log☇︎
3700+ entries in 0.037s
asciilifeform: diana_coman: try the local variant. but i expect even across 1 router hop you will see similar picture.
asciilifeform: diana_coman: i.e. there's a tiny reassembly buffer, and if it is occupied while new frags fall in, it drops
asciilifeform: diana_coman: it's consistent with what i know of the braindamaged frag reassembly mechanism on most iron
asciilifeform: diana_coman: do you have a list of the ones that reached the other end ?
asciilifeform: diana_coman: tcpdump -i eth0 udp port YOURPORT -vv -X
asciilifeform: diana_coman: the other thing, entirely possible that yours got fragged and never unfragged, this'll be path-dependent ( as discussed in orig thread )
asciilifeform: diana_coman: what was your random size range ?
asciilifeform: diana_coman: unless i misread the docs -- yes
asciilifeform: diana_coman: iirc max packet that linux will actually send, is 65507 byte
asciilifeform: diana_coman: as always lemme know if you find other sharp edges
asciilifeform: diana_coman: if you make this fix, i'ma backport , otherwise i'ma get to it in the coming days
asciilifeform: diana_coman: you read correctly, i did not apparently touch that item
asciilifeform: diana_coman: in this one, oughta replace with spaces
asciilifeform: diana_coman: np. and lemme know if you catch sumthing i missed.
lobbesbot: diana_coman: Sent 1 hour and 23 minutes ago: <asciilifeform> http://www.loper-os.org/?p=2557 updated with udp_errata.asciilifeform.vpatch/sig ( still old-fashioned v, for nao ) , minor corrections discussed in prev thread
asciilifeform: i'ma move this item to new-type v, eventually ( if diana_coman publishes a new-type regrind of it as part of smg , i'ma sign that and repost , or otherwise later )
asciilifeform: !Q later tell diana_coman http://www.loper-os.org/?p=2557 updated with udp_errata.asciilifeform.vpatch/sig ( still old-fashioned v, for nao ) , minor corrections discussed in prev thread
mod6: thanks hanbot, diana_coman, asciilifeform
PeterL: diana_coman: http://ossasepia.com/2018/02/15/eucrypt-chapter-10-oaep-with-keccak-a-la-tmsr/comment-page-1/#comment-4218
a111: Logged on 2018-09-19 13:14 asciilifeform: diana_coman: the way i'd do the size shuffle, is to use a kindergarten prng (e.g. mersenne) for repeatable sequence, and put ~in~ the packets, a serial num. then simply record what comes out.
asciilifeform: unless diana_coman or mod6 remembers
a111: Logged on 2018-09-19 15:32 asciilifeform: http://btcbase.org/log/2018-09-19#1851642 << btw mod6 , diana_coman , one possible cut of the knot would be if new vtron were to have algo where it tries keccak 1st, then if fails, tries sha512 and ~loudly warns~ ( can be off by default , and enabled on cmdline )
asciilifeform: mircea_popescu: right now asciilifeform , diana_coman , errybody, has buncha vdiff laying around, not only on www but on own disks, and serious problem if gotta stab at it by hand erry time to see which it is. ☟︎
a111: Logged on 2018-09-19 12:34 asciilifeform: diana_coman: this is how we typically do standardizations, yes
nicoleci: diana_coman: i think its what hanbot calls the bathroom :)
nicoleci: diana_coman: lol!
asciilifeform: diana_coman: i'd say this formulation describes bureaucrat, not engineer, but yes i am aware that they've recently hybridized into a strange mule species, bureau-engineer
mircea_popescu: diana_coman right you are indeed. i take it you're also familiar ?
asciilifeform: diana_coman: asciilifeform reads things he dun fully understand, all day, erry day, i dun disagree.
mircea_popescu: diana_coman and necessarily derive no benefit from it.
