asciilifeform: ( i.e. sequence where only ADD, SUB, MUL, SHR/SHL, MOV, all the way from head to toe , a,b,c in, a^b mod c comes out )
asciilifeform: the other thing, if one is writing a proggy where only 1 ffawidth is used, it is possible to unroll all of the loops, derecursivize karatsuba, etc. and get straight line of instructions
asciilifeform: ( and it is fortunately much easier to find mistake in 5k loc , with ~one~ execution path, than in 150kloc with 900000 paths... )
asciilifeform: diana_coman: sadly i do not know how to 'guarantee perfection'. all i know how to do is to bake maximally 'fits in head' and bank that the folx here will find mistake if it turns out that i made one.☟︎
asciilifeform: ( where yer tryin' to save coad space )
asciilifeform: btw it is possible to dispense with FZ_LoMul , and simply discard upper chunk of ordinary karatsuba product. this costs ~10% cpuwise, and may be justified for some proggies.
asciilifeform: ( hence the somewhat pedantic treatment of the subj, 2 separate proofolade articles. was going for 'measure 7777 times, cut 1ce' there. )
asciilifeform: you may find it interesting that asciilifeform's 1st shot at barrett managed to have a subtly fatal mistake in the proof, and correspondingly a proggy that failed on <0.1% of input space.
asciilifeform: this is the place where slow an' careful is the Right Thing.
asciilifeform: ( the key work for diana_coman et al, from ffaology pov, is to eat the proofs )
asciilifeform: let's say scout reported a 1k-man garrison in the point of origin. and nobody knows precisely how many hauptmann fuckface will be sending.
asciilifeform: let's hear mircea_popescu's logical end of the stick tho. mircea_popescu finds himself partizan commander again, in the field. has 300 men. how he decides how many to send to ambush nazi convoy ?
asciilifeform: i'll be surprised if i haven't nailed the last comment typo by the time done adding chapter-sized knobs, tbh
asciilifeform: thing's about 50-60% comment by mass.
asciilifeform: imho the haskellism where 'program carries proof' is generally a sad thing and results in soup
asciilifeform: ( and it still needed conventional docs, illustrations, etc )
asciilifeform: mircea_popescu: if someone knows how to make ffa moar 'docs is the program'-y, oughta write in and tell asciilifeform , cuz i came as close to it as i knew how.
asciilifeform: mircea_popescu: upper bounds can be known in advance . ( in ffadom, at least. ) in larger meta-heuristics, naturally cannot be ~known~, but can be estimated.
asciilifeform: mircea_popescu: if i v-versioned the ~docs~, we'd be at 50-ish, nao, i suppose
asciilifeform: ( where it dun make sense to make anymoar changes without simply baking new item out of the dismembered corpse of old )
asciilifeform: mircea_popescu: how does bridge constructor decide 'it should hold 50 tanks time margin of 3' and not 55 and not 45.
asciilifeform: fwiw i gave orig vtron, 'v.py' , 100. it's on 97 (phf's revision) and when i get around to massaging it, i suppose will be 96, and i dunsee why to touch it again after that..
asciilifeform: i.e. 'if this thing needs >255 vpatches after today, then asciilifeform is a tard and somebody else ought to do the whole fucking job over'
asciilifeform: ( i.e. that branch predictor nukes ~all~ win from skipping zeros )
asciilifeform: possibly mircea_popescu is simply less thick than asciilifeform and expected the find
asciilifeform: the copy on my www ( and later cleaned further by diana_coman ) is from gpg 1.4.10 ( itself from signed tar that mircea_popescu dug in in '14 )
