ben_vulpes winces at the self-crit
mircea_popescu: in other words : if small yellow guy dun wanna be disturbed, it's possibly not because last week's festival.
ben_vulpes: does the dude have otherwise spot-on analyses of his weaknesses and folks' opionions of him?
mircea_popescu: as opposed to "i suck so much, they keep bumping me to various projects"
mircea_popescu: and i was thinking "they moved YOU. why are you thinking in terms of the company/job/whatever ?"
ben_vulpes: i have no firsthand idea, but i got the impression the party was for an entire week and nobody was going to be interested in hearing from auslanders for the duration.
mircea_popescu: oh the mooncake thing.
mircea_popescu: the who now ?
ben_vulpes: apparently it is bad manners to bother chinese people about business during the mid-autumn festival
mircea_popescu: from the above linked malaele piece.
mircea_popescu: "when my wife saw how many bottles of wine i bought for new year's, she bid me take them to the bathroom and dump them down the drain. i did as she said, because the wife is that person one shares with all life's happiness and unhappines. and sadness. and misfortune. and catastrophe. and all other disasters that wouldn't have happened had one not married."
asciilifeform: whenever i see 'pontiac' auto, i think of this .txt, having literally NO other context for it
asciilifeform saw these on bbsen in '90s
BingoBoingo predicts pantsuit attacks on sportspuck because racist by the end of November
BingoBoingo: Which gives the bored beer drinking crowd an alternative noise to drink to
BingoBoingo: AHA, and it doesn't help that sportball season overlaps the calendar with sportspuck season.
mircea_popescu: much much more vulnerable to trump than the liberal "endowments" ie hedge funds masquerading as universities.
BingoBoingo: The commisioner a bureaucrat.
BingoBoingo: The league is a bureacracy
mircea_popescu: most of the dollars--are green.
BingoBoingo: Lol, in this sportsball most of the players ARE black
mircea_popescu: trump certainly picked the right "issue" though.
mircea_popescu: in which case, i suspect federation rules requiring all black players wear a yellow star and apologize for being there before every game have even odds next year.
BingoBoingo: The pantsuit states aleady tuned out of sportsball because "concussions" and "mean"
BingoBoingo: This sportball was the financially healthiest reaping all the ad money, but... On the strength of marketing to Texas and the south.
mircea_popescu: iirc, they've all been struggling for economic relevancy in the past decade, if this even vaguely proves itself as a mass movement the whole shitshow's canned overnight.
mircea_popescu suspects the "sports" entertainment franchises won't weather the anti-pantsuit strom any better than dnc did.
BingoBoingo: But this isn't Pence pretending to know to ape after you leaving. This is Pence being the ape following ordinary sportball watching Trump voters leaving.
BingoBoingo: Oh, the sportsball players didn't stand for the song of country before game in protest of "Police r mean to us black folk" as they had been for the last couple weeks. Mike Pence went to game and left before it started.
mircea_popescu: dude didn't do the "proper" ustard gymnastics or what was it ?
BingoBoingo: And from the sports desk, Mike Pence: "I left today's Colts game because @POTUS and I will not dignify any event that disrespects our soldiers, our Flag, or our National Anthem."
mircea_popescu: "One has to marvel at the wide acceptance of our existing punctuation marks and the sociology of their acceptance." << hm. naggum doesn't actually have the ability to understand things he didn't set himself deliberately to work at understanding, does he ?
asciilifeform: [insert the typical demise of `usenet thread here...]
mircea_popescu: eventually they stop somewhere for the night, and in the wee hours of the morning monk goes and shakes the other awake. "brother... that woman... how could you fuck her ?"
mircea_popescu: asciilifeform dja know that koan with the two monks, meeting a woman at a bridge ? "please, one of you, fuck me! fuck me!" she says. one of them recoils in disgust, the other fucks her, then they continue over the bridge.
mircea_popescu: "SGML is a good idea when the markup overhead is less than 2%." << holy shit check him out. that's... a lot narrower than i'd have guessed.
asciilifeform: btw this imho was not accidental, various 'yarvins' ( incl. the original, who confessed to being part of this set ) want to bury their usenet past , 'unseemly'
mircea_popescu: hey, wikipedia (that bit it recently) still links them.
mircea_popescu: "continue" just redirects me to the original url.
mircea_popescu: with people this smart, who can wonder why the united states is the foremost leader of the world!
