Array/Sequence/Object destructuring/unpacking macros attempt to quench the thirst of (Python, ES6, et al.)-spoiled people.
nimble install unpack
import unpack
let someSeq = @[1, 1, 2, 3, 5]
someSeq.unpackSeq(a, b, c) # creates a, b, c with 'let'
# is expanded into:
# let
# a = someSeq[0]
# b = someSeq[1]
# c = someSeq[2]
# or equivalently:
[a2, b2, c2] <- someSeq
someSeq.unpackSeq(a3, _, _, b3) # use `_` to skip index
# is expanded into:
# let
# a3 = someSeq[0]
# b3 = someSeq[3]
echo a, b, c # 112
someSeq.unpackSeq(var d, e) # creates d,e with 'var'
# or equivalently:
[var d2, e2] <- someSeq
someSeq.aUnpackSeq(d2, e2) # assigns someSeq[0] to d2, someSeq[1] to e2
# or equivalently:
[d2, e2] <-- someSeq
# yes, <-- for assignment; <- for definitions.
# This is not a typo.
type
Person = object
name, job: string
let tim = Person(name: "Tim", job: "Fluffer")
# create name, job with let and assign respective member values to them
tim.unpackObject(name, job)
# or equivalently:
# {name, job} <- tim
# is expanded into:
# let
# name = tim.name
# job = tim.job
# you can also unpack into custom names using 'as'
{job as someJob, name as otherName} <- tim
# or equivalently:
tim.unpackObject(job as someOtherJob, name as someOtherName)
# is expanded into:
# let
# someOtherName = tim.name
# someOtherJob = tim.job
var
secreteState, arg = 0
proc someProcWithSideEffects(person: Person, input: int): Person =
secreteState += 1
{var job, name as newName} <- person
newName &= $input
result = Person(name: newName, job: job)
# using this at the end of proc chain will not invoke proc chain multiple times
tim.someProcWithSideEffects(arg).unpackObject(name as tim0, job as job0)
# or equivalently:
# {name as tim0, job as job0} <- tim.someProcWithSideEffects(arg)
# is expanded into:
# let someUniqueSym1_212_498 = tim.someProcWithSideEffects(arg)
# let
# tim0 = someUniqueSym1_212_498.name
# job0 = someUniqueSym1_212_498.job
# if you haven't noticed,
# this means we can unpack named tuples like objects
type
SomeTuple = tuple[x, y, z, i, j, k: int; l, m: string]
let someTuple = (1, 3, 7, 0, 3, 6, "so", "lengthy").SomeTuple
# with vanilla nim, to get arbitrary fields
let (_, diz, _, iz, _, _, _, it) = someTuple
# it gets lengthy
# with this package
{y as diz2, i as iz2, m as it2} <- someTuple
# Mind. Blown.
# also, if you only care about the first three items
[nice, n, sweet] <- someTuple
# with vanilla nim
let (youNeedTo, writeSoMany, underscoresMan, _, _, _, _, _) = someTuple
# also supports nested unpacking
let nestedTuple = ((123, 321), (-3, (1, 2, 3, 4)))
# This nesting is unnecessarily deep, but it's supported.
[ [run, outOf], [names, [_, _, toUse, now]]] <- nestedTuple
# to be continued...
See tests/theTest.nim for more usages.
Following features are implemented but the syntax or their actual effects are still in question.
Like in Python (*a,b = range(5)
) and modern JavaScript let [a,...b] = someArray
, you can put the rest of the sequence into a new sequence. I haven't decided on the actual prefix to use yet, but I am settling on *
as used in Python for now. If you have better ideas, please start an issue to discuss other options.
import unpack
let mamaHen = @[3, 4, 5, 6, 7]
[a, b, *sneakyFox] <- mamaHen
# is expanded into:
# let
# a = mamaHen[0]
# b = mamaHen[1]
# sneakyFox = mamaHen[2..^1]
assert(sneakyFox == @[5, 6, 7])
[*sloppySavior, e] <- sneakyFox
assert(sloppySavior == @[5, 6])
# Perhaps the variable naming may be a bit mis-leading,
# since mamaHen[x..y] creates a new sequence and copy the slice into it,
# so rather than stealing, the sneakyFox actually cloned(?) whatever mamaHen had with her
# You can use *_ to skip the beginning
[*_, pickyFox] <- mamaHen
assert(pickyFox == 7)
# It's okay to take the middle chunk too.
[f, g, *randomFox, _, h] <- mamaHen
assert([f, g, h] == [3, 4, 7])
assert(randomFox == @[5])
# Due to restriction from nim's grammar, `*` following `var`
# is not allowed. Adding `_ as` before it is the current hack I chose to bypass this.
[var _ as *boldFox, i, j] <- mamaHen
assert([i, j] == [6, 7])
assert(boldFox == @[3, 4, 5])
# They are indeed created with var.
i = 12
boldFox[2] = 123
assert(i == 12)
assert(boldFox == @[3, 123, 5])
Under the hood, unpack
just attaches [countFromStart..^countFromEnd]
to whatever you throw at it, so anything that has slice operator implemented should work. Which also brings us to our first caveat.
Unless you implement the ..
operator (and its friends) yourself though.
[*a, *b, c] <- someSeq
is not allowed. It might be possible, but I think it will be really messy (plus I am lazy). Same restriction applies to both Python and JavaScript, so I think it's okay to skip this part for now.
However, using rest operator in different parts of the nested sequence is fine, since they have different index counters, so this will work:
[[*a, b], [c, d, *e], *f, g, h] <- someNestedSeq
# is expanded into:
# let
# b = someNestedSeq[0][^1]
# a = someNestedSeq[0][0..^2]
# c = someNestedSeq[1][0]
# d = someNestedSeq[1][1]
# e = someNestedSeq[1][2..^1]
# h = someNestedSeq[^1]
# g = someNestedSeq[^2]
# f = someNestedSeq[2..^3]
Since we have no way to know the sequence length at compile time, (well, at least I don't know a way). We can't know if you are trying to do something goofy like:
[a, b, *c, d, e] <- @[1,2,3]
Yes, I also wanted to have the natural [let x, y] <- someSeq
syntax for defining new symbol with let, [x, y] <- someSeq
for assignment, but the compiler deems it illegal. I ended up settle with more verbose assignment syntax since I anticipate it being used less often.
- Docs
- Maybe we can also support tables?
- More informative error message for out of bound sequence access during unpacking.
- rest operator for objects/tables.
Especially if you know how to make this macro easier to use. Also, if you know any other existing package that does this kind of stuff better, please let me know, thank you.