Tips and tricks - nim-lang/Nim GitHub Wiki

Nim Tips and tricks

A mix of bits of code, cookbook recipes, deep knowledge, tips and tricks about Nim programming.

Compilation

Syntax

Check syntax

Use nim check myfile.nim to check your program for errors, without code generation. This makes the process quicker. Also, nim check displays more error messages than nim c and it can help finding the cause of an error.

Using reserved words as identifiers

You can use the slanted quote string syntax (stropping) if you need to use a reserved word as an identifier.

var `type`: int

Conditional compilation

Defining a compile time symbol

You want to define useSomeFeature at compilation time and use it in the code.

  • Do it when you compile: $ nim c -d:useSomeFeature myApp
  • Put it in the yourfilename.nims file located in the same directory as your source file yourfilename.nim as switch("define", "useSomeFeature")
  • Put it in the nim.cfg file located in the same directory as source file and write define:useSomeFeature 
when defined(useSomeFeature):
  const someSubFeature = true
  type idType = int32
  var defaultUserName = "Frank"
else:
  type idType = int16
  var defaultUserName = "Molly"

# `when` does not open a new scope, so `idType` and `defaultUserName` are available here.

when isMainModule:
  # this code is ran only when this module is the main compilation module
  echo "This module was compiled on ", CompileDate, " at ", CompileTime

Errors

Error: attempting to call routine: 'a'

Nim likes spaces. This is a quirk of the command syntax where a +b is evaluated as a(+(b)). Use spaces, preferably, like a + b or else a+b, but don't mix space and no-spaces usage.

Error: expression '...' has no type (or is ambiguous)

  • Have you tried to discard the value from a proc that does not return any value?
proc noReturnValue =
  echo "Foo!"

discard noReturnValue()

Error: '...' doesn't have a concrete type, due to unspecified generic parameters.

  • A inner proc "inherits" the including proc generics declaration and you don't need to re-define them.
proc foo[T](x: T) =
  proc bar[T](y: T) =    <== Error is here
    echo "In bar"
  echo "In foo"

You must write:

proc foo[T](x: T) =
  proc bar(y: T) =    # Generic type T is defined by englobing proc
    echo "In bar"
  echo "In foo"

Error: undeclared field: 'major' for type foo.Version

To access fields of a public object, these fields must be public too.

This is causing the error, where Version is defined in foo.nim:

type
  Version* = object
    major: int

You must write:

type
  Version* = object
    major*: int

Error: template instantiation too nested

Remember that templates replace template call with template body recursively. This can occur even in non-recursive templates in case of parameter clash. For instance:

template foo(body: untyped) =
  template bar(body: untyped) =
    ...

Here the body parameter is not seen both by foo and bar as different symbols. When foo instantiates, it will replace the occurrence in bar parameters. To prevent the symbol name clash, you must write:

template foo(fooBody: untyped) =
  template bar(barBody: untyped) =
    ...

Error: identifier expected, but found '...

See also Error: template instantiation too nested

Error: type mismatch: got <X> but expression is of type: X

This can happen when a type is defined twice, probably because of a file include. This can occur too when you create duplicated types in macro code.

Error: ambiguous call; both procName...

You have forward declared procName to define recursive macro procs but the compiler complains that it found two declarations of procName. Additionally, there's a hint that procName is declared but not used!

This happens because the forward declaration does not use the same parameter names as in the proc definition .

# Forward declaration of procName that will cause a compile error.
# Use argument `stmt` instead of `arg` to have it compile.
proc procName(arg: NimNode): NimNode

proc procName(stmt: NimNode): NimNode =
  ...

Tests

How to test that the compiler will reject a piece of code?