mircea_popescu: the pigs, lol.
mircea_popescu: Datskovskiy, Stanislav (29 June 2010). "The Wisdom of Erik Naggum". Loper OS (blog).
asciilifeform: the pigs?!
mircea_popescu: asciilifeform shit, you're in there!
asciilifeform: pioneered by the fat white men
mircea_popescu: "protector" of youth that sees no problem as long as the cumguzzler is black etc.
asciilifeform: lol re the pig ban
jhvh1: mircea_popescu: Green Dam Youth Escort - Wikipedia: <https://en.wikipedia.org/wiki/Green_Dam_Youth_Escort>; China's Green Dam : The Implications of Government Control ...: <https://opennet.net/chinas-green-dam-the-implications-government-control-encroaching-home-pc>; China's " Green Dam Youth Escort " Software - CircleID: <http://www.circleid.com/posts/20090608_chinas_green_dam_youth_escort_software/
mircea_popescu: it's really hard not to like the chinese, try as one might.
jhvh1: asciilifeform: Grass Mud Horse - Wikipedia: <https://en.wikipedia.org/wiki/Grass_Mud_Horse>; Grass - mud horse - China Digital Space: <https://chinadigitaltimes.net/space/Grass-mud_horse>; Song of the Grass - Mud Horse (Cao Ni Ma) - YouTube: <https://www.youtube.com/watch%3Fv%3DwKx1aenJK08>
asciilifeform: they tend more to 'grass mud horse'
mircea_popescu: i suspect chinese has a helluva lot of these.
a111: Logged on 2017-10-08 15:49 mircea_popescu: (because of reasons discussed in http://trilema.com/2014/the-problem-of-ideal-social-systems-reprint/ socialisms can't have categorical terms, defined in the normal manner, but must always include the ethical color of all words in the words. so "movement" becomes either provocation (bad) or progres (good) and so following for everything, stalin's cup is named by a different cupword than hitler's self-same identical cup)
mircea_popescu: well... he was "provocative". fine example of consensus-building word in world of http://btcbase.org/log/2017-10-08#1722690 because it's "bad" to one group and "good" to the oither so it all works out.
mircea_popescu: in other lulz, wikipedia : "Erik Naggum (June 13, 1965 June 17, 2009) was a Norwegian computer programmer recognized for his work in the fields of SGML, Emacs and Lisp. Since the early 1990s he was also a provocative participant on various Usenet discussion groups.[1]" << accidental hit of the consensus process, because "provocative" seemed both sufficiently insulting to the idiots and actually flattering to the fanbois.
asciilifeform: you do it again, and corrected version gets in the tagset.
mircea_popescu: and if it gets cut, the logparser sure as fuck has no way to fix it now.
mircea_popescu: because if i don't have the liberty to overload an url i can find no incentive to ever tickmark.
mircea_popescu: experience convinces me that anchor-first is correct way and the url-first html spec a kludge./
mircea_popescu: or else you know, in other `sluts:http://78.media.tumblr.com/8292708d5f10c32dd03e840cbf38b52f/tumblr_msoh7wWpev1snc48co1_400.gif
mircea_popescu: reason i prefer the `tag:url format to the `tag url format is that space is too syntactically meaningful.
asciilifeform: the url is now head of the list of what comes out when you click 'overcompetent' in logview
mircea_popescu: it never has to be done again whether done by ":" or by " " lol
mircea_popescu: who's gonna enforce the onceness.
mircea_popescu: or what is "the clean way" here
mircea_popescu: "further cost" aye.
asciilifeform: there will be ugh
mircea_popescu: i imagine the sane way to represent "outlyers are interesting. i find an `overcompetent:http://trilema.com/2013/whore-strat/ teenager appealing just as i find an `oversubmissive:http://trilema.com/2017/little-miss-pretty/#selection-247.0-251.710 `overinnocent lawyer appealing." in html is via the usual anchor mechanism, as "outlyers are interesting. i find an <a href=http://trilema.com/2013/whore-strat/>overcompetent</a> teen
mircea_popescu: gets point of reference baked in into tags ; gives me an incentive to even tagmark in the first place.
mircea_popescu: asciilifeform the more complex construction, `tag:url
mircea_popescu: evolvospec, for the record, is not a spec, exactly in the manner a bad girl may be great fun to be with but notrly a wife.
mircea_popescu: if the argument's good enough, at any rate.