asciilifeform: http://btcbase.org/log/2018-09-19#1851642 << btw mod6 , diana_coman , one possible cut of the knot would be if new vtron were to have algo where it tries keccak 1st, then if fails, tries sha512 and ~loudly warns~ ( can be off by default , and enabled on cmdline ) ☝︎☟︎
asciilifeform: diana_coman: i like the new v format, but i also very strongly wish to avoid coming to resemble the urbit people, with their 'aah we restarted the universe for the 7th time' ☟︎
asciilifeform: diana_coman: entirely fair point. i simply dun have the new one in my workflow yet, is all
asciilifeform: diana_coman: the way i'd do the size shuffle, is to use a kindergarten prng (e.g. mersenne) for repeatable sequence, and put ~in~ the packets, a serial num. then simply record what comes out. ☟︎
asciilifeform: diana_coman: how did you make the receiver eat unexpected sizes ?
a111: Logged on 2018-09-14 16:02 asciilifeform: the only lang feature that gets in the way of this partitioning is 'generics', but iirc diana_coman did not use these anywhere
asciilifeform: diana_coman: there's a caveat tho
asciilifeform: diana_coman: the 1 thing that still bothers me is that same file extension was kept for new v as for old
asciilifeform: diana_coman: what's going on is that asciilifeform has buncha boxen with old tooling, and also tends to rely on mircea_popescu to proclaim 'let's nao new format'
asciilifeform: diana_coman: i'm satisfied that your keccak worx
asciilifeform: diana_coman: i'm not averse to regrinding , if diana_coman already switched ( i managed to sleep through this , somehow, but yes looks like you're in the new format )
asciilifeform: diana_coman: this is how we typically do standardizations, yes ☟︎
asciilifeform: diana_coman: but to answer the orig q -- i have nuffin against the new format; but not switched yet.
asciilifeform: diana_coman: and if there was a moment in the l0gz where mircea_popescu proclaimed 'hey this is out of beta, errybody plox to retire the classical vtrons' -- i missed it? plox to link.
asciilifeform: diana_coman: believe or not, i still havent switched to phf's .
asciilifeform: iirc most recently diana_coman's
asciilifeform: diana_coman ^
mimisbrunnr: Logged on 2018-09-18 14:08 mircea_popescu: http://btcbase.org/log/2018-09-18#1851125 << incidentally, we very much want to do this. diana_coman mind setting up a large testing harness, send a soup of all packets lengths from 1 to 65536 bytes each hour back and forth for a week or two ?
a111: Logged on 2018-09-18 16:06 diana_coman: anyways, rounding up, it seems my next step here is to 1. set up the testing harness http://logs.bvulpes.com/trilema?d=2018-9-18#431515 2. put asciilifeform's lib to use in smg.comms
asciilifeform: diana_coman: last (for nao) observation -- 1) it is possible to make the thing 'fancier' in 2 ways -- can make Socket a 'controlled type' ( as i did in mmap, see http://btcbase.org/log/2018-09-14#1850368 ) , then it can close itself when going out of scope. i did not do this, as it adds a bit of overhead 2) it is possible to make the lib a 'generic' ( again see horsecocks re how ) , and make udptrons of different packet length runti ☝︎
asciilifeform: diana_coman, mircea_popescu : nobody noticed, but it is troo -- i forgot to close the socket in the demo ( this has 0 effect, os closes ). but in next rev will correct this.
mimisbrunnr: Logged on 2018-09-18 14:08 mircea_popescu: http://btcbase.org/log/2018-09-18#1851125 << incidentally, we very much want to do this. diana_coman mind setting up a large testing harness, send a soup of all packets lengths from 1 to 65536 bytes each hour back and forth for a week or two ?
asciilifeform: diana_coman: somewhere i also have glue for unix signals support, so proggy can do the Right Thing when you ctrl-c, or kill, etc. but this i'll dust off later (or if somebody has a dire need)
asciilifeform: diana_coman: there'll be no getting away from inline asm once we start planting things on naked iron.
asciilifeform: diana_coman, mircea_popescu : i thought about including timeout in 'procedure Receive...' but sat and thought and could not think of why , so omitted.
