tmsr - Log 2018-09-27

asciilifeform: !Q later tell phf are http://barksinthewind.com/2018/vtools-keccak-regrind/ old-style or new-style vpatches ?? my vtron won't press'em, and there is no way to distinguish , nor anything in the post to indicate, unless i'm thick

lobbesbot: asciilifeform: The operation succeeded.

asciilifeform: !Q later tell phf nm, distinguished by hand... but my vtron doesn't verify the sigs (not immediately sure why) and mod6's -- sees only Leaf: vtools_vpatch_newline.vpatch (phf)

lobbesbot: asciilifeform: The operation succeeded.

asciilifeform: diana_coman, mircea_popescu , et al -- plox to reveal what you pressed this tree with , thx

mod6: anyway, it's been a while, so I don't know for sure, but if memory serves, I believe his vtree is incompatible with my vtron, on the whole.

asciilifeform: http://nosuchlabs.com/pub/v_noworky.tar.gz << the complete tarball with both variants, patches, seals, .wot, my attempt thus far. ☟︎

mod6: one has to, again, if i remember correctly, remove some of the vpatches from the "right" side. or something to that effect, to press successfull.y

mod6: im not positive, although, I think another troublesome one was hanbot's wp-mp, which BingoBoingo used esthlos' to press. i think.

asciilifeform: maybe i'm thick, and didn't shuffle correctly, or wat. phf can i persuade you to give a working recipe ??

mod6: asciilifeform: yeah, i dunno, it's been a while. i don't have time to tinker with it at the moment, gearing up for month end here.

asciilifeform: mod6: i sat down and resolved to give diana_coman the promised item, but presently loox like won't happen tonight

asciilifeform: this thing is supposedly battlefield-ready, but presently loox to me like a ball of yarn for which no working build instruction is posted anywhere

a111: Logged on 2018-09-26 18:46 diana_coman: asciilifeform, udp tester seems to be nicely running uk->uy and uy->uk so my next step on this will be to publish all the relevant code; since this is effectively on top of udplib, I'd patch it on your tree; mind moving it though to keccak?

mod6: aha, ok. in the past, i've alwasys built the sha256 stuff for my own purposes. let's see if i can get the keccak stuff build.

asciilifeform: mircea_popescu & diana_coman specifically asked asciilifeform to rebake udp proggy in keccak.

asciilifeform: BingoBoingo: if i'm not mistaken, you built and used this thing? with what didja do it ?

asciilifeform: loox like the matter may have to wait until the folx with working keccaktrons wake up and tell me what i'm missing.

mod6: I took your tarball, dumped in the latest version of my vtron, dropped in the .wot (phf) and seals from what I had in my sandbox previously.

mod6: Alright, I guess the makefile doesn't build the vpatch binary by default, so I built it manually:

mod6: lol, I don't know how to make it do stuff either, but I guess that's a different problem :]

esthlos: http://btcbase.org/log/2018-09-27#1854912 << I plan to integrate Keccak into the thing this weekend. At that point it should be fully operational. ☝︎☟︎☟︎

a111: Logged on 2018-09-27 02:03 trinque: does it now properly verify hashes?

trinque: obviously I want such a vtron for the cuntoo final cut, or what's the use of the build process producing a vpatch ☟︎

BingoBoingo: <trinque> does it now properly verify hashes? << I am still at the point where I hand read input, patch, and output

diana_coman: http://btcbase.org/log/2018-09-27#1854882 -> got your .tar.gz but I could not find in it the .wot and .seals? at any rate, if I copied over the .wot and .seals dirs, it pressed perfectly fine with v here (9999 K version); ftr I ran also precisely the v.pl you have in there and it also worked! ☝︎☟︎☟︎

a111: Logged on 2018-09-27 01:34 asciilifeform: http://nosuchlabs.com/pub/v_noworky.tar.gz << the complete tarball with both variants, patches, seals, .wot, my attempt thus far.

diana_coman: re using vtools, I currently first check sig (so by hand, separate 1-line) and if ok, then feed patch to vtools

a111: Logged on 2018-09-27 02:27 esthlos: http://btcbase.org/log/2018-09-27#1854912 << I plan to integrate Keccak into the thing this weekend. At that point it should be fully operational.

