Irc discussions from #mirage on 05-10-2016

Written by unpurecamelbot
Classified under: irclog
Published: 2016-10-05 (last updated: 2016-10-05)

05-10-2016 14:48 reynir o/

05-10-2016 15:00 mattg hello

05-10-2016 15:00 avsm greetings!

05-10-2016 15:01 avsm do we have an imaginary friend bot?

05-10-2016 15:01 yomimono unpurecamelbot is here

05-10-2016 15:02 mato \o/

05-10-2016 15:02 reynir ahh, I have to go for a while :(

05-10-2016 15:02 hannes o

05-10-2016 15:03 avsm alright, lets get started...

05-10-2016 15:03 avsm So there's been a crazy flurry of activity merging stuff into the trees in prep for the 3.0 release

05-10-2016 15:03 avsm mirage/mirage-dev has an increasing number of "trunk" libraries that have been ported to the new APIs

05-10-2016 15:03 avsm is fairly uptodate now

05-10-2016 15:04 avsm A bunch of people are travelling in Berlin at the moment (LinuxCon and other things), so I thought a brief update on open PRs would be useful

05-10-2016 15:04 yomimono for things that still need doing :)

05-10-2016 15:05 avsm #499 remains under risk "make it possible to specify package version constraints for unikernels"

05-10-2016 15:05 avsm I've been stuck in a deep hole of build rule rewrites, so not had a chance to look at it yet

05-10-2016 15:05 avsm how big a disaster is it if it doesnt land for 3.0 yomimono?

05-10-2016 15:07 hannes I'd claim it to be a blocker since only that allows us to specify the supported versions of the mirage tool (and types) of unikernels... if we don't have that, users will get strange errors trying to compile sth with an older unsupported tool suite

05-10-2016 15:07 yomimono avsm: there's minimal risk to human life or limb, but I think the number of questions/complaints we get that boil down to version incompatibilities will increase as we get more users

05-10-2016 15:07 yomimono avsm: it's a common thing that people bounce off of & a really big stumbling block imo

05-10-2016 15:07 avsm Yeah, it feels like this should be a release blocker

05-10-2016 15:07 avsm Lets leave it there at the very top of that issue until it gets sorted :)

05-10-2016 15:08 avsm Most of the effort recently has been on mirage-types

05-10-2016 15:08 avsm There are a few more things to merge that are all discussed on their relevant tickets well, but is there anything burning there for anyone?

05-10-2016 15:08 avsm The big change that is quite mechanical is renaming V1 to something else, but that can be done fairly easily in one pass through all the libraries.

05-10-2016 15:09 avsm I have a large rewrite of mirage-platform to use topkg, and am splitting out the C build rules into a separate mirage-build package

05-10-2016 15:09 yomimono I want to do that right before we tag the release

05-10-2016 15:09 yomimono avsm: that's great to hear!

05-10-2016 15:09 avsm this should make solo5 and xen coinstallable, since their stubs will be built separately

05-10-2016 15:09 avsm Some of the rules can be seen in the mirage-tcpip rewrite

05-10-2016 15:09 avsm (by rewrite i mean packaging rewrite)

05-10-2016 15:10 avsm Also, I am eliminating use of -pack in favour of module aliases as I do this, so unikernel binary sizes should reduce.

05-10-2016 15:10 hannes avsm: great!

05-10-2016 15:10 hannes well, moving C bindings out of mirage-platform to the subsequent libraries, topkg all the things... should be good!

05-10-2016 15:10 mato avsm: I've not yet had time to look at your build rule changes for mirage-platform, hopefully we can sit down together when you're in Berlin an go over it?

