Irc discussions from #mirage on 18-05-2016

Written by unpurecamelbot
Classified under: irclog
Published: 2016-05-18 (last updated: 2016-05-18)

18-05-2016 15:01 Drup yeah, I didn't

18-05-2016 15:01 djs55 I think we have a good agenda here:

18-05-2016 15:01 engil indeed unpurecamelbot

18-05-2016 15:01 Drup ca_: you will have to create a new keys yourself

18-05-2016 15:01 Drup (and use it inside your unikernel

18-05-2016 15:02 djs55 shall we start the meeting?

18-05-2016 15:02 GemmaG Yes :D

18-05-2016 15:02 mort___ yes!

18-05-2016 15:02 yomimono I call this meeting of MirageOS conspirators to order

18-05-2016 15:02 yomimono bangs gavel

18-05-2016 15:03 mort___ lead us oh leader!

18-05-2016 15:03 djwillia hello all!

18-05-2016 15:03 mort___ sits down

18-05-2016 15:03 djs55 The first topic appears to be "Reproducible builds" — Hannes' suggestion:

18-05-2016 15:03 djs55 Anyone got any thoughts? :)

18-05-2016 15:03 hannes i didn't suggest anything, was only poked by holger (who has sent a list to the mail, awaiting moderator approval!??)

18-05-2016 15:04 djs55 hehe ok

18-05-2016 15:04 djs55 Who is a list moderator? It's not waiting for me is it?

18-05-2016 15:04 hannes would be nice if someone would try that out 4.03 + reproducible mirageos unikernels and explain why it is the case (i poked xavir to have this in 4.03 out of the box)

18-05-2016 15:04 yomimono I wondered what the set of experiments to "prove" a reproducible build might look like

18-05-2016 15:05 yomimono I had a look at the baserock example which seemed most thoroughly documented

18-05-2016 15:05 hannes yomimono: build xxx once, record checksums, build it again. record checksum. wipe ~/.opam, build again same versions

18-05-2016 15:05 yomimono hannes: change hostname, change date, change locale...

18-05-2016 15:06 hannes yomimono: sure, if you like. and document what you changed and what works :)

18-05-2016 15:06 yomimono none of these should affect our builds, but I'll admit I can't prove it :)

18-05-2016 15:06 talex5 Someone on the Qubes list tried building mirage-firewall twice with 4.02 and complained it had a different hash. Seems to work with 4.03, but didn't try wiping opam.

18-05-2016 15:06 hannes afaict it should work... is the mantis thingy ftr

18-05-2016 15:06 djs55 ok, I've flushed the list moderation queue — sorry about that

18-05-2016 15:08 djs55 ok I see also on the agenda there's a note "MirageOS and QubesOS integration -> ppx_sexp_conv removal -> updated to support 4.03" — this sounds good

18-05-2016 15:09 hannes djs55: there might be more libraries which have accidentally recorded the ppx into runtime dependencies (rather than build deps)

18-05-2016 15:10 djs55 That's a good point — I think I'm probably guilty of that

18-05-2016 15:10 hannes (since you were doing lots of the ppx stuff, you might remember which to check; i did it for mirleft+mine)

18-05-2016 15:10 talex5 Sounds like a build tooling problem with oasis.

18-05-2016 15:10 talex5 Does topkg help here?

18-05-2016 15:10 djs55 I'm sure topkg solves all problems

18-05-2016 15:11 yomimono I don't see a dbuenzli to provide a counterargument

18-05-2016 15:11 hannes talex5: there's builddepends and xmetarequires which you can specify separately

18-05-2016 15:11 djs55 ah xmetarequires

18-05-2016 15:12 djs55 I've just been called away IRL but talex5 has volunteered to help run the rest of the meeting — sorry about that!

18-05-2016 15:12 talex5 I feel there should be some kind of automatic link between the {build} tag in opam and the requires in META.

18-05-2016 15:13 talex5 Anyway, it's working for now. mirage-firewall went back down from 10 MB to 5 MB.

18-05-2016 15:14 mort___ cool :)

18-05-2016 15:14 talex5 OK. Moving on...

18-05-2016 15:14 pris is here or #docker-dev the right place for hyperkit discussion?

18-05-2016 15:14 talex5 Mirage Dashboard and Lwt tutorials. Who added that?

18-05-2016 15:15 yomimono pris - normally here, but we're smack in the middle of our fortnightly meeting. it usually runs for another ~20 minutes or so

18-05-2016 15:15 pris lurks to let meeting continue

18-05-2016 15:15 yomimono thanks pris :)

18-05-2016 15:16 talex5 pris: just checking with a hyperkit developer, who's sitting next to me.