diana_coman: http://barksinthewind.com/2018/vtools-keccak-regrind/ -> gotta ask here, phf, am I missing something or what Wednesday was that there in the first line meant to be? ☟︎

a111: Logged on 2018-09-27 02:28 trinque: obviously I want such a vtron for the cuntoo final cut, or what's the use of the build process producing a vpatch

mircea_popescu: asciilifeform phf's thing presses fine, since at least june ( http://thewhet.net/2018/06/mp-wp-genesis-regrind/ )

diana_coman: adding to the above: old v can still be used to check the sigs, to avoid manual check; so use old v to check the sigs and once satisfied, just feed patches to vpatch

diana_coman: since I need to get the work done on this, I reground the UDP lib and I'll proceed from there; asciilifeform, phf and anyone else interested, keccak-patches are on my Code Shelf as usual: http://ossasepia.com/reference-code-shelf/#selection-477.0-477.19 ☟︎

asciilifeform: http://btcbase.org/log/2018-09-27#1854880 , http://btcbase.org/log/2018-09-27#1854913 << interestingly, this worked, mod6 ! wtf is 99994 K one ? where did i even get it ? ( i take it , was buggy release ? ) ☝︎☝︎☟︎

a111: Logged on 2018-09-27 01:33 mod6: http://thebitcoin.foundation/v/V-20180222.tar.gz << >> http://thebitcoin.foundation/v/V-20180222.tar.gz.mod6.sig

a111: Logged on 2018-09-27 02:06 mod6: asciilifeform: http://p.bvulpes.com/pastes/EItV0/?raw=true

asciilifeform: http://btcbase.org/log/2018-09-27#1854926 << confirmed ( and somehow nobody last night noticed ? ) apparently tar by default refuses to tar up 'hidden' ( i.e. starts with . ) dirs ! ☝︎

a111: Logged on 2018-09-27 09:31 diana_coman: http://btcbase.org/log/2018-09-27#1854882 -> got your .tar.gz but I could not find in it the .wot and .seals? at any rate, if I copied over the .wot and .seals dirs, it pressed perfectly fine with v here (9999 K version); ftr I ran also precisely the v.pl you have in there and it also worked!

asciilifeform: http://btcbase.org/log/2018-09-27#1854926 << ~this~ is pretty strange; diana_coman wouldja mind sharing your working set there ? ☝︎

a111: Logged on 2018-09-27 09:41 mircea_popescu: did you follow diana_coman 's http://barksinthewind.com/2018/vtools-keccak-regrind/#comment-27 ?

asciilifeform: and apparently this is not a full vtron, but only replaces 'diff' and 'patch' ....

asciilifeform: aaand it loox like diana_coman already did my chore for me... i'ma do the elementary test on it nao, and sign/mirror.

diana_coman: asciilifeform, np; re vtron yes, currently it is only vdiff and vpatch functionality; I use old v to see the flow (since it checks the seals but doesn't care about the hashes until press time) and then the vpatch to actually press; looking forward to esthlos' vtron ☟︎☟︎

diana_coman: I added the manifest + comments in it; otherwise *all* code is precisely what I got from pressing your patches

asciilifeform: diana_coman: http://www.loper-os.org/?p=2557 updated, with mirror and seals (yours, mine)

diana_coman: asciilifeform, ok, I'll mirror your sigs too and link to the page anyway; a bit later today

asciilifeform: diana_coman: is there a version of mod6's vtron that uses the new vdiff / vpatch ? or what do you use in erryday work ?

a111: Logged on 2018-09-27 13:50 diana_coman: asciilifeform, np; re vtron yes, currently it is only vdiff and vpatch functionality; I use old v to see the flow (since it checks the seals but doesn't care about the hashes until press time) and then the vpatch to actually press; looking forward to esthlos' vtron

asciilifeform: this is what i thought, and why i did not hurry to start the regrindings earlier

diana_coman: I kept waiting on phf and esthlos since they were working on it as far as I could tell; and http://btcbase.org/log/2018-09-27#1854921 ☝︎

a111: Logged on 2018-09-27 02:27 esthlos: http://btcbase.org/log/2018-09-27#1854912 << I plan to integrate Keccak into the thing this weekend. At that point it should be fully operational.