mircea_popescu: asciilifeform maybe we can push a year 0 on the misfortunate kind hearts trying to follow this evolvospec.
mircea_popescu: somehow he manages to need much much less hammer to hammer in a point, for which i confess to admiring the man.
mircea_popescu: http://btcbase.org/log/2017-10-08#1723104 << ahahah! i really enjoy these apeloyee-alf discussions, for the record.
asciilifeform: iirc we found that every other char's been clobbered to death already
a111: Logged on 2017-10-08 22:13 asciilifeform: they still got'em in mircea_popescustan?!
mircea_popescu: http://btcbase.org/log/2017-10-08#1723088 << the stan-advantage is that you can get anything if you want to.
a111: Logged on 2017-10-02 19:52 mircea_popescu: so what is the idea here, if i wish to review the state of this, other than asking you, i could also what ? !#s ffa ?
asciilifeform bbl, will continue the misadventure later
asciilifeform makes a http://wotpaste.cascadianhacker.com/pastes/bnsZv/?raw=true and then notices that it not only is always off by smallint, but there is no way to fix, because the carries from the ( not used for anything else ) bottom halves of the muls, now are lost .
asciilifeform: ( notice, we only use the bottom halves of XHi*YLo and XLo*YHi )
asciilifeform: ( conceivably on a box with smaller cache, there could be speedup. )
asciilifeform: bad is that apeloyee was not only right re 'no more than 10%' but in fact there is NO observed speedup:
asciilifeform: or hm, there's gotta be a mistake in my test, bbl
asciilifeform: there is a missing term
a111: Logged on 2017-10-08 18:17 apeloyee: compute the higher part of product X*Y as XHi*YHi+ShiftRight(XLo*YHi+XHi*YLo, K), where K is size of XLo and YLo
asciilifeform: and large multron is made of a quantity of half-sized ones, they - half-sized again, and so on.
a111: Logged on 2017-10-08 20:57 mircea_popescu: i expect you just get one mult for the largest size and reuse it indefinitely, nfi what he's on about with the 2n luts for every sum
asciilifeform: http://btcbase.org/log/2017-10-08#1723072 << mul is not add, can't do this, the wider register costs you nonlinearly moar time
asciilifeform: http://btcbase.org/log/2017-10-08#1723064 << gcd does not need a karatsuba, the karatsubatron can be doing something else while gcd happens
a111: Logged on 2017-10-08 20:12 apeloyee: why not lak waht's already public: the list of small primes. that' not a compromise.
asciilifeform: they still got'em in mircea_popescustan?!
mircea_popescu: sooo... just had a lightning so close, made the landline phone ring.
mircea_popescu: they adhere to anything.
mircea_popescu: from what i hear they've a new model now.
mircea_popescu: http://www.beatmort.ro/aberatii/internalnote.php << in ancient lulz, rather related to http://btcbase.org/log/2017-08-01#1691973 ; this was genuinely a major piece of ro internet culture late 90s, when ustards started to "invest"/move in.
mircea_popescu: i expect you just get one mult for the largest size and reuse it indefinitely, nfi what he's on about with the 2n luts for every sum
apeloyee: *neither
apeloyee: nether does proposed method require 8x
apeloyee: not 8x, but nether does proposed method
apeloyee: as usual, if primitives don't leak- this won't leak anything but the small primes. this is obvious. or are we doing associativity proof in peano arithmetic or other hair-shirt arcana (c)
apeloyee: why not lak waht's already public: the list of small primes. that' not a compromise.
asciilifeform: no camel's nose in the tent.
asciilifeform: ( i.e. every time you write down a '+' that's a minimum of 2N LUTs used up, that cannot be used for anything else, where N is the operand width )
apeloyee: what, the small primes are a secret?
a111: Logged on 2017-10-07 21:48 apeloyee: http://btcbase.org/log/2017-10-05#1721485 << alternatively, can *construct* numbers which don't have very small factors. pick a nonzero remainder mod 2, mod 3, ... mod largest-prime-fit-in-your-primorial and find what number of primorial is congruent to it using chinese remainder theorem
asciilifeform: and incidentally i was not joking when said 32kb, it is fully my intention to eventually put whole thing on fpga where there will be certainly not even half MB of working space.
asciilifeform: barrett needs large scratch buffer for the mults; gcd can happen in-place.
