without it being divulged?
Bob van der Linden
@bobvanderlinden
Nov 25 2015 22:50
ya, there isn't really a 'task' that can run after all parallel builds are done
if you mean private key for deployment, that should be doable
Michael Haberler
@mhaberler
Nov 25 2015 22:50
that would be needed for pkg uploads but could be used to record passed sha's say in a redis server on the apt-repo
right
well I guess you need to leave some pro-forma work for John anyway ;)
but that doesnt look like phd-required work
Bob van der Linden
@bobvanderlinden
Nov 25 2015 22:52
haha
Michael Haberler
@mhaberler
Nov 25 2015 22:52
no, really - you and @kinsamanka did an absolutely stunning job!
Bob van der Linden
@bobvanderlinden
Nov 25 2015 22:52
well @zultron can do pull requests at github.com/bobvanderlinden/machinekit I'm still wondering how well that works
thanks! it's really awesome services like travis and docker are available nowadays for free for projects like these
Michael Haberler
@mhaberler
Nov 25 2015 22:54
yep, big lever - good it made me have to learn docker a bit, super useful
Bob van der Linden
@bobvanderlinden
Nov 25 2015 22:54
indeed
Michael Haberler
@mhaberler
Nov 25 2015 22:55
the same scheme can be used for all the few private packages we still require, that will take a lot less magic downstream
Bob van der Linden
@bobvanderlinden
Nov 25 2015 22:56
yes! for most of those libraries, it should be a lot simpeler (no flavors, though they do have different architectures)
we could also already look at building machinetalk-protobuf
Michael Haberler
@mhaberler
Nov 25 2015 22:56
this is an issue.. we are on purpose using pretty bleeding edge stuff in zeroMQ and Alex suggests some changes which would require switching to a newer zeroMQ (they still have to freeze a release)
yes
it occurred to be both mk-rt and mk-cnc will have to depend on that one
Bob van der Linden
@bobvanderlinden
Nov 25 2015 22:57
yes, because of the nanopb stuff right?
Michael Haberler
@mhaberler
Nov 25 2015 22:57
I think you, me and Alex should hash out a short requirements text for that package
no, halcmd/rtapi_app comms uses it, and mk-cnc will once we start dumping NML/RCS for good
and the cython bindings use it too
Bob van der Linden
@bobvanderlinden
Nov 25 2015 22:58
they'll be different packages... it might be a good idea to start with python as it is standardized and can be deployed to pypi automatically
Michael Haberler
@mhaberler
Nov 25 2015 22:58
good point
Bob van der Linden
@bobvanderlinden
Nov 25 2015 22:59
it will be nice to have a .deb for the C/C++ stuff as well
not sure if bare .proto files should also be an package
Michael Haberler
@mhaberler
Nov 25 2015 22:59
so we'd have a mk-protobuf which has the C++/C descriptor libs, and the protos
Bob van der Linden
@bobvanderlinden
Nov 25 2015 22:59
maybe integrate that into the C/C++ package
Michael Haberler
@mhaberler
Nov 25 2015 22:59
or will the protos be in the py package?
Bob van der Linden
@bobvanderlinden
Nov 25 2015 23:00
nah, i don't think that fits the py package
not sure though
Michael Haberler
@mhaberler
Nov 25 2015 23:00
yeah, maybe just a separate package for the protos, why not
the c++/c stuff doesnt need it post build, but other folks downstream might
there's runtime parsing of protos possible even with C++, then protobuf.js needs it, so not a 1:1 mapping
Bob van der Linden
@bobvanderlinden
Nov 25 2015 23:01
i need it for node... the node package mostly contains the .proto files, because protobufjs can load those at runtime
Michael Haberler
@mhaberler
Nov 25 2015 23:01
I assume there is no way to express a deb prerequisite when doing a pypi install?
Bob van der Linden
@bobvanderlinden
Nov 25 2015 23:01
hmm not sure
but i don't like that
distro-dependent ;)
Michael Haberler
@mhaberler
Nov 25 2015 23:02
well shouldnt be too hard to come up with an instructive error message if the protos arent found
Bob van der Linden
@bobvanderlinden
Nov 25 2015 23:02
maybe some install assertion with pkg-config
indeed
for node I just include them in the package, independent of the system it's running on
that means it'll work on windows and mac osx as well
Michael Haberler
@mhaberler
Nov 25 2015 23:04
I would want to avoid a situation where two sets of protos are installed, recipe for trouble
Bob van der Linden
@bobvanderlinden
Nov 25 2015 23:04
by just installing machinetalk-protobuf with node's package manager (and in turn by just installing node-machinetalk (and in turn by just installing the webgui))
Michael Haberler
@mhaberler
Nov 25 2015 23:04
let's do this in an issue in machinetalk-protobuf and bring in Alex
Bob van der Linden
@bobvanderlinden
Nov 25 2015 23:04
well, how would you know the application is compatible with the protobuf that's on the system?