asciilifeform: if phf's item came with a standalone keccak hasher , would then be quite simple to retool e.g. mod6's vtron, to follow the new format. but as it is loox like i'ma have to wait for esthlos , to get full working replacement for old vtrons

a111: Logged on 2018-09-26 22:00 asciilifeform: mircea_popescu: prolly like most folx who actually work on proggies, asciilifeform has '9000' vtrees on various disk, on various boxen, that are in classical format, and many not even intended for publication, the ones that see daylight naturally will become newtype

a111: Logged on 2018-09-26 22:01 asciilifeform: unlike some folx (as late as '16 some confessed to git ) asciilifeform does not use heathen versiontrons internally, strictly 100% v from aug '15

a111: Logged on 2018-09-27 13:26 asciilifeform: http://btcbase.org/log/2018-09-27#1854880 , http://btcbase.org/log/2018-09-27#1854913 << interestingly, this worked, mod6 ! wtf is 99994 K one ? where did i even get it ? ( i take it , was buggy release ? )

phf: asciilifeform: this has been extensively discussed in the logs. there was never a need for "new style v". v works just fine. one of the design goals of vtools was to nail down (and improve upon) the patch format. vtools are explicitly designed to be used with an existing v (for example i use it with v.py). vdiff produces wellformed vpatches and vpatch presses them ensuring that the format is valid and that the hashes stand. there's no ☟︎

phf: need for the separate keccak hashing tool, because vpatch ensures that the result is valid.

lobbesbot: phf: Sent 14 hours and 10 minutes ago: <asciilifeform> are http://barksinthewind.com/2018/vtools-keccak-regrind/ old-style or new-style vpatches ?? my vtron won't press'em, and there is no way to distinguish , nor anything in the post to indicate, unless i'm thick

lobbesbot: phf: Sent 14 hours and 6 minutes ago: <asciilifeform> nm, distinguished by hand... but my vtron doesn't verify the sigs (not immediately sure why) and mod6's -- sees only Leaf: vtools_vpatch_newline.vpatch (phf)

phf: asciilifeform: but if you want a full "v replacement and i don't want to think about none of that" then just use esthlos's item. i believe he has a working keccak already ☟︎

asciilifeform: phf: try an' see from my pov : i get a 'goddamn stop using old v' , go an' uncrate the replacement, and turns out it ain't a plug-in replacement but a set of pieces and a roll of duct tape

asciilifeform: and yes it is the picture i got from log. the gap in my head is where diana_coman switched to the new format; i then assumed there is a 100%-complete new vtron, and that 'hmm simply missed this, is in log somewhere' , turns out nope, notyet

asciilifeform: apparently diana_coman has been hand-cranking it, sorta like asciilifeform's 1st yr of trb pre-vtron

asciilifeform: ( recall, i introduced the patches thing considerably prior to ~vtronic~ patches )

asciilifeform: possibly this is not 100% fair picture , http://btcbase.org/log/2018-09-27#1854956 is only half-handcranked, but still, semi-automatic ☝︎

asciilifeform: phf: this is not a criticism of the format. but turns out asciilifeform has been looking for a 'i missed it in logz?' piece that doesn't exist quite yet ( hopefully will in 3 days... )

asciilifeform: once it ~does~ exist, and fully displaces the duct tape, then yes i'ma start regrinding other things , and i expect then mod6 -- trb, etc

asciilifeform: afaik nobody's used esthlos's item in battlefield yet, it'll need good bit of test.

asciilifeform: i'd also like to see a continuing existence of multiple working vtrons, so eventually would like to rework mod6's to use newtype format ( unless mod6 would prefer to do it himself )

phf: asciilifeform: you're just spreading fud, i don't know where to start unpacking this conversation ☟︎

asciilifeform: my ancient v100 , could also be reworked, but i'd like to put it to rest ( for one thing, it dun check hashes at all )

asciilifeform: this thread is not meant to insult phf et al , but i open box where i thought was hammer, but there is only a hammer head

phf: when i started working on vtools there was already a handful of battle tested V implementations, that relied on vdiff.sh/gnu patch combination. vtools is a "drop in" replacement for the later. you get valid vpatch on the input, valid press on the output. the purpose is to nail down the format, and gradually replace out its parts (e.g. transparent sha->keccak transition)

mod6: I had started on a ada vtron last year, but I got hung up with some of the string handling, and the fact that I had to use shell-outs for pgp. I'd like to get back to it at some point. I would love to dispense with the shell outs - and can probably do so, but not until 'peh' is finished.