asciilifeform: (i.e. you still win if you take 500x the cpu cycles, so long as you don't get cache-evicted)
asciilifeform: in practice on pc speed appears to be inversely proportional to memory used, rather than the cpu cycle count.
apeloyee: the fact that divisions are dog slow, for seconds << what barrett's reduction is for.
asciilifeform: the fact that divisions are dog slow, for seconds
apeloyee: bernstein's batch trial division would seem to straightforwardly ffaize. where's the problem?
apeloyee: because the acceptably fast algorithms are simpler.
asciilifeform: but hypothetically it may even be possible to ffaize bernstein's tree. or even to do it in such a way that doesn't wipe out the cpu winning from it. and even possibly to prove that it works and doesn't leak bits and doesn't let composites through once in a while.
asciilifeform: why the hell should i keep random crud in a table to pick up later.
asciilifeform: ( in our concrete case, r, a random , and p, a primorial -- for the pre-mr litmus test )
asciilifeform: ( and potentially for other primality tests, though i can think of some cryptosystems where it is handy )
asciilifeform: gcd is for the pre-mr sieve, is all.
apeloyee: anyway, I was saying that, if spending a week, may spend a small fraction of the time on the supposed-deterministic test
asciilifeform: situation where rsa is breakable, but no one can yet break it, makes it the sane option . because alternative is to become a donkey fucker ( rely on face to face for all comms , hope that nobody invents listening bug, etc )
asciilifeform: problem is that the historical period where crypto was a contest of bullet vs armour, rather than 'absolute bullet exists'/'absolute armour exists' is not over.
asciilifeform: and all of what they're selling -- stinks.
asciilifeform: in that it may be an actual problem , but NONE of the folx who ever publicly discussed it, have any business being taken seriously.
apeloyee: hey, before quadratic sieve was invented, they used to say that breaking 512-bit rsa will take eleventy zillion years and it's therefore Totally Secure (tm)
asciilifeform: mircea_popescu: if it were physically possible as the sole primality test, we'd all use.
asciilifeform: rather than, e.g., 'rsa broken OR aes broken OR prng broke OR riemann is false OR ...'
asciilifeform: but in light of this, a correct rsatron is still one that stands on nothing BUT the assumption that rsa is hard.
asciilifeform: it remains possible that -- somehow -- they do not
mircea_popescu: if your expectation is that the fifth attempt did not resolve the problem in a manner such as the fifth million would, there's deeper problems.
mircea_popescu: asciilifeform but the test that takes longer and costs more does not consist of manic re-measuring of the same one length, repeated millions of times.
mircea_popescu: oh and in other news
asciilifeform: but there does not.
asciilifeform: and incidentally if there existed an UNBIASED constructor of primes, i'd use that
mircea_popescu: asciilifeform yes but this is just the artistic side in you.
asciilifeform: ( i.e. i regard the proof behind strength of the probabilistic ver, as fundamentally stronger than the other's )
mircea_popescu: as per the ancient "doctor, random things in the house are talking to me, am i losing it ?" "have you started answering ?" "not yet" "then not yet"
asciilifeform: i'll take the p(failure) to the week's power, over the possibility of hypothesis falling and ALL keys fucked.
mircea_popescu: well, the running maybe not, but ~believing~ that it achieved something, surely.
apeloyee: doesn't run in geological (e.g. saxena) time << if you have faith in generalized riemann hypothesis and correctness of work on deterministic miller test - you have it. I don't, but running test for a week is imo greater crackpottery than believing in that.
asciilifeform: ( very often abuse of terminology, what people actually mean by 'deterministic version' is 'probabilistic with prng supplying the random' )
mircea_popescu: and in other bucharest tax office, https://i2.wp.com/www.cronicipebune.ro/wp-content/uploads/IMG_1539.jpg
asciilifeform: which he'd rather have -- key that he genned inside 50cent chip, staying there, or primality-torture on his fleet of pentiums etc
mircea_popescu: famously, maple misidentified the guy's number. not because of rng, eiher.
asciilifeform: mircea_popescu: chance of these without sabotaged rng is < chance of meteorite
mircea_popescu: but we don't have to start low. and we don't really want to, either.
