11:51 <Andres__> I would like to write a stateful protocol (over tcp) for some app similar to tcp ack-based protocol stack in ocaml. For instance, I have a start handshake, and then data passing, together with sending acks. Is there any piece of code that you can point me to as a best practice so I can start the coding using that structure?

12:57 <Andres__> Is there anyone in here that can answer my questions on mitrage tcp library?

15:01 <talex5> hi

15:01 <yomimono> hi!

15:01 <yomimono> must be time for a meeting!

15:02 <yomimono> The agenda is at https://github.com/mirage/mirage-www/wiki/Call-Agenda .

15:02 <yomimono> There's still time to add items if there's something you'd like to discuss :)

15:03 <Drup> Does "wee ICFP" counts as agenda item ?

15:03 <yomimono> I'll add it! :D

15:03 <yomimono> Might be nice to get a sense of who's going - I am :)

15:03 <Drup> Me too!

15:04 <Drup> I'm arriving on saturday evening, if people want to get out to eat

15:04 <yomimono> first up, "wee ICFP" - the International Conference on Functional Programming (I think?) is happening in Oxford next week

15:04 <yomimono> there will be some Cambridge contingent as well, I hear :)

15:05 <yomimono> as in, some folks coming from Cambridge

15:05 <mato> Oh, it's in Europe! I didn't know that.

15:05 <mato> Err... near Europe. Or something.

15:05 <yomimono> Still time for last minute flights :D

15:05 <yomimono> http://icfp17.sigplan.org/ for more info :)

15:07 <yomimono> g2p is presenting the storage work (Wodan) at OCaml Workshop, and there are lots of other talks that might be interesting to Mirage folks

15:07 <yomimono> If nothing else on that, let's move on --

15:07 <yomimono> I had put an item up asking about solo5, because I realized I lost track of what was going on there

15:08 <yomimono> we'd talked about a 3.1 release for the refactored and ARM-ready solo5

15:08 <yomimono> is that at a point where it needs some release managing, or can I go on being blissfully unaware?

15:08 <mato> Unfortunately I don't have an update on this today, I've been swamped by other work at $CORP and had no time for Solo5 :-( :-(

15:09 <yomimono> whew, then nobody's mad at me for being too funemployed to keep an eye on it!

15:09 <djwillia> ricarkol and I are in the process of writing up a paper about record/replay for ukvm/solo5 unikernels

15:10 <mato> But, TL;DR, I have a list of things in my head that need to be done for a next release which would surface all the new goodness we have since the last one, so perhaps I should put those down in an issue somewhere at least?

15:10 <yomimono> mato: that would be very appreciated!

15:10 <djwillia> we haven't talked to mato about how/when any of that should be merged in though

15:10 <djwillia> when we finish a draft we would love to share it for some feedback

15:10 <yomimono> djwillia: that sounds very cool! is there somewhere people can look for more info, or is it still under wraps?

15:11 <yomimono> djwillia: oh, jinx :) I'll look forward to seeing it!

15:11 <djwillia> we hope to send it to Eurosys, which has a deadline at the end of october

15:12 <yomimono> good luck, although I doubt you need it!

15:13 <djwillia> thanks! we'll take all the luck we can get :)

15:14 <yomimono> If nothing else on solo5 --

15:14 <yomimono> I also put in an agenda item for dropping OCaml 4.03.0 support and adding 4.05.0

15:15 <yomimono> but more generally speaking, it would be useful to have some guidelines for how we intend to keep up with compiler releases, now that they're timeboxed and not feature-boxed

15:15 <avsm> ahem, apologies for tardiness

15:16 <yomimono> nbd, glad you're here :)

15:16 <yomimono> was just saying that I'd like to add 4.05.0 support and drop 4.03.0

15:16 <avsm> yes. the only blocker to 4.05.0 support is unfortunately rebasing xen to use the new freestanding minios

15:16 <yomimono> and have some policy written down about which compilers we aim to support, e.g. the two most recent major releases

15:16 <mato> FWIW ocaml-freestanding (and thus mirage/solo5) got 4.05.0 some time ago and should "just work"

15:17 <avsm> i parked my trees that had that work; i'll have time to finish it right after icfp

15:17 <avsm> i am very keen to retire our minios fork and just use the -freestanding variant for both xen and solo5

15:18 <yomimono> I think you're not alone :)

15:18 <avsm> I've been working through the new CI and have an almost working node registration and command executing capnp thing :)

15:19 <avsm> many thanks to talex5 for helping with the basic questions. it's all looking quite nice so far

15:19 <avsm> there's a Logs library backend that funnels any Logs usage through capnp to an aggregator

15:19 <avsm> and I'm putting an Irmin backend on it

15:19 <yomimono> I'll make a post to the mailing list or a PR to mirage-www about compiler versions in the next few days, let's move on to capnp in response to demand :)

15:20 <talex5> Lots of new stuff here!

15:20 <talex5> First, it now has TLS support built-in, so you get encryption and authentication by default.

