bb010g has quit [Quit: Connection closed for inactivity]
<XgF>
eternaleye: 9P just isn't a good protocol for a Unix VFS layer... It really doesn't map well without abominations like 9P2000.L
<XgF>
Ironically, the best network FS protocol for Unix might just be... SMB2!
jadewang has quit [Remote host closed the connection]
<XgF>
(The SMB2 core protocol maps very well to various VFS layers)
isd has quit [Quit: Leaving.]
<asheesh>
: D
funwhilelost has joined #sandstorm
<eternaleye>
XgF: 9p is a much better protocol IMO - all the _features_ of a Unix VFS can be expressed cleanly on 9p, and in a capability architecture many of them shouldn't exist at all anyway (access control should be done via membranes, not ACLs, for example)
<eternaleye>
XgF: I consider the failing to be on the side of the designers of the Unix VFS layer.
<XgF>
eternaleye: The Unix VFS layer is complicated and convoluted and there is nothing we can do to fix it now
<eternaleye>
XgF: It has poor separation of concerns, and promotes too many things into the core when they could instead be done over a simple base.
neynah has joined #sandstorm
<XgF>
(Besides, ACLs and membranes/capabilities are complementary: when you have a tree like an FS, capabilities/membranes don't work)
<eternaleye>
XgF: Strongly disagree - membranes work just fine there; it's what differentiates them from simple interposition in fact
<eternaleye>
XgF: In addition, Cap'n Proto/Sandstorm _isn't dealing with a Unix VFS_ - it's dealing with stuff written against it.
<eternaleye>
XgF: It's _exactly_ an opportunity to fix it.
<XgF>
eternaleye: That makes 9p even more silly!
<eternaleye>
XgF: And finally, it's not such a poor match that Linux can't mount it :P
<XgF>
(Anyway, Membranes are the manifestation of an access control policy - such as an ACL)
<XgF>
Also - how on earth do I cache files accessed over 9p?
<eternaleye>
XgF: That's a fallacy, specifically the "ACLs as rows, capabilities as columns" one
<eternaleye>
XgF: I suggest reading "capability myths demolished"
<XgF>
eternaleye: I *have* read it
<XgF>
ACLs are how you do permissions on some form of global namespace
<XgF>
(a global namespace being intrinsically somewhat anti-capability but also somewhat essential when humans are involved
funwhilelost has quit [Quit: My Mac has gone to sleep. ZZZzzz…]
<XgF>
Consider sandstorm sharing links: They're each essentially a capability to a membrane which puts access control over a resource. The collection of them is essentially an ACL
<eternaleye>
XgF: And caching is hard _period_, at least if multiple clients are involved
<eternaleye>
XgF: In 9p, I'd say to have a cache-control FS mounted as well, which you read and write lease and invalidation information to/from
<eternaleye>
XgF: ACLs are static and stateless. A membrane can in fact be _stateful_ over the client's past behavior.
<fkautz>
mmm, this conversation reminds me of a nacl plugin i need to write :x
<eternaleye>
XgF: Path-dependence is a powerful tool.
<XgF>
eternaleye: Sure, but in that case the code is essentially the ACL entry
<XgF>
And one of the advantages of SMB2 is that it integrates support for caching
<eternaleye>
XgF: No, it's a monad over past behavior that computes a verdict
<fkautz>
eternaleye: is there such thing as a dyanmic/stateful equivalent of an acl?
<eternaleye>
XgF: These have _very_ different expressive power
<eternaleye>
fkautz: Revocation!
<XgF>
It's ridiculous to claim "9P is network transparent!" when the protocol has no answer to the question of "how do I cache things on the client side?"
<eternaleye>
XgF: I'll be sure to tell that to the Linux kernel's caching framework for it.
<fkautz>
one of the purported benefits of 9p is it is supposed to "encourage caching" :x
<fkautz>
i don't know enough about 9p to substantiate that though
<XgF>
v9fs has no support for caching writable mounts
<fkautz>
eternaleye: thanks for the term, i'll look it up
<fkautz>
so is v9fs not cacheable because of an implementation detail then, or because of the design of the protocol?
<XgF>
Because of the design
<eternaleye>
XgF: Also to the people running 9p clustering on Blue Gene using exactly that transparency
<XgF>
eternaleye: sure, probably using massive RDMA networks without caching
<fkautz>
ah, i see... single user caching is fine, but multi user there is no mechanism to report modifications to other users
<eternaleye>
XgF: Also, the protocol is _extensible_. Define a 9p2000.c extension that adds support for claiming a lease on a FID and getting invalidation notifications.