18-05-2016 15:16 talex5 He suggests #docker-dev

18-05-2016 15:17 talex5 Does anyone have anything to say about Lwt tutorials? I think there was an out-of-date one on the mirage site somewhere...

18-05-2016 15:18 yomimono Yes, is still there

18-05-2016 15:18 avsm Hello! That Lwt tutorial does need quite a bit of work.

18-05-2016 15:18 talex5 It still talks about using the campl4 syntax extensions.

18-05-2016 15:18 yomimono among other things it teaches the use of lwt syntax extensions D:

18-05-2016 15:18 mort___ agree syntax extensino should go

18-05-2016 15:18 yomimono Perhaps we should just remove it and link to the ocsigen docs?

18-05-2016 15:19 mort___ but given RWO is all about async, several students this year found it v useful

18-05-2016 15:19 mort___ so i don't think it should be removed completely without a good simple alternative tutorial

18-05-2016 15:19 avsm There's been active discussion about what to do with Lwt and Async as well recently. An Lwt adapter to Async is a distinct possibility... I'm interested in ensuring that Async_kernel is Mirage compatible in the short term so we can experiment with that approach.

18-05-2016 15:19 mort___ (can't recall if such is in ocsigen docs)

18-05-2016 15:19 talex5 I could go through the tutorial and at least make sure it's up-to-date and mentions error handling.

18-05-2016 15:20 mort___ talex5: that'd be great. i really think that losing it without a good replacement is a bad idea

18-05-2016 15:20 talex5 OK. Next item.

18-05-2016 15:20 talex5 Security adv. hannes?

18-05-2016 15:21 hannes well received, we should do that again. (I didn't add that topic)

18-05-2016 15:21 Drup if I may, I (and others) would appreciate if lwt tutorials were contributed to .. lwt :)

18-05-2016 15:21 yomimono +1

18-05-2016 15:21 aantron +1

18-05-2016 15:21 hannes my article ended up in the ocaml weekly news..

18-05-2016 15:22 talex5 Drup: I'll check how specific it is to Mirage.

18-05-2016 15:23 talex5 Do we have any plans for any kind of security audits?

18-05-2016 15:23 hannes talex5: I do continuous ones... would be nice to get yomimono afl stuff up and running... would also be nice to get professionals look over OCaml code, but so far I only met people who can read C and assembly ;)

18-05-2016 15:24 talex5 I suspect that "grep FIXME" in a C stubs would be quite effective.

18-05-2016 15:24 hannes and fixing all the nasty C compiler warnings you mean? ;)

18-05-2016 15:24 yomimono I'd personally be hesitant to call anything short of a professional engagement a "security audit", but maybe a thorough code review wouldn't go amiss

18-05-2016 15:24 yomimono (I should restate: a thorough code review would certainly be a good idea)

18-05-2016 15:25 hannes and in any case, design documents, as discussed last time, should predate this..

18-05-2016 15:25 avsm I would focus on getting the AFL tooling integrated into at least one of the protocols

18-05-2016 15:25 avsm it's very very close to being useful...

18-05-2016 15:25 yomimono it's quite a bit easier with the (unfortunately massive) PR to mirage-tcpip I submitted yesterday

18-05-2016 15:26 talex5 I had a go with AFL on the 9p libraries. Very easy to use and effective.

18-05-2016 15:26 talex5 Is it in opam yet?

18-05-2016 15:26 avsm yomimono: awesome!

18-05-2016 15:26 yomimono is what in opam yet? the AFL branch to 4.02.3?

18-05-2016 15:26 talex5 yomimono: yes

18-05-2016 15:27 hannes (sorry, have to leave in 5 minutes to catch a train; I'd like people thinking about a potential mirage-3.0 (maybe this time with a press release with LF/Xen (amir reminded me of that possibility)... imho having a solution to error handling (wip afaik) and logging is sufficient, plus no camlp4, and 4.03 support.

18-05-2016 15:27 yomimono talex5: I believe so, but am checking to make sure

18-05-2016 15:27 yomimono hannes: good points all, thanks :)

18-05-2016 15:27 talex5 Regarding breaking all the APIs. I really think we need a way to upgrade APIs smoothly.

18-05-2016 15:27 hannes there are various things to be improved in mirage-types: rename V1 to something not depending on a version (versioning is done at the opam level), CLOCK interface improvements, NETIF, ...)

18-05-2016 15:28 hannes talex5: no, please let us use opam for that. there's no need for another SAT solver on the module layer

18-05-2016 15:28 talex5 At the moment, if I want to change the NETWORK sig, I have to update every network implementation in existence at the same time,

18-05-2016 15:28 yomimono talex5: do you think that providing some mechanism for releases of all libraries in lockstep would suffice?