15:20 <talex5> It generates capnp:// URLs which include the server fingerprint, so there's no messing about with CAs.

15:20 <talex5> With that, I've added a TCP transport (it was just Unix-domain sockets before).

15:20 <avsm> cor blimey

15:20 <talex5> The URIs also include a secret "swiss number", both identifying the target service within the vat and granting access to it.

15:20 <talex5> That means we now have working access control :-) Also, you can now export multiple services as sturdy refs (e.g. "admin" and "user" interfaces).

15:20 <talex5> I'm currently working on full persistence support, which will allow creating sturdy refs dynamically, backed by some storage (filesystem, irmin or whatever).

15:21 <talex5> There have also been lots of updates to the docs. And there's a TOC now, so it's easier to find things:

15:21 <talex5> https://github.com/mirage/capnp-rpc/blob/master/README.md

15:21 <avsm> can the access control be expressed in terms of an rpc interface, or is it at the connection level?

15:21 <avsm> i.e. for my command line CI tool, can i do a github auth or something and then allow particular uids to go through

15:21 <talex5> It's capability based - if you have the sturdy ref, you can access it.

15:21 <avsm> i see, that's easy then -- just need an Auth RPC service

15:21 <avsm> and then manufacture a sturdy ref to persist on the CLI

15:22 <talex5> Yes, go the GitHub auth and give the user a capability back (sturdy or otherwise).

15:22 <avsm> how much effort will it be to make the runtime mirage-friendly?

15:22 <avsm> i think core_kernel is the culprit atm

15:22 <talex5> There might be some core_kernel in the capnp runtime.

15:22 <talex5> Might need to remove that, but should be easy.

15:23 <talex5> The RPC split has a separate unix package with the unix-specific bits, as usual.

15:23 <avsm> I also tried the python bindings and they work great after some hassle to connect over a unix domain socket

15:23 <talex5> s/split/library/

15:23 <avsm> but it sounds like tcp transports fix that too

15:23 <talex5> Python stuff is documented too now:

15:23 <talex5> https://github.com/mirage/capnp-rpc#is-there-an-interactive-version-i-can-use-for-debugging

15:24 <avsm> this is awesome :-)

15:25 <avsm> i was thinking about what a ppx would look like to remove the boilerplate in bits of it

15:25 <avsm> but I'm not entirely sure it's needed. Just for things like parsing params and remembering to release the capabilities

15:26 <talex5> Maybe the capnp compiler could spit out some rough templates too?

15:27 <avsm> that's a good idea, particularly for the Service object definitions

15:27 <avsm> editing templates doesn't seem very ML-ish to me, but in this case it's hard to abstract over object definitions

15:27 <talex5> The schema language does support annotations if we need them.

15:27 <talex5> That could direct code generation.

15:28 <avsm> ah yes, and very rich annotations too

15:28 <talex5> I'm not sure how much effort the wrappers really are though (it seems like a lot when writing the tutorial, but it's only a small part of a real system).

15:29 <avsm> I'm hoping to build the web ui for the CI using the capnp js backend, and possibly with reason or jsoo

15:29 <avsm> I'm spending quite a lot of time defining the data model in OCaml, and then the conversion to and from the capnp. But it's a one-off thing

15:29 <avsm> refactoring is kind of painful since so many modules and objects need to be renamed

15:30 <avsm> overall it is a Very Pleasant Experience

15:30 <talex5> :-)

15:31 <talex5> If anyone has any questions or feedback, feel free to post it on https://discuss.ocaml.org/ (tagged capnp).

15:31 <avsm> i'm planning to release the Logs backend separately as it can be used by other projects, so i'll post there

15:32 <avsm> hoping to have a few re-usable components

15:32 <yomimono> awesome :D

15:32 <avsm> i notice capnp has include support, which i assume can be used to compose libraries of rpcs

15:32 <avsm> incidentally, the reason for all this is just how incredible it makes unikernel programming

15:33 <talex5> Yes, includes should work.

15:33 <avsm> since this is the multilanguage substrate for communicating processes

15:33 <avsm> for anyone wondering why capnp is relevant to mirage :-)

15:33 <avsm> i'm sort of wondering what a unix shell based on capnp might look like...

15:33 <avsm> instead of sudo, you have to rpc to something to grab a capability

15:33 <avsm> but anyway...

15:34 <avsm> maybe the opposite of systemd lies down that road though :-)

15:34 <talex5> Well, there was http://www.cs.jhu.edu/~seaborn/plash/html/

15:35 <avsm> thanks, will read. there's also http://shill.seas.harvard.edu which is a language along the same lines that is more recent

15:35 <talex5> Ah, hadn't seen that.

15:36 <talex5> Anyway, that's all from me.

15:36 <avsm> and me.

15:37 <yomimono> And that's it for the agenda :)

15:37 <avsm> see several of you in Oxford at ICFP next week!

15:37 <yomimono> I'll be in Cambridge for a few days after ICFP, if any lurkers want to meet up -- 10th through 12th, drop me a line :)

15:37 <avsm> +1 :-)