asciilifeform: diana_coman: libc, specifically ( i like musl, but it doesn't belong on the fyootoor all-adatronic box )
mircea_popescu: diana_coman i can see it.
asciilifeform: diana_coman: i'm not married to the c glue, and it'll eventually go. i am quite fond of my api tho, it completely rids user of having to think in unixisms , imho
asciilifeform: diana_coman: at the very minimum, survival rate, and latency. ( could also try to measure reorderiness, but immediately obvious how )
mimisbrunnr: Logged on 2018-09-18 14:08 mircea_popescu: http://btcbase.org/log/2018-09-18#1851125 << incidentally, we very much want to do this. diana_coman mind setting up a large testing harness, send a soup of all packets lengths from 1 to 65536 bytes each hour back and forth for a week or two ?
a111: Logged on 2018-09-18 14:33 mircea_popescu: http://btcbase.org/log/2018-09-18#1851125 << incidentally, we very much want to do this. diana_coman mind setting up a large testing harness, send a soup of all packets lengths from 1 to 65536 bytes each hour back and forth for a week or two ?
a111: Logged on 2018-09-18 13:46 diana_coman: it's not fully clear to me if it's something needed /desired atm; at any rate, compared to where I was 2 days ago, it's great - all of a sudden it went from "need to do this from scratch, ugh" to "there are 2 republican libs with 2 approaches, which one fits best my needs?" ; I'm rather delighted to be honest
mircea_popescu: http://btcbase.org/log/2018-09-18#1851125 << incidentally, we very much want to do this. diana_coman mind setting up a large testing harness, send a soup of all packets lengths from 1 to 65536 bytes each hour back and forth for a week or two ? ☝︎☟︎
a111: Logged on 2018-09-18 13:17 diana_coman: as I was saying earlier: atm the fixed packet length might clash a bit with what I need but it's not even fully clear it's not *better* to have a fixed packet length anyway
a111: Logged on 2018-09-18 13:05 diana_coman: for completeness, version 3. GNAT.Sockets.Thin that is an Ada wrapper on C system calls containing however questionable approaches (e.g. returning access to String so effectively a pointer but worse than this: allocating memory on the heap and leaving the de-alloc to the caller...)
mircea_popescu: http://btcbase.org/log/2018-09-18#1851102 << this is such, SUCH terribly bad practice, srsly now. scope the fucking memory handling, so both ends always happen in the same context. either the caller allocates and then also deallocates, diana_coman style, or else the callee allocates and deallocates. none of this insane scope bridging jesus christ. ☝︎
mircea_popescu: diana_coman: to round this whole thing up: 2 days ago it seemed I had only the gnat.sockets/ thin layer option which wasn't fit for purpose; now I have 2 more options: << it's been an epic few days! (what happened ?)
asciilifeform: diana_coman: lemme know if you think of further 'why did he do that'-s.
asciilifeform: diana_coman: virtually all nontrivial programs eat or display an ip in txt form somewhere, i've found. incl the demo. hence, included.
asciilifeform: diana_coman: i already nailed down a format, observe, ip is stored always as native-endian 32bit.
asciilifeform: diana_coman: imho that functionality belongs in a udp lib ( given as it demands knowledge of how ip is represented ) but prolly oughta be a troo ada thing, not a callout. however implementing it would double the mass of the proggy
asciilifeform: diana_coman: i handle all eggogs
asciilifeform: diana_coman: imho variable-length-datagram is for the birds. hence i jettisoned good bit of complexity by omitting it
asciilifeform: diana_coman: anyffing useful and conspicuously missing ?
asciilifeform: diana_coman: observe, my lib is somewhat unorthodox, it tries to abstract entirely over the os socket, nails down a fixed packet length, ipv4 4evah, no 'nonblockisms' - who wants these, can implement via ada task; no dnsism supported at all, etc
asciilifeform: diana_coman: afaik the only useful option re udp avail on linux is reuseaddr
mimisbrunnr: Logged on 2018-09-18 12:37 diana_coman: to round this whole thing up: 2 days ago it seemed I had only the gnat.sockets/ thin layer option which wasn't fit for purpose; now I have 2 more options: 1. ave1's ADA implementation of UDP sockets using directly ASM inline 2. asciilifeform's light UDP sockets lib that uses C code for needed UDP sockets calls but provides an Ada wrapper so that any code using the lib can call Ada methods
lobbesbot: diana_coman: Sent 3 hours and 51 minutes ago: <asciilifeform> http://www.loper-os.org/?p=2557 .
