Arbitrary precision integers (bignum)

[sanjayjain159]

Is there a convenient way to introduce arbitrary precision integers (bignum) in Janet? Perhaps through long strings?

[sogaiu]

For some use-cases, may be this is suitable?

An interest was expressed in a wrapper for GMP here, not sure if there was work on that.

I think at one point there was some comment about it NOT being likely for there to eventually be something built into Janet but I'm fuzzy on that (^^;

[amano-kenji]

Scheme has full-numeric tower which includes big integer and big rational numbers.

[tttuuu888]

Related to this topic, I think one of Janet's current bignum-related problems is that it appears to support bignum but behaves strangely.

Below are some examples. (tested on Janet 1.33.0-f91e5994 linux/x64/gcc)

repl:24:> 11111111111111111
1.11111111111111e+16
repl:25:> 11111111111111110
1.11111111111111e+16

repl:27:> (- 11111111111111111 11111111111111110)
2

repl:28:> 1111111111111111111111111111111111
1.11111111111111e+33
repl:29:> 1111111111111111111111111111111110
1.11111111111111e+33

repl:31:> (= 1111111111111111111111111111111111 1111111111111111111111111111111110)
true

repl:34:> (- 1111111111111111111111111111111111 1111111111111111111111111111111110)
0

Wouldn't it be better to generate an error rather than allowing abnormal behavior like above?

In the long term, I hope that Janet will support bignum natively.
Janet-big mentioned above looks great. But unfortunately I found same abnormal behavior like below.

repl:5:> (=  (big/int 11111111111111111111)  (big/int 11111111111111111112))
true
repl:6:> (=  (big/int "11111111111111111111")  (big/int "11111111111111111112"))
false
repl:7:> (-  (big/int 11111111111111111111)  (big/int 11111111111111111110))
<big/int 0>
repl:8:> (-  (big/int "11111111111111111111")  (big/int "11111111111111111110"))
<big/int 1>

As above, Janet-big works normally only when bignum is provided as a string.
I'm not sure if this is due to a limitation of current Janet or Janet-big is missing something.
But if bignum is added to Janet, It would be nice if bignum could be freely used as a number rather than a string.

[CosmicToast]

The reason this happens is because Janet uses floating point numbers built-in. Loss of precision in floating point use-cases is normal and expected, there are some use-cases that explicitly depend on this.

Typically, the way you're meant to use bignums is by starting from something that isn't a bignum, turning it into one, and then letting it "contaminate" the rest of the numbers it comes into contact with.
For example, you might do a `(inc (big/int "11111111111111111111")) and it would turn into 11111111111111111112, and you'd go on from there.

I finally got around to this point on my list (thus why I'm commenting now) and am working on bindings to libtommath (which will coincidentally provide a lot of the desired functionality from prior discussions, such as next-prime).
Currently I have very little to show (I can instantiate a zero and then output it to a buffer 😄), but the scaffolding to get seriously building is more or less "up" now.
I hope to have something ready (up to my quality standards) by the end of February, but it might end up being longer if work is too busy.