IRC channel logs

<euleritian> https://fosdem.org/2024/schedule/event/fosdem-2024-3095-one-way-forward-finding-a-path-to-what-comes-after-unix/ The speaker in this talk voiced the opinion that Plan 9 is better than Linux and better than the Hurd. Is that so? Would/could guix and/or gobblins work well on/with Plan 9?

<Zarutian_iPad>euleritian: hmm... does that speaker know why Linus came to be? It was because Linus Torvalds wanted a working kernel for the GNU userland stuff that Hurd was meant to be the kernel for. All inspired or whatever by Unix

<Zarutian_iPad>plan 9 took the unix idea of "everything is a file" to whole another level, perhaps that is what the speaker meant

<dthompson>euleritian: I don't see guix/goblins couldn't work on plan 9 if it's unixy enough. I've never used plan 9 or hurd

<dthompson>I'm all for experiments with alternative kernels

<Zarutian_iPad>ACTION is trying to recall if Hurd is a micro kernel setup or something else

<dthompson>but for the sake of practicality everything needs to work on linux too

<Zarutian_iPad>I wouldnt mind seeing seL4 used with a Genode and Plan 9 kind of userland

<dthompson>Zarutian_iPad: yeah it's a microkernel

<Zarutian_iPad>dthompson: hence target the CloudABI / Capsicum subset of posix syscalls

<Zarutian_iPad>shame that context switching on x86 is so fracking expensive

<euleritian>The main message of the speaker was: linux is so huge, it is not manageable anymore. And it's conceptually flawed, important stuff like graphics, network, containers with all their namespaces. In his opinion, microkernels failed, and if I understood the argument correctly, it's because the message passing is too severe of a bottelneck. His suggestion is to use Plan 9, which he says has graphics, network and isolation of

<euleritian>processes built in by design and is much smaller than linux. Plan 9 is too different from posix that todays apps and programs will not work on it, but for that problem he suggests to start a stripped down linux kernel (with no hardware support) inside each process just to run the one application in it.

<euleritian>Anyway, I'm not so sure that Plan 9 is a microkernel after that talk. And wikipedia claims that it's a monolithic kernel type.

<euleritian>Ups! *important stuff like ... has been bolted on as an afterthought (and does not use the filesystem properly).

<Zarutian_iPad>was looking at AmbientTalk again and notices it has something that Goblins lacks, annotations on eventual sends

<Zarutian_iPad>there is @OneWay and @TwoWay which is pretty much like `(<-np)` sendOnly and `(<-)` respectively

<Zarutian_iPad>( E(target).method() and E.sendOnly() in Agoric js btw)

<Zarutian_iPad>but no @Due() that parameterizes a timeout and expiry for the message

<Zarutian_iPad>both for the purpose of not attempting to resend the message after the expiry but also telling the receipient that (in case of normal @TwoWay eventual send) not to bother to process or send an reply after the timeout / expiry

<Zarutian_iPad>now I am mulling if these kind of annotations can solve the current non-expressibilty of the promise pipelining case where the sender does not care about the result of an eventual send other than to refer to it in a subsequent eventual send

<Zarutian_iPad>suchan annotation in AmbientTalk could be@PPO standing for promise pipelining only

<Zarutian_iPad>these annotatiions would/could be handy also for use with PostalRefs and Smart Messages

<Zarutian_iPad>sneek, botsnack.

<sneek>:)

<randy__>Some of this stuff should be over at community.spritely.institute. will get lost here on irc..

<sneek>Zarutian_iPad: Greetings :)

IRC channel logs

2024-02-06.log