apeloyee: each round of miller-rabin is mostly a modexp which makes some tests on the intermediate results. so I don't see how you can avoid a different version of modexp
a111: Logged on 2017-10-07 19:28 asciilifeform: http://btcbase.org/log/2017-10-07#1722358 << point was exactly to compare like items. i.e. heathendom does NOT get to 'win' by 'oh hey the hamming weight of exponent is only 2, not 4096, so we only do 4 modexps and not 8192'
mircea_popescu: incidentally, since we're on m-r : do we actually pick 4096 bit bases to avoid the arnault number problem ? to leverage the ffa flatness, as in http://btcbase.org/log/2017-10-07#1722376 ?
asciilifeform: mircea_popescu: review the mr algo , it is actually surprisingly easy to ffaize, just replace all 'return true' with flag := flag OR true, etc
asciilifeform: then all ACCEPTED primes took exactly same number of cpu clocks, to produce.
asciilifeform: apeloyee: each individual test has to be fixedtime though. then -- yes.
asciilifeform: mircea_popescu: all that means is that one of the inputs comes from rng.
mircea_popescu: it naturally makes assumptions about the item you're testing.
apeloyee: if you have N ffa-eligible tests, bailing early out after one of them failed is not a problem.as per above.
asciilifeform: yes there is.
asciilifeform: there is to be no compromise on leak.
asciilifeform: this is why ultimately entire primality test algo must be constant time, just like the other pieces.
asciilifeform: via the interval it took you to ACCEPT.
asciilifeform: and then if your miller-rabin is variable time, you have leaked key bits.
asciilifeform: for miller rabin you may end up ACCEPTING the test
mircea_popescu: apeloyee no, because as he well points out, the time it takes is not unrelated to the key.
mircea_popescu: so this is more a r-m problem altogether. as that's not linear.
asciilifeform: because all you do is get NEXT N bits from rng, they have ( if rng is proper ) 0 relation to previous N
asciilifeform: rejecting rng result that doesn't pass the gcd sieve -- leaks nothing
asciilifeform: possibly constantly, depending on the rsa keying system
mircea_popescu: i don't see what the problem is, practically. so you leak ... how many times you had to try to get a prime ?
asciilifeform: otherwise sieve is waste of time.
asciilifeform: apeloyee: i see your point. either we dispense with the sieve, or decide to count from the moment after sieve.
mircea_popescu: http://wifc.com/news/articles/2017/oct/06/russia-greek-ruling-on-russian-bitcoin-fraud-suspect-is-illegal/ in other coke machine competitions.
apeloyee: so if the number fails initial sieve, do you proceed to miller-rabin?
asciilifeform: *variability allowed not in the test, but in output
asciilifeform: apeloyee: no contradiction. the variability of time is in the ~test~, not the output result , which naturally will vary depending on what rng gave you
a111: Logged on 2017-10-05 19:38 asciilifeform: for the initial sieve ~prior~ to miller-rabin
a111: Logged on 2017-10-08 00:16 asciilifeform: the ONLY correct method of generating cryptoprimes, is to 1) get N bits from FUCKGOATS 2) determine, in fixed spacetime every single time, whether that string of bits constitutes a usable prime.
apeloyee: on a different topic, http://btcbase.org/log/2017-10-08#1722429 and http://btcbase.org/log/2017-10-05#1721484 seem to contradict each other. what's an initial sieve for if the algo must run in fixed time? i've interpreted it as "successful test must run in fixed time, failures can be variable-time", and make proposal accordingly.
asciilifeform: because on pc most of the wait time is for memory access.
asciilifeform: so far almost all of my theoretical predictions re which optimizations will be worth the effort, were wrong
apeloyee: your choice to not implement it. i've come to the conclusion that it's not worth the effort anyway.
apeloyee: right, unclear again. the muliply of N and floor(A*R/4^K) can be calculated mod 2^(K+1)
apeloyee: i.e. the lower part.
a111: Logged on 2017-10-07 21:25 apeloyee: the multiply-by-approximate quotient in barrett's also needs only the lower part (plus 2 extra bits to the left), and lower part of product can be computed exactly (since rounding is not a problem)
apeloyee: http://btcbase.org/log/2017-10-07#1722397 << I was unclear. Let A be the number to be reduced mod N, R the approximate reciprocal, K the ffa bitness fitting the modulus, then we know that 0<A - N*floor(A*R/4^K) < 2*N <2^(K+1). So might as well calculate A - N*floor(A*R/4^K) modulo 2^(K+1).