Use compiles.

import unittest

suite "valid syntax":
  test "syntax errors":
    let a = "foo"
    let b = 5
    assert not compiles(let c = a + b), "Can't mix string and int"

Enhance documentation with examples

You can provide examples of in your documentation, with runnableExamples. When the documentation is extracted from the source with $ nim doc myfile.nim, these examples are extracted and put in a temporary file, compiled and tested. This is a good way to provide examples for your API and check that they are still valid.

proc isOdd(i: int): bool =
  ## Test if its argument is odd.
  runnableExamples:
    assert 5.isOdd, "5 is an odd number"
    assert not isOdd(3 * 2), "6 is not an odd number"

  result = i mod 2 == 1

Templates

Templates overloading

Like proc or funct, template follow overloading rules. You can use normal types for parameters but they can have meta-types too: typed, untyped and typedesc.

There are lazy type resolution rules for untyped parameters.

For typed parameters, you can use parameter constraints to reduce the scope of the parameters.

template foo(x: typed{lit}) =
  echo "foo matched the literal '", x, "'"

template foo(x: typed{ident}) =
  echo "foo matched the identifier '", x, "'"

Template scopes

You have defined a dsl: and you want to define an options: template that can be used only within dsl:. How do you do this? Use block: scopes and declaredInScope, like following:

template dsl*(body: untyped) =
  block:
    const inDsl {.inject, used.} = true
    # Here define the dsl code...

template options(body: untyped) =
  when not declaredInScope(inDsl):
    {.fatal: "`options` can be used only in `dsl` block".}
    # Here define the options code...

Don't try to create englobing templates! Templates are text code rewriting macros and you'll get unexpected results...

Macros

Dump macro expansion

expandMacros shows how a macro call is expanded at compilation time. It's very useful for debugging macros...

dumpTree can also be used to print how a block of code is parsed at compile time.

import macros

let x, y = (4, 8)
expandMacros:
  echo "x > y is expanded to ", x > y
dumpTree:
  echo "x > y is expanded to ", x > y

Output

Echo to stderr

echo writes to stdout. If you want to print to stderr use write and writeLine

writeLine(stderr, "Write to stderr")

How do I echo in a {.noSideEffect.} proc?

You can use debugEcho for this.

Enums

Converting string to enum

Strings can be converted to enum using parseEnum from strutils. Beware that a ValueError is raised if the enum is not found. If you don't want to manage the exception, provide a default with parseEnum.

import strutils

type
  Fruit = enum
    Apple,
    Banana,
    Cherry

  let fruit = parseEnum[Fruit]("cherry")
  let pineapple = parseEnum[Fruit]("pineapple", Banana)

Tuple

Enumerate the fields of a tuple or object.

fields and fieldPairs will help you discover the fields and values of a tuple or object. But as Nim is strongly typed, you can't build a result mixing types. Either use conditional compilation for different code paths or use overloading.

type
  Custom = object
    foo: string
    bar: bool

proc `$`(x: Custom): string =
  result = "Custom:"
  for name, value in x.fieldPairs:
    when value is bool:
      result.add("\n\t" & name & " is " & $value)
    else:
      result.add("\n\t" & name & " \'" & value & "\'")

proc `$1`(x: string): string = $x
proc `$1`(x: bool): string = $x

proc `$1`(x: Custom): string =
  result = "Custom:"
  for name, value in x.fieldPairs:
    result.add("\n\t" & name & " is " & `$1`(value))

let o = Custom(foo: "Hi there!", bar: false)

echo "o=", $o
echo "o1=", `$1`(o)

Seq and lists

Heterogeneous lists

Nim being strongly typed, you can't mix multiple types in a seq. andrea came out with heterogeneous lists that fit that special need:

# hlist.nim

type
  HNil* = object
  HCons*[H, T] = ref object
    h: H
    t: T
  HList = HNil or HCons

let hNil* = HNil()

proc cons*[H; T: HList](hd: H; tl:T): auto =
  HCons[H, T](h: hd, t: tl)

template `<>`*(hd, tl: untyped): untyped = cons(hd, tl)

proc head*[H, T](c: HCons[H, T]): H {.inline.} = c.h

proc tail*[H, T](c: HCons[H, T]): T {.inline.} = c.t

They can be used like in this short example where l contains a string and an int.