18-05-2016 15:28 hannes talex5: no, conflicts: "mirage-types">3.0.0

18-05-2016 15:30 talex5 What changes have we successfully made to mirage-types in the last 12 months?

18-05-2016 15:30 hannes which one have been successfully proposed? most recently mindy's ICMP

18-05-2016 15:30 yomimono ISTM this might be a discussion for the mailing list

18-05-2016 15:30 talex5 I only see changes to TCP, which has only one consumer anyway.

18-05-2016 15:31 hannes talex5: I don't get your point, sorry

18-05-2016 15:31 hannes (have to leave now)

18-05-2016 15:31 yomimono I agree that this is a problem, but I don't think it's one we'll solve before hannes has to catch a train, and I think it's one we can usefully get more perspectives on.

18-05-2016 15:31 yomimono In particular, there may be some release automation that can help us with this.

18-05-2016 15:31 talex5 Yes, let's move it to the list.

18-05-2016 15:32 avsm just one thought on mirage-types and versioning

18-05-2016 15:32 avsm why dont we have an opam package per module type?

18-05-2016 15:33 avsm that way we can refine them independently. It was just an artefact of packaging at the time that I bundled them together into one OPAM package

18-05-2016 15:33 avsm (thus possibly defeating the entire point of having a dependency-free set of module types)

18-05-2016 15:33 avsm (but yes this is more appropriate for the mailing list :-)

18-05-2016 15:34 talex5 OK. Any other thoughts on Mirage 3.0?

18-05-2016 15:34 yomimono I'd like to make a poll of users of low-level libraries like block and network to see whether folks have requests beyond error reporting and logging

18-05-2016 15:35 yomimono for massive API breakage

18-05-2016 15:35 talex5 Good idea. Are you planning to send something to the list?

18-05-2016 15:35 yomimono If we're going to start talking about mirage-3.0.0 more concretely, yes.

18-05-2016 15:36 avsm It's a good time to do so.

18-05-2016 15:37 yomimono under "mirageos 3.0" we have an item "schedule? release engineer? press release via xenproject/LF?"

18-05-2016 15:38 yomimono in the past we've ended up scheduling around deadlines imposed by the press release, and samoht/avsm have done our release managing

18-05-2016 15:38 yomimono I suspect that they're not too eager to repeat that experience

18-05-2016 15:39 talex5 I thought the plan this time was to sneak all the features into 2.x and then use 3.0 just for PR?

18-05-2016 15:39 yomimono hm, so massive API breakage in 2.10 or something then?

18-05-2016 15:40 talex5 Yes. The press release would talk about "changes since 2.0".

18-05-2016 15:40 avsm This time, I would not impose an artificial deadline around a press release

18-05-2016 15:40 avsm There's simply too much code to quickly rejig everything as we used to do (and too many users of said code!)

18-05-2016 15:41 avsm So planning a more punctuated set of feature changes is better. mirage 3 should only include the absolute core APIs that we want to define and leave as much as possible to the library ecosystem

18-05-2016 15:43 talex5 But the press release will talk about the whole ecosystem, right?

18-05-2016 15:45 yomimono We have to come up with some compelling stuff for a press release to say, and "we broke a bunch of APIs" is probably not super compelling

18-05-2016 15:45 yomimono So I imagine that talex5 is right, and any press we do will center the features that all the API breakage enables.

18-05-2016 15:46 talex5 I guess we should continue this discussion when Amir is around.

18-05-2016 15:46 yomimono That's probably a good idea.

18-05-2016 15:47 talex5 OK, next: samoht linked to:

18-05-2016 15:47 talex5 and

18-05-2016 15:47 talex5 (old GitHub issues and PRs)

18-05-2016 15:48 talex5 Mmm. IPv6 support.

18-05-2016 15:48 yomimono It does say DO NOT MERGE YET

18-05-2016 15:49 avsm It's a guideline not a rule I'm sure

18-05-2016 15:49 avsm nojb did a great job with that PR -- it was just configuration that was blocking it (pre functoria)

18-05-2016 15:49 avsm Also, we also now have actual IPv6 endpoints to test it with on Rackspace

18-05-2016 15:50 yomimono I'm willing to sit down and try to go through a bunch of these, but it'd be nice if someone with complementary expertise to my own has time to do that with me

18-05-2016 15:50 yomimono (i.e. someone who knows about stuff that isn't mirage-tcpip)

18-05-2016 15:51 talex5 Maybe grab a conference room here once you're back in the UK?