asciilifeform: phf: makes sense. my frustration is from walking the logs for past 2 days and looking for who has the complete shaft+head hammer, and not finding

mod6: The string handling, discussed previously in the logs, is basically a solved problem - would just need something similar to what alf or others have done before - character by character.

phf: i think v.pl is a venerable tool, it's battle tested, it has established interface, it's been worked on for three years now. i don't see any reason to throw it out. ☟︎

asciilifeform: phf: right . what i was looking for is variant of same that calls out to keccak instead of sha512 ( mod6's latest vtron actually checks hashes ) . but apparently still needs baking.

phf: i guess it's presumptuous on my part to think that it's exactly obvious how to take vtools and plug it directly into an existing v's, but that's all that's needed ☟︎

asciilifeform: phf: i don't dispute that it is obvious, but thought it had been done, this was my mistake.

phf: asciilifeform: well, you're one of the v maintainers, i'd think you would want to do it to your own tool.

asciilifeform: phf: my orig vtron has been sad and unmaintained, and afaik unused by anybody by me, for long time

phf: re standalone keccak hasher: i'm not sure that it's needed, i think the relevant phase can just be dispensed with altogether. `vpatch' verifies the hashes as it goes along

asciilifeform: then not direly needed. though i'd still like to be able to post keccak hash in the log next time we're doing a thread with fw images or the like.

phf: you can basically do (cat foo.vpatch bar.vpatch qux.vpatch) | vpatch and expect the resulting press to be fully valid, hashes and all ☟︎

asciilifeform: ( i.e. i'd like to dispense with sha512sum util entirely , aside from existing ancient items which have sha512 hashes in the l0gz from years ago )

asciilifeform: btw phf , if you're willing to post your modified v.py ( i.e. http://btcbase.org/log/2018-09-27#1854990 ) i'ma sign ☝︎

a111: Logged on 2018-09-27 15:33 phf: asciilifeform: this has been extensively discussed in the logs. there was never a need for "new style v". v works just fine. one of the design goals of vtools was to nail down (and improve upon) the patch format. vtools are explicitly designed to be used with an existing v (for example i use it with v.py). vdiff produces wellformed vpatches and vpatch presses them ensuring that the format is valid and that the hashes stand. there's no

asciilifeform: imho the situation where 'errybody made own hack' but no one posted 'because obvious' is a barbarism, really ought to have a civilized 'here is the whole thing' sitting on www somewhere. ☟︎

trinque: phf: did the problem which caused folks to have to delete one branch of your vpatch tree to press the other get fixed in v.pl?

asciilifeform: prolly will also makes sense to standardize the calling for it eventually ( and i'ma rework mine to fit, etc )

phf: trinque: it hasn't been, and i'm doing omission by not mentioning that vtools had a later mandate for graph resolution. meanwhile esthlos v was supposed to supposed to fix the issue, i suspect that different v's will eventually catch up.

phf: trinque: vtools has not been design or intended to compete with any particular v implementation. it's a set of tools that you can use in a v workflow (hence the name). at least initially it was two matching tools vdiff and vpatch that know how to produce and consume a canonical vpatch. the conflation came up, because i also published vtools in a form that broke existing canonical v, v.pl, and was tasked with fixing the situation. ☟︎

asciilifeform: phf: i finally sat down and properly read the thing, and i think i finally grasp.

phf: re esthlos's work i think it's a shame that he chose to reimplement own keccak, but is still calling out to gnu patch. he could just focus on graph resolution/signature/wot checking, and offload the validation on vpatch: construct the press list, feed the patches in-order to vpatch, if vpatch succeeds then you know _for certain_ that the sequence of patches is valid, hashes and all

mod6: yeah, this is the right approach imho. it's what i'd do with my ada-vtron, if it ever does breathe.