import hlist

proc printAll(n: HNil) =
  discard

proc printAll[H; T](hl: HCons[H, T]) =
  echo hl.head
  printAll(hl.tail)

let l = "hi" <> (2 <> hNil)
printAll(l)

Iterators

Recursive iterators

From a forum post demonstrating how to simulate recursive iterator by converting a recursive proc to generate all character permutations of a string. This macro does not support break or continue statement, but read the referred post to find more complete example.

import std/macros

macro toItr*(x: ForLoopStmt): untyped =
    ## Convert factory proc call for inline-iterator-like usage.
    ## E.g.: ``for a,b in toItr(myFactory(parm)): echo a,b``.
    let
        forVars = x[0..^3]
        call    = x[^2][1] # Get foo out of toItr(foo)
        body    = x[^1]
        itrSym  = ident"itr"
        #itrSym  = genSym(ident="itr")
    var forTree = nnkForStmt.newTree()  # for
    for v in forVars: forTree.add v     # for v1,...
    forTree.add(nnkCall.newTree(itrSym), body) # for v1,... in itr(): body
    result = quote do:
        block:
            let `itrSym` {.inject.} = `call`
            `forTree`
    #debugEcho result.repr # uncomment to print the generated code

when isMainModule:
    proc lex_perm[T: seq|string](s: T, idxStart=0): iterator(): (T,int) =
        var idx= idxStart
        result = iterator(): (T,int) =
            if s.len == 0: yield (s[0..<0], idx)
            for i in 0 ..< s.len:
                for s2,_ in toItr lex_perm(s[0 ..< i] & s[i+1 .. ^1], idx):
                    yield (s[i ..< i+1] & s2, idx)
                    inc idx
    
    for s,idx in toItr lex_perm("123", 1): echo idx,":  ",s

External libraries

Managing external libraries memory allocations

Some external libraries provide an API to give access to data structures that they allocated on their own. They also provide a reference counting process in order for the client code to manage when the library can release the memory or resources used by these data structures. The use of destructors and compiling with --gc:arc or --gc:orc can ease bridging to these API. In the following sample, apiX() are calls to the external library API.

type
  MyObject* = object
    ## The Nim proxy object that is used in Nim code
    apiObject: pointer
      ## The pointer to the library object

proc initMyObject*: MyObject =
  ## Get a new library object.
  result = MyObject(apiObject: apiGetObject(...))
  # We increment the reference counter for this object
  apiIncRefCounter(apiObject)

proc `=destroy`(o: var MyObject) =
  ## When a ``MyObject`` is destroyed in Nim code, we need to decrement
  ## the API reference counter of that object.
  if not isNil(o.apiObject):
    apiDecRefCounter(o.apiObject)
    o.apiObject = nil

proc `=`(a: var MyObject; b: MyObject) =
  # Do nothing for self-assignments:
  if a.apiObject == b.apiObject:
    return

  # Destroy what is allocated in a
  `=destroy`(a)

  # Copy new content to a and increment reference counter
  if not isNil(b.apiObject):
    a.apiObject = b.apiObject
    apiIncRefCounter(a.apiObject)

Don't forget to compile with --gc:arc or --gc:orc to enable destructors. Notice how MyObject has public visibility while its apiObject field is kept private.

Language details

Shallow vs. DeepCopy

import vs include

import imports the public symbols (the ones with * after the name) from the imported module. You can't access the non-public symbols. import lines can be used only at the top level of a module.

include replace the include line with the content of the included file. You can access public and non-public symbols as they are part of the code now. include is mainly used to break big Nim files into smaller ones. include statements are not limited to the top level of module. But include files can create strange error messages too (see Error: type mismatch: got <X> but expression is of type: X).

Optimization via C compiler flags (LTO, PGO, ...)

Clang

toast

⚠️ **GitHub.com Fallback** ⚠️