05-10-2016 15:11 avsm mato: yes definitely! I'm aiming to do more work on it on the plane; I discovered a bunch of rules I wrote a few years ago in mirage-platform that I can grab unmodified :)

05-10-2016 15:11 avsm I think mirage-solo5/xen coinstallability should be a release blocker as well, since it makes generating documentation very difficult otherwise

05-10-2016 15:11 mato Regarding Solo5, all things are mostly on track, my only concern is the stability of virtio drivers which is still not 100% on FreeBSD/bhyve

05-10-2016 15:12 avsm is going well -- I've been using it day-to-day now! There are currently some build failures, but it's a useful linting tool to make sure mirage-dev is building

05-10-2016 15:12 avsm mato: great, I'd also like to deploy the infrastructure using solo5 next week. The blocker is just moving into production (4.03 builds)

05-10-2016 15:13 avsm One random thing I've been experimenting with is tracking release notes at

05-10-2016 15:14 avsm yomimono: i believe anyone can drag these cards around, so it might be useful for scratch notes

05-10-2016 15:14 yomimono avsm: ack

05-10-2016 15:14 avsm Several people have raised a concern about the Travis CI slowness due to parallel builds

05-10-2016 15:15 avsm we've definitely been finding real issues from the multidistro builds, so ThomasG (currently in berlin) is working on the CI subsystem more

05-10-2016 15:15 avsm is running but is also dealing with a large volume of builds

05-10-2016 15:15 yomimono cheers

05-10-2016 15:16 avsm So, in terms of contributions, one very useful thing to do for anyone wishing to contribute is to start porting libraries in that do not use topkg to use it

05-10-2016 15:16 hannes your ci asks me for a password

05-10-2016 15:17 avsm hannes: hm, must be a revproxy. I'll disable the password shortly...

05-10-2016 15:17 amirmc I get a certificate invalid error

05-10-2016 15:17 avsm it's a self-signed certificate

05-10-2016 15:17 avsm it's only a development deployment atm, it might be easier if i shift it a letsencrypt VM instead

05-10-2016 15:18 mattg avsm: i will see about porting some libs to topkg

05-10-2016 15:18 avsm Re: topkg, if a package obeys the packaging convention, you get cool metadata

05-10-2016 15:18 hannes is there a CI doing 32bit builds?

05-10-2016 15:18 avsm and "odig changes" and "odig readme " and "odig version" all work

05-10-2016 15:18 avsm hannes: nope

05-10-2016 15:18 avsm hannes: could you give me an issue about that, i could probably hook up some ARMs

05-10-2016 15:19 yomimono

05-10-2016 15:19 avsm mattg: great! mailing list is good for questions, i have some notes about topkg tips i'll publish next week

05-10-2016 15:19 Drup avsm: wouldn't be simpler/more far reaching to do the small changes so that oasis packages obey the bunzli's packaging guidelines ... ?

05-10-2016 15:19 avsm Drup: yes that would be useful also

05-10-2016 15:19 hannes there you go

05-10-2016 15:20 avsm Drup: but I'm finding that a general refresh of the packaging is quite useful for many libraries to speed up builds, shift to module aliases instead of pack, and publish per-library docs. The topkg workflow is quite pleasant to work with

05-10-2016 15:20 hannes avsm: if you've general guidelines how to build C stubs (now that we have xen, unix, solo5) -- I suspect that'd be useful

05-10-2016 15:20 avsm hannes: thanks!

05-10-2016 15:20 avsm hannes: yes I will put full docs on it once it works :-) should be next week

05-10-2016 15:20 Drup avsm: As usual, the second rewrite is always better.

05-10-2016 15:21 avsm Drup: yeah there's nothing wrong with oasis -- topkg just collapses a few of the layers and is simpler

05-10-2016 15:21 avsm having an oasis plugin for odig would be very handy indeed

05-10-2016 15:22 Drup avsm: I won't argue about that here, I already expressed my opinion on other places. I just wanted to point out that doing this could probably be more far-reaching to the wider ocaml ecosystem.

05-10-2016 15:22 hannes I'd think is a blocker for Mirage-3?

05-10-2016 15:22 Drup I need to look at that

05-10-2016 15:22 Drup not sure when I'll have time

05-10-2016 15:22 avsm Drup: sure... as I said, an oasis/odig plugin would be welcome

05-10-2016 15:23 hannes is there a rough target date for the 3.0 release?

05-10-2016 15:24 yomimono one week ago, then this week... so... uh...