trinque: phf: in principle I'm entirely in favor of the "unix philosophy" approach of simple, single-purpose tools

asciilifeform: phf: imho your approach , i.e. dispensing with gnupatch, is The Right Thing, historically there was quite a bit of grief from gnupatch's habit of eagerly attempting to apply an invalid (by vtronic lights) but 'partially ok' by barbarian lights patch ☟︎

asciilifeform: trinque: recall when i thought that this nonsense could be disabled by flags, hah

phf: right vpatch applies 'diff -ruN' style patches exactly. it also keeps track of both the hash of the patched file and the hash of the result state as it's reading/patching (there's no double read happening), and errors out if either fails to match the hashes in the header

phf: asciilifeform: i have a ksum PoC for you, http://btcbase.org/patches/vtools_ksum signed but it's from workbench, potentially buggy. "ksum foo bar qux" gives you shasum style <hash> foo\n<hash> bar\n<hash> qux\n

deedbot: http://bimbo.club/?p=31 << Bimbo.Club - TMSR Log Summary - 9/22/2018

BingoBoingo: asciilifeform: I have returned from the envirorast office, about to fire off a message to DHL informing them to try again as I am a provisionally acredited importer of packaged goods for commercial use ☟︎☟︎

a111: Logged on 2018-09-27 12:47 diana_coman: since I need to get the work done on this, I reground the UDP lib and I'll proceed from there; asciilifeform, phf and anyone else interested, keccak-patches are on my Code Shelf as usual: http://ossasepia.com/reference-code-shelf/#selection-477.0-477.19

mircea_popescu: http://btcbase.org/log/2018-09-27#1854952 << what you apparently did was completely ignore the matter for five months, then discover like children that you actually need tools at the time you started on the task (late at night etc) and so forth. ☝︎☟︎☟︎

a111: Logged on 2018-09-27 13:35 asciilifeform: call me an idjit, but i thought there were a new vtron...

mircea_popescu: keep your tools in state of good repair, you won't have to start fixing fence by fixing hammer and nails.

mircea_popescu: in any case shit in the logs ain't gonna to "just go away", this isn't linuslands. ignoring it is like ignoring hot coals. ☟︎

deedbot: http://trilema.com/2018/fetlife-the-next-derperation/ << Trilema - Fetlife -- The Next Derperation

mircea_popescu: http://btcbase.org/log/2018-09-27#1854988 << we came up with this clever thing sometime in 2015 or so iirc. not sure what is gained by doing 99995 --> 99994 etc in light of experience, but i clearly recall how cool it looked at the time. ☝︎☟︎

a111: Logged on 2018-09-27 15:12 asciilifeform: ( fughot that they decrement ! )

asciilifeform: http://btcbase.org/log/2018-09-27#1855080 << i stole the notion , in slightly modified form, from knuth. the logic is, decrementing versions, with finite initial N, support a http://btcbase.org/log/2018-09-18#1851362 flow , where 'if you fucked it N+1 times , nao forced to call proggy something else' -- specifically in opposition to the heathen plague of runaway ver nums (e.g. linux kernel, emacs, firefox, etc) ☝︎☝︎

a111: Logged on 2018-09-27 20:00 mircea_popescu: http://btcbase.org/log/2018-09-27#1854988 << we came up with this clever thing sometime in 2015 or so iirc. not sure what is gained by doing 99995 --> 99994 etc in light of experience, but i clearly recall how cool it looked at the time.

a111: Logged on 2018-09-18 18:22 asciilifeform: this is the other thing, 'changes are expensive' promote imho a sane view of software, where you actually try to perma-stabilize yer proggy, rather than to keep up the classic 'open sores' eternal cauldron of bubbling liquishit

asciilifeform: http://btcbase.org/log/2018-09-27#1855075 << yea if i'd smoke-tested it earlier, would have found. on top of this, naively assumed that diana_coman has a working and complete keccaktronic v , given as she's moved smg to newform ☝︎

a111: Logged on 2018-09-27 19:52 mircea_popescu: http://btcbase.org/log/2018-09-27#1854952 << what you apparently did was completely ignore the matter for five months, then discover like children that you actually need tools at the time you started on the task (late at night etc) and so forth.

asciilifeform: http://btcbase.org/log/2018-09-27#1855070 << neato BingoBoingo . plz lemme know asap if you need anyffing from my end ☝︎

a111: Logged on 2018-09-27 19:17 BingoBoingo: asciilifeform: I have returned from the envirorast office, about to fire off a message to DHL informing them to try again as I am a provisionally acredited importer of packaged goods for commercial use

asciilifeform: http://btcbase.org/log/2018-09-27#1855078 << trickier item. recall 'the song of the sirens, some have escaped, but their silence -- no one' . how to deal with 'but ~was~ it in the l0gz'. ☝︎☟︎☟︎

a111: Logged on 2018-09-27 19:52 mircea_popescu: in any case shit in the logs ain't gonna to "just go away", this isn't linuslands. ignoring it is like ignoring hot coals.

