nim-wiki/Nim-for-Python-Programmers.md

---

DISCLAIMER!
Unofficial, work in progress! This is still a stub. Please help extending it. There may be inaccuracies in this guide.

---

This is a guide for people with experience in Python or a similar language. The guide assumes some intermediate knowledge.

The general tutorials can be found here:
http://nim-lang.org/docs/tut1.html
http://nim-lang.org/docs/tut2.html

The manual provides a more or less complete overview of the language: http://nim-lang.org/docs/manual.html

At a glance

Similarities and differences.

Feature	Python	Nim
Execution model	Virtual Machine, JIT	Machine code via C*
Meta-programming	Python (decorators/metaclasses/eval)	Nim (const/when/template/macro)
Memory Management	Garbage-collected	Garbage-collected and manual
Types	Dynamic	Static
Dependent types	-	Partial support
Generics	Duck typing	Yes
int8/16/32/64 types	No	Yes
Bigints (arbitrary size)	Yes (transparently)	Yes (via nimble package)
Arrays	Yes	Yes
Bounds-checking	Yes	Yes
Type inference	Duck typing	Yes (extensive support)
Closures	Yes	Yes
Operator Overloading	Yes	Yes (on any type)
Custom Operators	No	Yes
Object-Oriented	Yes	Minimalistic**
Methods	Yes	Yes
Multi-Methods	No	Yes
Exceptions	Yes	Yes

*Other backends supported and/or planned
**See section below.

Philosophy

The key to understanding Nim is that Nim was designed to be as fast as C, but to be much safer. Many of the design-decisions are based on making it harder to shoot yourself in the foot. In Python there are no pointers (everything is treated as a reference). While Nim does give you pointers, Nim gives you other, safer tools for your everyday needs, while pointers are mostly reserved for interfacing with C and doing low-level system programming.

Contrarily to Python, most Nim code can be executed at compile time to perform meta-programming. You can do a lot of the dsl-s possible with Python decorators / metaprogramming with Nim macros and pragmas. (And some stuff that you can't!). Of course this requires some different patterns and more type safety.

Arrays

In Nim, arrays are much more strict. They are pass-by-value (meaning they're copied at assignment). When passing an array to a proc in Nim, the argument is a read-only reference, meaning it can't be assigned to. Take the following example:

Nim:

proc foobar(z: array[0..3, int]) = 
  z[5] = 5  # Error: Cannot assign to z
  echo z[5] # Error: Index out of bounds.

var x = [1, 2, 3, 4]

foobar(x)

The Nim code will not compile. If you mean to change the array that was passed to the procedure, you can change the the signature of the procedure to proc foobar(z: var array[0..3, int]). Now you will only get index out of bounds error. If you change the index in both lines to 1, the code will compile. If the index is a variable, Nim will include run-time checks on the bounds of the array.

In C, you can pass an int[3] to the foobar function, and the compiler will not complain. In this case Nim would not compile. You can use an openarray to accept an array of any size, and you can use low(z) and high(z) to query the bounds of the array.

Nim arrays can also be indexed from any number. That is, z: array[1..4, int] is an array of int indexed from 1 to 4. Trying to access z[0] would throw an index out bounds error.

In Nim you are strongly discouraged from using pointers, and you can accomplish almost everything you'd otherwise use pointers for with normal arguments, "var" arguments, variables, and "ref".

Object-Orientation

Objects in Nim behave quite differently from classes in Python. Objects support inheritance (not multiple inheritance). Python-like object methods do not exist: procs are not bound to objects (however, you can use them in ways very similar to Python, see the example below)

You can call a proc on objects with the anObject.foobar(), but you can do that on any type (e.g. ints and arrays) as well. You can have methods on object, but you can have methods on any types, and for all the arguments, not just the first (in C++, implicit) one.

Nim does not have an implicit self.

It is possible to implement object-orientation features through libraries, thanks to the extensive meta-programming features of Nim. These are at the moment mostly work-in-progress.

Python-like object orientation example:

type Animal = ref object of RootObj
  age: int
  name: string

type Cat = ref object of Animal
  playfulness: float

# A proc that can access and *modify* the object
proc increase_age(self: Cat) =
  self.age.inc()

var c = Cat(name: "Tom")
c.increase_age()
echo c.name, " ", c.age

Tip: use objects by reference

Inheritance

A Python programmer when first trying Nim can try to write something like this:

# TODO: extend this example with the declaration of the objects w/o ref
var heterogeneousArray: seq[ParentObjectType] = @[]
# TODO: Add an object inheriting from ParentObjectType to the seq

That will return a compilation error, because the objects have different sizes. Like in C, you can't make an array of structs with disparate types and sizes. What one can do is an array of references to the objects, all references having the same size but pointing to heterogeneous objects. This is what Python does by default.

