the k tree

« first day (702 days earlier) ← previous day next day → last day (833 days later) »

juanez

00:43

assuming the "int is the char themselves" - in the ascii table?
why not simply 'seed' your global list with the ascii table, you can then treat everything the same ?

2 hours later…

Razetime

02:43

just saying, would be a good idea to link a K documentation page/ tutorial somewhere in the description

coltim

@Razetime not a bad idea. I think a good starting point is the oK manual (at least for oK and ngn/k)

Razetime

yep, ngn told me

searching up "K documentation" shows schooling websites

coltim

@Razetime my trick with googling k is to use "kdb" instead. the official kx/q docs are mostly applicable (although they're focused on q and not on the underlying k)

Razetime

nice

chrispsn

@ngn yes

Razetime

03:15

@coltim this page is GREAT

coltim

@Razetime all credit to @JohnE!

Hao Deng

03:29

@ngn 1bit int is very useful, like chess bit board.

is ngn/k no longer open source?

Razetime

04:19

@HaoDeng I think it's on sourcehut, still open source

chrispsn

05:07

yes: git.sr.ht/~ngn/k/tree

3 hours later…

JohnE

07:53

it seems that my documentation has proved at least as useful as my k interpreter for the broader open-source vector languages community

ktye

absolutely

Razetime

@JohnE yes, very helpful

2 hours later…

ktye

09:44

@ngn you could also store compact symbols as 6chars lowercase only in 4 bytes.

ngn

10:28

@juanez the int is the chars. i mean, if the identifier is "abc", it's represented as 'a'|'b'<<8|'c'<<16. this way it can be converted back to a char* simply by casting (assuming little-endian arch).

if it's longer than 3 bytes, it becomes the (negative of) an index in a global grow-only list, and getting back the char* takes some pointer chasing, but so what.

ktye

in my current experiment i parse symbols (max 6 chars) as:
if x >= 'a' && x <= 'z' { // parse symbol
Y *= 32
Y += x - '`'
}

ngn

@ktye that takes a bit more effort to encode and decode, compared to no-op casting. but it could make sense if you're expecting longer identifiers on average.

ktye

i also need some bits (of these 32) for type-tagging.

ngn

@ktye 32 is 2^5, and you said you can fit 6 chars, so that's 5*6=30 bits, so there are only 2 bits left for tagging

ktye

:56986348 no problem
..000 pointer(list)
....1 int x>>1
...10 symbol x>>2
..100 operator x>>3

ngn

10:40

@ktye why not Y += x - 'a'?

ktye

because the parser is "basic" and does not keep state. Y=0 means it's not currently parsing a symbol. same for numbers.

ngn

who needs more than 3 chars anyway :)

52letters * 62alphanumerics^2 ≈ 200000, that should be enough for everybody

ktye

@ngn how do you use symbols for variable lookup?

ngn

@ktye globals are stored in an ordinary dict (there's room for optimization there), locals are looked up by index (like offsets from some base pointer on the stack)

ktye

10:58

@ngn that means you have to do a linear scan over all keys to find the location of a global.

ngn

@ktye yeah

some day i'll get to optimizing the bytecode a bit more, and then a reference to a global can be a direct pointer

ktye

in my k, symbol values are offsets for both: to lookup the string from from the list of keys and to lookup values of global variables.

chrispsn

i'm surprised this gives 1 in both k9 and ngn/k

  97~"a"
1b

ngn

@ktye so if i name my variable "abc", its offset would be 1091 (i.e. 32/1 2 3)?

ktye

@ngn no. the above 6 char thing is another language. the offsets are in my k implementation.

ngn

11:06

@chrispsn thanks, i'll fix that

ktye

searching is done at parse time only for symbols. if the key exists, it's offset in the key list is used as the value (otherwise appended). the global value list must be kept in sync (has always the same length). this means ab`c also creates 3 global variables even if they are never used. as a consequence, there is no undefined error.

ngn

@chrispsn actually, it already works. are you using an old version?

chrispsn

@ngn huh. i was using the web repl... clearly going crazy

ngn

@chrispsn why does k9 print 1b instead of 1? are booleans not ints?

chrispsn

they are ints to the user but they display as booleans and i think they are booleans behind the scenes

boolean display/syntax may go away entirely

ngn

11:22

@Razetime in this chat room's description?