mircea_popescu: i was going to say, the calculated %s rarely match. but theoretically, it should be less than 0.1
asciilifeform: apeloyee: theoretically. but cache locality win from smaller memory segment sometimes gives surprising winning. the example above, for instance, gives 2x speedup rather than my predicted 25%.
asciilifeform: in the end might even release different variants that have different complexity tradeoffs.
asciilifeform: and then bernsteinian karatsuba, possibly, and whatever else i can think of.
asciilifeform: which i will also make, and decide if it was worth the cost
asciilifeform: it is! but much smaller than, for instance, the secretshift-barrett.
asciilifeform: for instance unrolled comba wins 20-25% speed, but i did not use it in place of the generic because it is longer and harder to read.
asciilifeform: apeloyee: my strategy so far was to introduce moving parts very, very reluctantly ( started with egyptian multiplier, for example ) when there is absolutely no choice.
apeloyee: so, are you putting off the top-half-multiply
a111: Logged on 2017-10-07 00:38 asciilifeform: mod6: you will notice that the barrett in 'crc handbook' is more complicated : it shrinks the x and then compensates later. this relies on normalization , and constanttimeized incarnation of it would have to work as apeloyee described ( i'ma try it much later, once i see what can be had re speed strictly from having asymmetric karatsuba instead of the current mega-waste )
apeloyee: the best case pretty much.
apeloyee: 2 half products out of 3 on the first level of recursion, 4 of 9 on second, and 8 of 27 on third, assuming 64-bit words and unrealistic 2-fold speedup of comba for half-multiply, and no overhead in karatsuba,
apeloyee: and most products for which the comba is called, are full products, not half products
apeloyee: see, it does three recursive calls, meaning the speedup is wholly dependent on the speedup of comba for half-multiply
apeloyee: compute the higher part of product X*Y as XHi*YHi+ShiftRight(XLo*YHi+XHi*YLo, K), where K is size of XLo and YLo
a111: Logged on 2017-10-07 21:14 apeloyee: http://btcbase.org/log/2017-10-07#1722289 << and the point of doing karatsuba is? you do 2 recursive calls to Mul_Karatsuba_TopOnly and one to Mul_Karatsuba. should've simply calculated upper_part(XLo*YHi), upper_part(YLo*XHi) and XHi*YHi
apeloyee: http://btcbase.org/log/2017-10-07#1722395 << why do karatsuba when you can just shift and add them, like in your W_Mul
trinque: that'll change, but as part of limiting the thing to this channel only
trinque: it's already in the logs.
trinque: isn't reflected until I actually credit the account
trinque: danielpbarron: give it another try
mircea_popescu: I GUESS THE WORLD WILL NEVER KNOW.
mircea_popescu: look at that, and before the archival bot kicked in, even.
mircea_popescu: aww ? they beleeted it already ?
mircea_popescu: even to the degree of http://trilema.com/2017/zuleika-dobson-or-an-proper-love-story/#selection-3413.588-3413.746 ; or rather... starting with.
asciilifeform: well yes but loox like intends to be a coherentwave of gurlz rather than randopolarized, if you will.
asciilifeform: mircea_popescu: for some reason i can't help but think of the old lul with chinese on footstools synchronously jumping
mircea_popescu: and in entirely other lulz : http://trilema.com/wp-content/uploads/2011/05/jsexnetrwork-soft-on-demand-s.jpg the exhibitionists are those girls who make the obviously deliberate effort of not covering their cunt with their hands.
mircea_popescu: and of course, "Alan Green, named ambassador to Romania by President George Bush [who called him "a good friend", "intransigent", "well introduced to my take on freedom and democracy"], died Friday in his home in Palm Springs, Calif. He was 75. Mr. Bush selected Mr. Green in 1989, and he moved into the American Embassy in Bucharest just two weeks before the dictator, Nicolae Ceausescu, was executed."
mircea_popescu: the important point for romania was that gorby wasn't going to deliver all the oil romania had contracted (and paid for). so ceausescu went to iran ; where he got ~40mn barrels with a further option, to be paid in romanian agricultural machinery.
mircea_popescu: anyway, the whole meeting went in that vein, ceausescu pointed out to soviet troops still at praga, gorby was liek "oh, that is a bilateral matter" "da, stiu, este un acord bilateral incheiat dupa ocuparea cehoslovaciei" (yea, i know... post-occupation bilateral). then gorby says they can't agree in this matter and ceausescu agrees with him.