asciilifeform: for completeness : 'drivers' subtree weighed 323M ; 'arch' (machine-specifics) : 126M; leaving a 176M (quite evenly spread across remainder) .
asciilifeform: the sheer magnitude of the bloat, is impressive even vs. microshit's (iirc 'win2k' kernel src weighed ~300M)
asciilifeform: not, incidentally, meaningfully documented anywhere (tho no shortage of 'these here are docs, cuz we said!' crapola) -- tho i do not expect this surprises anyone.
asciilifeform: to port the kernel to 'M', asciilifeform ended up having to read pretty good chunk of the arch/mips subtree, and it was comparatively small -- still bucket of sweat.
asciilifeform: may even be that the only practical approach is to discard ~100%~... ( who can meaningfully read 125M of obfuscated-c liquishit ? )
asciilifeform: (i suspect -- at least 80%. but do not know for fact.)
asciilifeform: still remains to be seen, what subset can be discarded w/out reading
asciilifeform: ( ~not~ including any particular gcc , or ~gcc's deps~, or 'binutils' req'd by the latter. just kernel. )
asciilifeform: meanwhile, in not-news, linux 3.16.70 src weighs... 625 MB .
asciilifeform: there was at one time a place that molded in-ear headphone to client's ear. assumed somehow that this exists for tits, and perhaps comes with the silicothing, lol
asciilifeform: http://btcbase.org/log/2019-07-30#1925619 << i've been seeing these for years nao. they're all contemptible scams, incidentally: profit model is that shit student ~pays~ to test for 'accidental' plagiarism , so that homework won't set off detector when his prof tests☝︎
asciilifeform: this incidentally is 9000x simpler than pc's page table. but still quite inconvenient under emulation (vs. physical iron, where it's a o(1) content-addressed ram lookup )
asciilifeform: ('asid' in practice corresponds to process id. but mips gives only 1 byte for it, so e.g. linux shuffles'em so to be unique at any given time in table )
asciilifeform: d0 means 'dirty' i.e. writes are permitted. v1/d1 are same as this, but for when bit-below-bottom-of-tag is ~odd~.
asciilifeform: 'G' means entry is 'global' (query 'asid' disregarded) ; V0 means 'valid when bit-after-bottom-of-tag in addr being translated is 0', and if match, translation replaces tag with pfn0. 1 means this but ~odd~, and pfn1, respectively;
asciilifeform: folx who dun give a shit can safely skip below, but imho oughta put in log :
asciilifeform: ( phunphakt: 1980s mipsen not only crashed, but would actually ~smoke~, if 2 identical tags! they had ultra-simplistic dual-ported sram as the table, would result in short circuit ! )
asciilifeform: 1 observation : given that the tags ~must~ be unique (an iron mips will actually hard-crash if you manage to -- it aint easy -- stuff 2 entries 2/ identical tags into the table) it may be possible to use some variant of crc as 'hash' for 'hash table' or similar (catch is, it gotta be ~fast~. as in, coupla instructions at most.)
asciilifeform: the solutions i tried, were : plain iterative search ; and the simd thing. both sucked.
asciilifeform: 1 naive solution i considered, was radix search, but with straight 3byte index, that's a 2^24-entry (4bit ea, i.e. index, so weights 1MB) table, would blow the pc cache to hell.
asciilifeform: writes to this structure have to be as fast as reads, as any time that a query results in 'not found', the table is written (typically one of the entries is replaced at ~random with a matching one, by the os)
asciilifeform: you have at most 16 entries, each of which has a 'tag' (3 bytes) that must match the query's tag .
asciilifeform: this kinda thing is where fab capability would be +9000 handy. but even without it, still doable, imho.
asciilifeform: key ~gen~ imho especially dun belong on pc.
asciilifeform: even if the 4096bit modexp that's 2sec on pc, is 200sec on battery-powered micro, still imho entirely usable.
asciilifeform: mp_en_viaje: ftr asciilifeform is of same position on 'where to rsa' as in 2013 -- ideally get it the hell off the pc.
asciilifeform: at 1 time i soldered 1 in, manually, but it was pretty cavernous lappy
asciilifeform: (lappy offers no convenient internal parking for it)
asciilifeform: the gnarly bit is that attaching fg to lappy is inconvenient, given the current form factor.
asciilifeform: no conceivable reason why it'd have to be laptop-sized. this being said, i'd rather use lappy with msdos peh than unix, for cipheration/signing.
asciilifeform: mp_en_viaje: atm folx use entire lappies (iirc there's a mp photo w/ same) for the job. so even the pictured item would be substantial improvement re portability.
asciilifeform: sorta like what we were orig wanting to build prior to fg.
asciilifeform: mp_en_viaje: why transmit the key over the (potentially loud) serial ? rsa in the box.
asciilifeform: ( was actually orig reason asciilifeform asked ave1 et al for mips gnat )
asciilifeform: i.e. variant where the end of the pipe is a device running also peh , talking over r232.
asciilifeform: mp_en_viaje: peh was specifically designed to work cleanly with ~this~ variant also. (hence 'cutout' mechanism)
asciilifeform: i did expect that folx would ask for this. there aint afaik a portable way of doing it, however, will require ugly c imports.
asciilifeform: ( 'build in' would require explicitly calling the keyboard, also, a la gpg, rather than 'pw from stdin', tape is already coming from stdin )
asciilifeform: mp_en_viaje: it is entirely possible to build-in. this doesn't really solve 'unix is leaky' tho.
asciilifeform: sorta why gluing lang-incompatible orcistans generally requires an external conqueror (ru, austrians, turks, ..) who impose yet third lang
asciilifeform: ( orc 'nationalist' wanks tend to follow language map, near as i can tell. )☟︎
asciilifeform: http://btcbase.org/log/2019-07-29#1925396 << by some accounts, this disease killed even crowned heads ( ceausescu ) , and saharov, and buncha other 'names' , large an' small , ( an' many yrs ago almost killed asciilifeform , who did not come from factory with mp's complete immunity ) and fuck knows whom else☝︎☟︎
asciilifeform: i wrote one, but it needs a bit of polish. imho it may be an unhealthy thing, though, gold standard for 'secrets kept unattended' would be a dedicated iron, rather than 'type in passphrase into this linux proggy, and 'faith-based' count on the os not to stow the keystrokes somewhere'
asciilifeform: given that 'peh' eats tape from stdio (this is by design, so that can 'chain' tapes) the only practical solution to symmetricizing seekritz kept on disk, afaik, is commandline 'serpent' util or similar.☟︎
asciilifeform: http://btcbase.org/log/2019-07-29#1925432 << needs keccak to actually rsa 'in anger', picked this up nao that asciilifeform returned from awol (where wrote 4l ln of ultimately failed asm experiment 'm') . but also thought about this point prior, will elaborate :☝︎☟︎