Razetime

@ngn yep yep

ngn

room topic changed to the k tree: see oK's manual for a quick introduction to k (k6 dialect) (no tags)

Razetime

looks like markdown doesn't apply at all to the description

ngn

room topic changed to the k tree: see github.com/JohnEarnest/ok/blob/gh-pages/docs/Manual.md for a quick introduction to k (k6 dialect) (no tags)

Razetime

cool

ngn

11:26

@Razetime SE chat is strange

chrispsn

i've been writing a 'welcome to k' doc but it's not really a tutorial - just a bunch of links - and it needs work gist.github.com/chrispsn/6b9ed1c3dc8bf5594d8e6022495f390a

Razetime

@ngn I think everyone unanimously hates SE chat lol

Make a K Wiki, pronounced "quickie"

ngn

@chrispsn that would be a very useful thing to link to

ktye

@chrispsn "...tables (one column per dimension)" is that correct?

i'd say that tables flatten higher dimensional arrays by catenating data AND adding an additional key column (which can be grouped on to recover an index into the dimension)

chrispsn

@ktye I think of the column count as the equivalent of matrix rank

ngn

11:37

@Razetime kona has a relatively well-developed wiki

chrispsn

@ktye specifically with this example in mind news.ycombinator.com/item?id=24168132

chrispsn

12:08

hmm going back to speed of lobsters

i wonder - if the fill char is pretty flexible - if anything could be done with {x*y{y_x,y=:*x}/x}

i mean this gets to the same char count (and with the right fill value) without the expand trick {`c$42|x*y{y_x,y~:*x}/x}, but it's not an improvement

it could be cool if "ab"|"cd" resulted in a string, not ints

@chrispsn (although it does now match ngn/k in k9 :) )

2 hours later…

coltim

14:01

@chrispsn going back to this, something I think could be interesting would be ?=list returning the original list. I would have to play around more, but there may be times where it's more convenient to work with the =x representation (the tricky part now with that is transforming it back to a flat list)

coltim

14:23

me too! playing around with
https://ngn.bitbucket.io/k/#eJxVUttu2zAMfc9XEN7LJgQYhmLAED85sixfI+sSS3YQFFmWtFvTdWnTxm7RfvvkOFdBECXykIc64HLwNr+u+58d5LjXzZf5oPlA9fvXXm8zeJvUbuO+TpeTxq2n773NxInAZzBiCjAbKcFSUCEBmRPig6dAhxEOIWVDqYiQ4EfEcR3M/Nagw5o/oIv16/KJFp1xpi2fIlJhT1oOJaIRtXWUJ+25aSGzp0NOBx7TNNQezYEX5UhVWWCB48515nlGq9vt7OYfWr80fzev98sumYvEw8aA1kHsQSzL2A8CDwodUNOmcRwUrV23X9iRLs+7ftk9u1IV6EpSVoQhxCzXEFFsIBubGEyWRdL4I1CQZkNbjeJxbNLhQZ+b85ItyzP6g+qTa4V+dhSJCSCmWUXiMh9C6Tlu3zlXeXEhTECAmTKB0iKIaWE2Xp/iDLSAIIsSG2Ja7G8PaPuIlve/7zpQTkmloGQVLjkHXxgFFBdeDKGBFCIDhmRRljFz6mXh9K9+fHL2LAnHaQJDCBOqRNiiWpY7p//t6…

8 hours later…

ngn

22:13

@JohnE the graphical demos are very useful too, to impress people

« first day (702 days earlier) ← previous day next day → last day (833 days later) »

all rooms