<fkautz>
so something like linux vfs would work fine, but network attached... different story
<XgF>
eternaleye: And now you're turning it into the SMB2 you hate
<eternaleye>
XgF: Er, no.
<eternaleye>
XgF: "get lease on fid" is a simple primitive operation, that if the client never calls has no complicating issues.
<eternaleye>
XgF: SMB is a LARGE protocol
<XgF>
eternaleye: The SMB2 core protocol is not
<XgF>
We are not discussing the entirety of SMB including 30 years of backwards compatibility
gemlog has joined #sandstorm
<eternaleye>
XgF: Mm. I'd need to look at the signatures of the messages to say more, but smb2 is better than it could be.
<gemlog>
How to manage users? My sandstorm is littered with various users from testing and playing around (most named 'gemlog)
<gemlog>
gemlog via github, gmail, email...
<gemlog>
How to have my admin respond to both github and gmail oauth and be the same account? Or delete people?
<eternaleye>
XgF: One thing I'm not fond of is that it requires SPNEGO as its auth framework - I do prefer the fact that 9p leaves auth up to the endpoints
<XgF>
eternaleye: Sure. 9p is first and foremost a local machine protocol for Plan 9; SMB is first and foremost a network protocol
isd has joined #sandstorm
kecolus has quit [Quit: Leaving]
jadewang has joined #sandstorm
ocdtrekkie has quit [Remote host closed the connection]
aldeka has quit [Remote host closed the connection]
aldeka has joined #sandstorm
ocdtrekkie has joined #sandstorm
mnutt has joined #sandstorm
jacksingleton has quit [Ping timeout: 272 seconds]
<eternaleye>
XgF: Not my meaning - One _can_ run SPNEGO over 9p's AFIDs; but one can also do SASL or EAP or...
<eternaleye>
XgF: I'd rather not impose that kind of presort.
<XgF>
You could tunnel SASL or EAP or whatever over SPNEGO/GSSAPI...
<XgF>
i,i SASL-GS2 over SPNEGO/GSSAPI over SASL-GSSAPI over ...
<XgF>
(but SPNEGO and SASL are already meta-authentication frameworks, please lets not have meta-meta-authentication-frameworks)
<zarvox>
scrolling and iframes on mobile safari are a source of pain; I don't think I'd seen issues with mobile chrome before, but if you're on iOS, it's still using the mobile safari engine, right?
<ckocagil>
yep
<ckocagil>
no Blink on iOS :(
gemlog has joined #sandstorm
<gemlog>
Anyone know anything about deleting users or combining login oauths?
<zarvox>
gemlog: dwrensha (who is currently on a plane) is working on a lengthy patchset for multi-identity support
<zarvox>
this would allow an account to assume multiple identities, so for instance you could sign in with either google oauth or github oauth, and access grains shared with either identity
<gemlog>
zarvox: thx very much, that's exactly it!
<gemlog>
I'd also like to nuke a few of my test-selves. How to do that?
<zarvox>
Hmmm, user deletion is not a currently-supported thing. I guess you could manually frob the mongo database, but that seems dangerous
<zarvox>
are you more interested in reclaiming disk space from their grains, or just not seeing them in the users list on the admin page?
<zarvox>
or something else?
<gemlog>
Just tidying up. No big deal. The bigger thing was combining oauth, which is already in work.
<gemlog>
Currently I have 4 gemlogs and an asheesh.
larjona has quit [Remote host closed the connection]
<gemlog>
3 of me and asheesh are unlikely to visit again :-)
<zarvox>
Heh.
<zarvox>
I dunno, asheesh seems like the kind of fellow that would check in in a few months!
<mnutt>
yeah, I had that issue. somehow I didn’t come across that github issue though
<gemlog>
zarvox: dunno if it's your area, but here's a thing from just now. Export from ethercalc. Works for csv and html but not libreoffice. Is that already a known thing or just me?
<zarvox>
it's pretty far toward the end - I was reading through all the old issues and seeing if any of them looked like they were fixed, or would be fixed by current PRs
<gemlog>
I meant excel
<zarvox>
gemlog: can't say I've used ethercalc's export features, but au is the ethercalc author; perhaps she would know more?
larjona has joined #sandstorm
<gemlog>
It's something I've long wanted and it's quite nice to use. I don't know au, I'll try to ask.
<zarvox>
also mnutt might https://github.com/sandstorm-io/sandstorm/pull/1141 make the owncloud mobile apps work? it looks like Sandstorm won't send a 401 unless the user-agent is whitelisted, and the owncloud mobile apps weren't
<mnutt>
yeah, thanks! I’ll give it a try
<zarvox>
gemlog: she idles this IRC channel, so maybe she'll answer if you highlight her :)
<gemlog>
au: is export to excel working as expected?