asciilifeform: ( asciilifeform plowed through the phf-vtronics thread when came back from voyage, and then second time when diana_coman requested keccak regrind, and both times failed to converge to correct answ re 'is there complete keccak vtron' )

deedbot: http://ossasepia.com/2018/09/27/tester-for-udp-communications/ << Ossasepia - Tester for UDP Communications

diana_coman: asciilifeform, ^ there is also a small .vpatch for that issue with null chars at ip

diana_coman: the rest then are for the tester only - let me know if you give it a spin and how it behaves

diana_coman: but otherwise the udp_tester.vpatch makes some changes to udp lib that are really just for testing (i.e. I think they should not be part of production use of udp lib)

diana_coman: it's the move to generic + relaxation of some constraints (because of generic and because of using Calendar for local time to log)

deedbot: http://qntra.net/2018/09/power-rangers-inserted-inflation-bug-into-core-bitcoin-network-client-in-2016/ << Qntra - Power Rangers Inserted Inflation Bug Into "Core" Bitcoin Network Client in 2016 ☟︎

mircea_popescu: asciilifeform no, i understand the principle, just... nothing, not even mozilla went through 10k versions to date.

asciilifeform: mircea_popescu: yea, 10k is imho much. but e.g. asciilifeform gave himself 100 in orig v

mircea_popescu: http://btcbase.org/log/2018-09-27#1855089 << i have no fucking idea how. i read the logs daily atm, mostly impelled by... outright fear. the best heuristic i know of, but otherwise this promises to be a first caliber bane as time goes by. ☝︎☟︎

a111: Logged on 2018-09-27 20:21 asciilifeform: http://btcbase.org/log/2018-09-27#1855078 << trickier item. recall 'the song of the sirens, some have escaped, but their silence -- no one' . how to deal with 'but ~was~ it in the l0gz'.

deedbot: Get your OTP: http://p.bvulpes.com/pastes/O1xwh/?raw=true

asciilifeform: http://btcbase.org/log/2018-09-27#1855104 >> bahahaha, and the disinfo machine already spinning at full speed, e.g. 'Everything older than 0.16.3 (and the corresponding 0.14 and 0.15 fix releases) is vulnerable to one exploit or another' etc ☝︎

a111: Logged on 2018-09-27 20:45 deedbot: http://qntra.net/2018/09/power-rangers-inserted-inflation-bug-into-core-bitcoin-network-client-in-2016/ << Qntra - Power Rangers Inserted Inflation Bug Into "Core" Bitcoin Network Client in 2016

asciilifeform: http://btcbase.org/log/2018-09-27#1855107 << the proggy author's duty, and nobody can save him from it. fwiw i go roughly by size/complexity of orig draft. ☝︎

a111: Logged on 2018-09-27 20:58 mircea_popescu: more's the point : HOW do we establush "100is much, 10k is enough, etc"

asciilifeform: http://btcbase.org/log/2018-09-27#1855108 << i read'em when waking, when going to sleep, and in between have scrolling in real time within peripheral vision. but evidently even this not guaranteed to suffice.. ☝︎

a111: Logged on 2018-09-27 21:02 mircea_popescu: http://btcbase.org/log/2018-09-27#1855089 << i have no fucking idea how. i read the logs daily atm, mostly impelled by... outright fear. the best heuristic i know of, but otherwise this promises to be a first caliber bane as time goes by.

deedbot: http://bingology.net/2018/09/27/peso-watch-september-2018-edition-with-bonus-uru-gay/ << Bingology - BingoBoingo's Blog - Peso Watch September 2018 Edition With Bonus URU-Gay

asciilifeform: diana_coman: built & emplaced your sender-receiver, it is running nao, asciilifeformistan <--> BingoBoingostan

asciilifeform discussed subj with asciilifeform's brother, who answered 'whaddayamean, what size packet, at $defunctgamesco we only ever used 1480, for decade, ideal' ☟︎

asciilifeform: tho as i understand it, they did not account for the 8 byte udp header size, and thereby still fragged.

asciilifeform: going by the current empirical test, however, a packet that frags into 2 or even 3, typically goes. tho it remains to be seen whether they start falling down once you saturate.