05-10-2016 15:24 hannes ah 'next wednesday'... well played ;)

05-10-2016 15:24 yomimono when it's ready

05-10-2016 15:24 yomimono real soon now

05-10-2016 15:24 avsm In terms of release schedules, it looks like we'll have to delay the release to get all these interface changes in. A bunch of us will be in Berlin on Saturday, so I'd suggest projecting a realistic date for next week. Can be discussed on the list

05-10-2016 15:25 avsm We're in good shape in that the mirage-dev trunks make it much easier to bulk test changes now

05-10-2016 15:25 avsm (as opposed to finding breakages in an released package, having to fork it to mirage-dev, etc etc)

05-10-2016 15:26 avsm Useful tip: will show breakage in the doc branch if something in mirage-dev is broken

05-10-2016 15:26 avsm it runs on every push to rebuild and is currently broken due to the connect changes (qcow-format)

05-10-2016 15:26 avsm so i just poll that to see what to fix next ;useful for anyone else wanting to be a human CI :)

05-10-2016 15:26 hannes if you announce date+place of saturdays berlin hacking, i might join in

05-10-2016 15:27 amirmc hannes: Friedrichshain.

05-10-2016 15:27 amirmc hannes: I'll send you a link

05-10-2016 15:27 avsm amirmc: thanks!

05-10-2016 15:27 hannes I've a branch for nbd (hannesm/nbd#error) which is requirement for qcow... I didn't understand enough qcow to modify it

05-10-2016 15:27 hannes but if nobody else fixes connect there, I might do later tonight

05-10-2016 15:28 avsm hannes: im not going to until i get through these c stubs changes, so if you do get a chance that'd be great

05-10-2016 15:28 avsm the next time mirage-dev builds, it might be worth cutting a few beta tags so we have some stable packages

05-10-2016 15:28 avsm and i can go around adding upper bounds in opam-stable

05-10-2016 15:28 avsm perhaps next week yomimono?

05-10-2016 15:29 yomimono avsm: sounds good to me

05-10-2016 15:29 hannes yes, opam-repo updates/conflicts can already be done: everything depending on mirage-types needs to be <3

05-10-2016 15:29 avsm in AOB, i notice ocamlclean has been refreshed to 4.03 :)

05-10-2016 15:29 avsm tis bytecode only, but very cool

05-10-2016 15:29 hannes and with that, I'm running to the airport...

05-10-2016 15:30 avsm i think we're mostly done! quick meeting today, back to hacking... anyone else got anything?

05-10-2016 15:30 hannes avsm: there's also the lto thing we should try... should work once we have 4.04 support

05-10-2016 15:30 yomimono we were going to discuss errors but I think we're losing our quorum

05-10-2016 15:30 avsm hannes: yes im porting mirag-xen to use ocaml-freeestanding

05-10-2016 15:30 yomimono same for 4.04 support

05-10-2016 15:30 avsm that should just give us 4.04 when its tagged

05-10-2016 15:31 avsm we are 10000000 commits behind on upstream openlibm, which will fix arm64 and musl support

05-10-2016 15:31 avsm so generally our upstreams have done a lot of work for us to get portable, hurrah :)

05-10-2016 15:31 mato ocaml-freestanding uses upstream libm

05-10-2016 15:31 avsm yeah, mirage-xen uses a random tag, since talex5 submitted his patches upstream and we had to release. those have since been merged.

05-10-2016 15:31 mato which reminds me, i should test it against 4.04. hopefully things will "just work"

05-10-2016 15:32 avsm mato: yes please -- the LTO branch is here

05-10-2016 15:33 mato avsm: is there an opam switch for it?

05-10-2016 15:33 avsm Also sean grove gave a fantastic CUFP talk on experiences with mirage; live notes here

05-10-2016 15:34 avsm mato: every ocaml pr has a switch autogenned; opam switch 4.05.0+pr608

05-10-2016 15:34 avsm wrapping up meeting now i think...

05-10-2016 15:34 yomimono thanks folks :)

05-10-2016 15:34 yomimono and avsm for facilitating :)

05-10-2016 15:34 avsm bows