<gemlog>
I get a 12.5k file that appears blank in libreoffice.
<gemlog>
html and csv are fine.
<larjona>
hi everybody. I have my debian jessie box updated; checkrestart reports 4 processes using old libraries: dash and nodejs, node and mongodb, all of them tied to sandstorm. I suppose I have to wait to sandstorm.io autoupdate and maybe some inside apps also update too?
<larjona>
(my about reports: Version: 0.130)
<gemlog>
larjona: is it actually causing a problem?
<zarvox>
hi larjona, 0.130 is the latest release. Do you know which libraries are still in use? My guess is glibc, since everything else should be provided by the sandstorm bundle...
<zarvox>
Ahh. You can just restart sandstorm and it should use the new files on disk.
<mnutt>
zarvox: regarding the ios scrolling, i’m curious if fixed positioning is needed at all for the sandstorm bar. couldn’t it be set absolute, and have the grain’s iframe also be set absolute? Are there some other edge cases there?
<larjona>
I rebooted the whole machine some minutes ago, to see if they were gone
<zarvox>
larjona: hmmm, that's curious then. It's possible that Sandstorm's self-containerization is confusing checkrestart, particularly if it isn't namespace-aware. All those files that are referenced *should* be coming from the sandstorm bundle under /opt/sandstorm/latest
sasattack-deskto is now known as sasattack
<zarvox>
Maybe checkrestart isn't looking for those files under /proc/pid/root to see if they are indeed visible at that path to that process?
<larjona>
I don't know
<mnutt>
ah, apparently I have forgotten a bunch of stuff about position: fixed and iOS. fixed is definitely needed there.
<larjona>
I'll file a bug on checkrestart, but in any case, I thought that sandstorm was self-updating every day, and getting those debian security patches
<zarvox>
mnutt: tell me more; I wanted the topbar to be sized and part of the content flow. I think the reason we didn't use it had something to do with the mobile menu and certain clipping rules
<zarvox>
sandstorm is definitely self-updating whenever we push a release
<mnutt>
sandstorm scrolling works fine with other apps, and davros works fine outside of sandstorm, it’s gotta be some sort of bad interaction between sandstorm and davros css
<mnutt>
possibly my use of flexbox
<zarvox>
ooooh flexbox
<zarvox>
I also wanted to use flexbox, but it's 1) a decently large commitment and 2) apparently still has a number of quirks in different cases in different browsers
<mnutt>
I’m using a library called ember-paper which is a port of google material design framework, it was great to get started with but as time goes on there are some weird things it does
<zarvox>
classic framework experience
* XgF
always enjoyed Polymer for not getting in the way
<mnutt>
especially one I don’t have an intimate understanding of
<zarvox>
(not ripping on frameworks; there's just essential complexity in many things that frameworks can't completely hide when your needs expand)
<mnutt>
not that one can have an intimate understanding of css, anyway
<XgF>
(hence why I love polymer: shoves every widget's CSS into its' own little box! :D)
<zarvox>
I've had a rather good experience with React in other projects, FWIW
<XgF>
(damnit Firefox, turn on the shadow DOM enabled feature flag, then it'd only be the crap browsers which don't support it :P)
<gemlog>
I'll try audrey via email z, thanks for pointing me.
aldeka has quit [Remote host closed the connection]
ocdtrekkie has quit [Remote host closed the connection]
gemlog has quit [Quit: Konversation terminated!]
neynah has joined #sandstorm
<mnutt>
btw, I (mostly) fixed the ios scrolling issue, via bisecting css removal. `min-height: 100vh;` was what was causing it
<XgF>
whats a vh?
<mnutt>
vertical height
<mnutt>
I think?
<mnutt>
essentially 100% vertical height
<zarvox>
viewport height
<mnutt>
ah, yeah. and you can do some interesting things like 100vh - 20px
<zarvox>
I think we had that because if you don't, then you can see through one app's not-the-whole-page-height box to the one stacked behind it or something :/
<mnutt>
ah
<zarvox>
would love for things to work more cleanly though
<zarvox>
I'm curious why you concluded that the topbar has to be position: fixed on mobile safari - does it just misbehave with absolute- or relative-positioned blocks?
<mnutt>
yeah, the explanation I’ve heard is around mobilesafari zooming, you’re never scrolling an inner container
<mnutt>
in the early days I think it didn’t even support position: fixed;