agred
*agreed
Michael Haberler
@mhaberler
Nov 25 2015 23:05
hum
ha, wonderful to discuss this, as the build stuff is nearing completion !
finally we'll be able to instill some sanity into the structure..
btw when we first talked, I advised against your plans to go for a cmake build
Bob van der Linden
@bobvanderlinden
Nov 25 2015 23:07
yes, I hope so
Michael Haberler
@mhaberler
Nov 25 2015 23:07
this was on the assumption that kmods/RTAI are still in scope
Bob van der Linden
@bobvanderlinden
Nov 25 2015 23:07
hehe yes I remember that
i understand
Michael Haberler
@mhaberler
Nov 25 2015 23:07
that is fry, so my remarks dont count anymore
Bob van der Linden
@bobvanderlinden
Nov 25 2015 23:08
i was a bit bummed out by it, because I don't like running a VM... I have a hard time running VMs on my laptop, because I already need Windows + NixOS installed on my system on a SSD that's only 250GB
Michael Haberler
@mhaberler
Nov 25 2015 23:08
that pushes the envelope, yes
I bit the bullet and got me a MB pro with 16GB/1TBssd
Bob van der Linden
@bobvanderlinden
Nov 25 2015 23:09
yes it's good to hear the difficult flavors are gone :P
oh wow
1TB ssd?
Michael Haberler
@mhaberler
Nov 25 2015 23:09
dont ask me on cost
Bob van der Linden
@bobvanderlinden
Nov 25 2015 23:09
oh like a hubrid
what did it cost? :P
Michael Haberler
@mhaberler
Nov 25 2015 23:10
nope, all ssd
4500ish or so
Bob van der Linden
@bobvanderlinden
Nov 25 2015 23:10
wow
Michael Haberler
@mhaberler
Nov 25 2015 23:10
one time pain
Bob van der Linden
@bobvanderlinden
Nov 25 2015 23:10
should be good for quite a while though
Michael Haberler
@mhaberler
Nov 25 2015 23:10
I hope
Bob van der Linden
@bobvanderlinden
Nov 25 2015 23:10
1TB ssd... that is nice :)
Michael Haberler
@mhaberler
Nov 25 2015 23:11
yeah, hashing out the final package structure and where to cut is next
maybe we should do this in a machinekit/machinekit issue,and bring in the machinetalk-protobuf related concerns there as well
almost nobody watches machinetalk-protobuf
Bob van der Linden
@bobvanderlinden
Nov 25 2015 23:12
yes, i hope the cmake stuff is going to make it easier to restructure things a bit
Michael Haberler
@mhaberler
Nov 25 2015 23:12
but that structure must fit all of the code not just pb
Bob van der Linden
@bobvanderlinden
Nov 25 2015 23:13
machinetalk-protobuf can be a separate issue i think, it's already potentially 3 separate packages
Michael Haberler
@mhaberler
Nov 25 2015 23:13
first task is to nail down what goes where
Bob van der Linden
@bobvanderlinden
Nov 25 2015 23:13
well, once the cmake stuff is near completion, I'd like to restructure the directories a bit
src/emc/task builds the linuxcnc if i remember correctly
Michael Haberler
@mhaberler
Nov 25 2015 23:14
keep git history and mergeability in mind
Bob van der Linden
@bobvanderlinden
Nov 25 2015 23:14
  • linuxcnc binary
yes, that's still tricky
Michael Haberler
@mhaberler
Nov 25 2015 23:14
breaks in history are a no-no
Bob van der Linden
@bobvanderlinden
Nov 25 2015 23:15
I had used git mv to restructure the directories, but I think the cmake stuff should be applied to machinekit/machinekit first and once that's building, we can restructure the directories
Michael Haberler
@mhaberler
Nov 25 2015 23:15
but as long as one separates a dir reshuffle from actual change commits git does pretty well
yes, first kill Submakefiles, then split
Bob van der Linden
@bobvanderlinden
Nov 25 2015 23:16
i changed the directories, because the current build system would create one big pile of .h files in one include directory... i wanted each thing separate
Michael Haberler
@mhaberler
Nov 25 2015 23:16
that include stuff is under "top arcana"
Bob van der Linden
@bobvanderlinden
Nov 25 2015 23:16
top arcana?
Michael Haberler
@mhaberler
Nov 25 2015 23:17
well bizaree
Bob van der Linden
@bobvanderlinden
Nov 25 2015 23:17
ah ok :P
well, we'll see... one problem at the time :)
i'm off to bed though
Michael Haberler
@mhaberler
Nov 25 2015 23:17
anyway, I guess I'll crash, off for two days
cu! superb job, Bob!
Bob van der Linden
@bobvanderlinden
Nov 25 2015 23:17
kk
see you later then
thanks!
Michael Haberler
@mhaberler
Nov 25 2015 23:18
that really moves things forward
Bob van der Linden
@bobvanderlinden
Nov 25 2015 23:18
yes, i hope so. still a LOT to do though :/
we'll see how that all goes
anyway, i'm off