18-05-2016 15:51 yomimono or do it over IRC sometime sooner -- seems silly to wait until June for that reason; it's a distributed project :)

18-05-2016 15:52 yomimono If anyone's interested, feel free to contact me here or by e-mail, or just responding to a bunch of old issues :P

18-05-2016 15:53 talex5 Yes, I guess people can just comment on issues and we'll check next meeting that we didn't miss too many.

18-05-2016 15:53 yomimono Unless there are other suggestions, let's move on...

18-05-2016 15:54 talex5 "Future hacking evening"

18-05-2016 15:54 yomimono A+ YES LET'S

18-05-2016 15:54 yomimono GemmaG - can you help us organize this in Cambridge?

18-05-2016 15:54 GemmaG Of course :)

18-05-2016 15:55 yomimono great! I'll take responsibility for working with you to get that going :)

18-05-2016 15:56 talex5 GemmaG, yomimono: thanks!

18-05-2016 15:56 GemmaG Was thinking that sometime in June could be good - while Daniel is visiting

18-05-2016 15:56 avsm I can supply sufficient wine to fuel a Pembroke IPv6 evening

18-05-2016 15:56 yomimono GemmaG: that sounds great. If there are no objections/suggestions/strident yelling, let's continue on

18-05-2016 15:56 yomimono avsm: YES EXCELLENT

18-05-2016 15:56 avsm as long as we dont end up with IPv5

18-05-2016 15:56 talex5 "4.03 flambda testing"

18-05-2016 15:57 talex5 Anyone tried flambda yet?

18-05-2016 15:57 yomimono Not since it was properly released.

18-05-2016 15:58 talex5 OK then. topkg?

18-05-2016 15:59 talex5 I haven't tried it, but it looks pretty nice.

18-05-2016 15:59 yomimono Does anyone have time to look at using it with some of our libraries in the next week or so?

18-05-2016 16:00 avsm It's on my list for cohttp

18-05-2016 16:00 avsm probably on plane back on Friday

18-05-2016 16:00 yomimono Hearing none, maybe it'd be a good idea to bump the mailing list thread that I seem to remember existing.

18-05-2016 16:00 yomimono Oh, cool! thanks avsm :)

18-05-2016 16:00 avsm flambda: I tried it, but it's not at all clear to me if every cmx file needs to be compiled with the flags or not

18-05-2016 16:00 talex5 I suspect people are waiting until it's available without pinning loads of stuff.

18-05-2016 16:00 avsm yes, but a remote should sort that out

18-05-2016 16:01 avsm I'm keen to figure out if a topkg setup would let Cohttp be published as both OCaml and Reason code (details TBD before anyone prods me for more)

18-05-2016 16:01 talex5 I mean, its purpose is to release your software, but any software you release won't work yet ;-)

18-05-2016 16:01 avsm hah, true :-)

18-05-2016 16:02 talex5 OK, final item:

18-05-2016 16:02 talex5 Outreachy

18-05-2016 16:02 yomimono It starts on May 23rd!

18-05-2016 16:02 talex5 wiredsister?

18-05-2016 16:03 yomimono Looks like she might be AFK. wiredsister, do let us know if you need anything before (or after!) the official start date :)

18-05-2016 16:04 avsm wiredsister is in OSCON! Must say hello while here!

18-05-2016 16:04 talex5 Was anyone else involved?

18-05-2016 16:04 yomimono Oh yes! she did a talk for OSCon Cultivate which I heard was fantastic.

18-05-2016 16:04 avsm I'll find her while in Texas and see if anything is needed to start Outreachy

18-05-2016 16:04 avsm There's a significant chance I'll be struck by lightning before that though

18-05-2016 16:04 wiredsister avsm: Hi!! I'm coming to your talk at 11:55

18-05-2016 16:04 avsm hello!!

18-05-2016 16:04 avsm Oooooops talk, almost forgot

18-05-2016 16:04 wiredsister currently sitting in the Go Unikernel's talk

18-05-2016 16:05 wiredsister lol

18-05-2016 16:05 avsm heading over now! Is it Ballroom A?

18-05-2016 16:05 engil Go Unikernel

18-05-2016 16:05 wiredsister let me check for you

18-05-2016 16:06 wiredsister avsm: Your talk is 11:55 in Ballroom A (first floor, I believe)

18-05-2016 16:06 talex5 OK. We've reached the end of the agenda. Any other things to discuss?

18-05-2016 16:06 talex5 OK then, let's close the meeting!

18-05-2016 16:07 wiredsister Hi everybody, I'm so excited to be your lackey. EOM.

18-05-2016 16:07 GemmaG Woop :)