# TODO: Show the proper way to do things, with emphasis on what changed

So, when using inheritance, be sure to inherit using refs, unless you know what you are doing. Also, check out Nim's generic programming capabilities as an alternative to OOP and inheritance.

From Python to Nim

Ranges

In Python, simple integer for loops use the range generator function. For the 1- and 2- argument forms of this function, nim's .. iterator works almost the same way:

for i in ..10:
  echo i  # Prints 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10

for i in 5..10:
  echo i  # Prints 5, 6, 7, 8, 9, 10

Note that the .. operator includes the end of the range, whereas Python's range(a, b) does not include b. If you prefer this behavior, use the ..< iterator instead:

for i in ..<10:
  echo i  # Prints 0, 1, 2, 3, 4, 5, 6, 7, 8, 9

Python's range() also has an optional third parameter, which is the value to increment by each step, which can be positive or negative. If you need this behavior, use the countup or countdown iterators:

for i in countup(1, 10, 2):
  echo i  # Prints 1, 3, 5, 7, 9

Python slices

The syntax for slice ranges is different. a[x:y] becomes a[x..<y]. Also, a[x..y] is inclusive.

let a = @[1,2,3,4]
a[0..0]  # returns @[1]
a[0..1]  # returns @[1, 2]
a[0..<2] # returns @[1, 2]
a[0..3]  # returns @[1, 2, 3, 4]

Python strings

Use double quotes: "foo" or """foo""", not 'foo'

Strings are always unicode. Remember to use runes() to iterate over unicode characters instead of bytes.

var a = "hello"
echo a[0..1]  # returns "he": ranges are inclusive! See the "Ranges" paragraph

A very detailed comparison

Python tuples

Nim Tuples are a lot like Python namedtuples in that the tuple members must have names in the type specification. However, they can be constructed and destructured without naming the fields. Also, you can have anonymous tuple types. See manual for a more in depth look at tuples.

var a = ("hi", "there")
echo a[0]  # hi

proc swap(tup: tuple[a: int, b: float]): tuple[c: float, d: int] =
  (tup.b, tup.a)

echo swap((1, 3.14))  # (c: 3.14, d:1)

let c, d = swap(8, 4.2)

In some cases where you might have used a tuple in Python, Nim Arrays can be used if you don't care what the fields are named. However, unlike tuples, arrays can only have one element type:

var a = ["hi", "there"]
var b = [1, 2, 3]
echo b[1]  # 2
var c = ["hi", 1]  # Error: type mismatch (no mixed types allowed in arrays)

Python lists

Use Nim sequences

Nim arrays and sequences can hold only items of the same type.

List Comprehensions

import future  # required for the list comprehension syntactical sugar

let x = @[1, 2, 3, 4, 5, 8, 8, 8, 10]  # create a Nim sequence

# syntax: lc[ item | ( item <- seq, (optional condition) ), type ]
# equivalent to python [ y for y in x if y > 5 ]
echo lc[ y | ( y <- x, y > 5 ), int ]
# Can also perform operations on item, e.g. y*2 as shown here
echo lc[ y*2 | ( y <- x ), int ]

Python sets

Python sets are not like Nim set type. If the values that will go in the set are known beforehand and finite, you can create an Enum for them. Otherwise you can emulate a Python set using a HashSet. The Nim set type is faster and memory-efficient. In fact, set is implemented with a bit vector, whereas HashSet is implemented as a dictionary. For simple flag types and small mathematical sets, use set.

Dictionaries

Use Tables for Python dicts, OrderedTable for Python ordered dicts, and CountTable for Python's Counter.

import tables

var a = {"hi": 1, "there": 2}.toTable
echo a["hi"], " ", a.len
assert a.hasKey("hi")

for key, value in a:
  echo key, " " ,value

ABC

See this

Ternary operator

var is_happy = if has_cat(): "quite" else: "not much"

Reading files line by line

for line in lines("yourfile.txt"):
  echo line

map and lambda

import sequtils  # for map
let temperatures = map(locations, getTemperature)
let temperatures = map(locations, proc (loc: Location): float = getTemperature(loc, scale=C))

See manual