a111: Logged on 2018-09-17 20:44 diana_coman: the question+kick&ban sounds good to me - kicking "silent" aka "I'm part of it because I hang about in here doing nothing" is even needed by now, I'd say; I can also see very well its usefulness for other channels; while atm #eulora tolerates the allah-spam, it could certainly do without it especially at less-quiet times
asciilifeform: !Q later tell diana_coman http://www.loper-os.org/?p=2557 .
a111: Logged on 2018-09-17 12:41 diana_coman: I can just about see next version of linux-with-feelings
mircea_popescu: anyway -- fuck 'em, i deeply fucking care some nobody-on-a-stick meanwhile joined the churchgoing group. whatever. but in matters of actual interest : you "were working" on an ada networking socking thing ; ave "was working" on ~same. had diana_coman not said "here's what ~~~I AM ABOUT TO DO~~~", we would have never known. if the wastefulness of this approach isn't directly obvious...
a111: Logged on 2018-09-17 11:42 ave1: diana_coman, code is here; http://ave1.org/code/suckit/, everything is pretty much in flux still (even then name could change, currently ip addresses are in network order but ports are in native order)
a111: Logged on 2018-09-17 09:06 diana_coman: arguably it helps if others know what you are working on - perhaps one can help at some point or at least not duplicate the effort, that's all
a111: Logged on 2018-09-17 12:41 diana_coman: sounds quite likely; and tbh that http://btcbase.org/log/2018-09-17#1850713 makes for weird/sad/hysterical reading depending on pov
mircea_popescu: diana_coman i am not reading on. i'm done with torvalds.
a111: Logged on 2018-09-17 09:45 diana_coman: lobbes, esthlos, hanbot, ave1 - do you have something against using categories on your blogs? I find well-defined categories to be really helpful in finding stuff but on your blogs everything is in the "Uncategorized" category
ave1: diana_coman, code is here; http://ave1.org/code/suckit/, everything is pretty much in flux still (even then name could change, currently ip addresses are in network order but ports are in native order) ☟︎
asciilifeform: then diana_coman may find it useful, i'ma hurry up & genesis ( realistically tomorrow posted. )
asciilifeform: diana_coman and mircea_popescu can read and use if they like it, i'ma genesis
a111: Logged on 2018-09-16 15:41 diana_coman: overall this leaves me with the fact that I'll have to make my own "thin layer" with only what I need, certainly thinner than gnat.sockets + gnat.sockets.thin
mircea_popescu: diana_coman but would the wrapper be c or ada ?
mircea_popescu: diana_coman would this be written in ada then ?
a111: Logged on 2018-09-14 16:48 diana_coman: I certainly have to consider that option too if all that gnat.sockets buys me is some bloat + streams
BingoBoingo: diana_coman: fxd, tyvm
a111: Logged on 2018-09-14 21:25 asciilifeform: diana_coman: found it, http://ossasepia.com/2018/04/17/rfc-euloras-communication-protocol-eucomms/ , aite
asciilifeform: i'ma stfu at this point, as i promised to mircea_popescu not to interfere in euloric matters. will answer q strictly re gnatology, if diana_coman runs into any puzzlers.
asciilifeform: diana_coman: found it, http://ossasepia.com/2018/04/17/rfc-euloras-communication-protocol-eucomms/ , aite ☟︎
asciilifeform: diana_coman: asciilifeform forgot that you already had the protocol. then you do not need any of this, indeed.
asciilifeform: diana_coman: if one receives an old (i.e. not currently expected ) packet ( e.g. straggler from a previous round ) it gets rejected in the same way as any other rubbish.