Show Idle (>14 d.) Chans

← 2020-11-16 | 2020-11-18 →
feedbot: << The Montevideo Standard -- Rocha Under Alert For Portuguese Man O' War
cgra: asciilifeform: a 256-bit peh tape like ".1 .FF LS .1 .3 MX" will bomb on me at here.
cgra: i haven't followed exactly a linear learning schedule, so some of the earlier chapters are somewhat unfamiliar to me... but, the reason appears to be this optimization.
cgra: say, for example: if X := 2**255 and M := 2**255 - 1, X - Q would end up negative (as that tape does in my case), because of the optimization.
cgra: as a side note, a 256-bit test tape "10k_shots_256bit_ffa_slid_rnd.tape" also bombs the same way on me
asciilifeform: cgra: thanks, will look.
asciilifeform: cgra: your example indeed bombs. the test tape doesn't, tho, interestingly (are you sure you linked the right one?)
cgra: i'll re-check
cgra: sha256sum of the test tape in q is "b6388e30fb881ebd7a37bf45fb07003e0420f2c923495dfc60c0705f22cca6a2"
asciilifeform: aa there we go.
asciilifeform: does bomb (initially when tried i did not specify sufficient run life)
asciilifeform: ... apparently did the same in the orig. bash script which ran this tape !
cgra: hehe :)
asciilifeform: ty for the tip-off, cgra . loox like ch14 will need same treatment as ch15 earlier.
cgra: np, glad i could provide finally some useful input
asciilifeform: cgra: most useful external input to date, in fact.
asciilifeform: the bomb in 15 i had to find on my own, despite certain pompous maffs phd supposedly 'read' and signed.
cgra: i dunno if i'd have spotted the issue by walking through the physical bounds proof. i stumbled upon it while experimenting
asciilifeform: cgra: after the incident w/ 15, i've been rereading & reproving continuously ch21b/c since april !
asciilifeform: at some pt will finally release it... apparently later rather than sooner.
asciilifeform: !!rated cgra
deedbot: asciilifeform has not rated cgra.
asciilifeform: !!key cgra
deedbot: Not registered.
asciilifeform: hey cgra wouldja get in deedbot already , so i can rate you ?
cgra: asciilifeform: how much do you have currently half-complete material for ffa?
asciilifeform: cgra: just about all of it
asciilifeform: ( 21 is done, as is 22, only hashism is entirely unwritten yet )
cgra: right
asciilifeform will have to unravel 14 and rewrite all of the material thereof, will prolly take a while.
cgra: i've been delaying the wot registration, because i don't have a proper understanding of best realistic approach to it yet
asciilifeform: cgra: elaborate ?
asciilifeform will be back in ~40m
cgra: asciilifeform: i mean, i thought i'd learn some rsa etc first, to be able to decide what's 'good enough'. in case i have more choices avail than the current wot people had at their registration time
asciilifeform: cgra: atm trinque's wot system is strictly rsa, afaik.
trinque: cgra: yes, use an RSA key, and there are key-length discussions in the old logs to be found.
trinque: that said, don't fester too much about doing things "right" when the foundation itself is a pile of accidental crap.
trinque: when we've moved to asciilifeform's crypter for this, much more can be said about right.
trinque: as of now deedbot sits on a gnupg.
cgra: trinque: ok
asciilifeform: wb trinque
asciilifeform: $ticker btc usd
btcinfobot: Current BTC price in USD: $17599.51
asciilifeform: !w poll
watchglass: Polling 16 nodes...
watchglass: : Alive: (0.082s) V=99999 (/ Jumpers=0x1 (TRB-Compat.) Blocks=657380
watchglass: : Alive: (0.080s) V=99999 (/ Jumpers=0x1 (TRB-Compat.) Blocks=657380 (Operator: asciilifeform)
watchglass: : ( Alive: (0.135s) V=99999 (/ Jumpers=0x1 (TRB-Compat.) Blocks=657380
watchglass: : Alive: (0.023s) V=99999 (/ Jumpers=0x1 (TRB-Compat.) Return Addr= Blocks=657213 (Operator: whaack)
watchglass: : ( Alive: (0.103s) V=99999 (/ Jumpers=0x1 (TRB-Compat.) Blocks=657380 (Operator: asciilifeform)
watchglass: : Could not connect! (Operator: jurov)
watchglass: : ( Alive: (0.140s) V=70001 (/ Jumpers=0x1 (TRB-Compat.) Blocks=657380
watchglass: : ( Alive: (0.115s) V=99999 (/ Jumpers=0x1 (TRB-Compat.) Blocks=657380
watchglass: : Violated BTC Protocol: Bad header length!
watchglass: : Alive: (0.167s) V=99999 (/ Jumpers=0x1 (TRB-Compat.) Blocks=657380
watchglass: : Alive: (0.183s) V=70001 (/ Jumpers=0x1 (TRB-Compat.) Blocks=657380
asciilifeform: cgra: pretty subtle mistake, btw. the proof aint wrong, per se -- Q <= X in all cases. and R in fact has at most Wm+2 bits. but in subtraction in step 5, it is necessary to add the carry to the result R, rather than bomb if carry != 0.
asciilifeform: in cgra's example, at step 5 :
snsabot: Logged on 2020-11-17 10:01:28 cgra: asciilifeform: a 256-bit peh tape like ".1 .FF LS .1 .3 MX" will bomb on me at here.
asciilifeform: x == 0x40000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000
asciilifeform: x - q == 1 ; but what gets computed is X(1 .. Rl) - Q(1 .. Rl) , i.e. :
asciilifeform: 0x00000000000000000000000000000000000000000000000000000000000000000000000000000000 - 0xFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF
asciilifeform: which yields 0, and a carry (borrow) bit 1 .
asciilifeform: which is the actual answer, given as we know that q <= x, and did not bother running through entire subtraction (but only for the length req'd to obtain the max known bitness of R)
asciilifeform: will chew on ch14 for a while, but currently afaik that's the whole story here.
← 2020-11-16 | 2020-11-18 →