Nim v2.0 released

01 August 2023 The Nim Team

The Nim team is proud and happy to announce Nim version 2.0.

This is an evolution (not revolution) of Nim, bringing ORC memory management as a default, along with many other new features and improvements.

Nim is a programming language that is good for everything, but not for everybody. It focuses on the imperative programming paradigm and enhances it with a macro system. Its customizable memory management makes it well suited for unforgiving domains such as hard realtime systems and system programming in general.

Installing Nim 2.0

New users

Check out if the package manager of your OS already ships version 2.0 or install it as described here.

Existing users

If you have installed a previous version of Nim using choosenim, getting Nim 2.0 is as easy as:

$ choosenim update stable

Alternatively, you can download Nim 2.0 from our nightlies builds.

Donating to Nim

We would like to encourage you to donate to Nim. The donated money will be used to further improve Nim by creating bounties for the most important bugfixes and features.

You can donate via:

If you are a company, we also offer commercial support.

New features

Better tuple unpacking

Tuple unpacking for variables is now treated as syntax sugar that directly expands into multiple assignments. Along with this, tuple unpacking for variables can now be nested.

proc returnsNestedTuple(): (int, (int, int), int, int) = (4, (5, 7), 2, 3)

# Now nesting is supported!
let (x, (_, y), _, z) = returnsNestedTuple()

Improved type inference

A new form of type inference called top-down inference has been implemented for a variety of basic cases.

For example, code like the following now compiles:

let foo: seq[(float, byte, cstring)] = @[(1, 2, "abc")]

Forbidden Tags

Tag tracking now supports the definition of forbidden tags by the .forbids pragma which can be used to disable certain effects in proc types.

For example:

type IO = object ## input/output effect
proc readLine(): string {.tags: [IO].} = discard
proc echoLine(): void = discard

proc no_IO_please() {.forbids: [IO].} =
  # this is OK because it didn't define any tag:
  echoLine()
  # the compiler prevents this:
  let y = readLine()

New standard library modules

The famous os module got an overhaul. Several of its features are available under a new interface that introduces a Path abstraction. A Path is a distinct string, which improves the type safety when dealing with paths, files and directories.

Use:

  • std/oserrors for OS error reporting.
  • std/envvars for environment variables handling.
  • std/paths for path handling.
  • std/dirs for directory creation/deletion/traversal.
  • std/files for file existence checking, file deletions and moves.
  • std/symlinks for symlink handling.
  • std/appdirs for accessing configuration/home/temp directories.
  • std/cmdline for reading command line parameters.

Overloadable enums

Overloadable enums are no longer experimental.

For example:

type
  E1 = enum
    value1, value2
  E2 = enum
    value1, value2 = 4

const
  Lookuptable = [
    E1.value1: "1",
    value2: "2"
  ]

The types E1 and E2 share the names value1 and value2. These are overloaded and the usual overload disambiguation is used so that the E1 or E2 prefixes can be left out in many cases. These features are most beneficial for independently developed libraries.

Default values for objects

Inside an object declaration, fields can now have default values:

type
  Rational* = object
    num: int = 0
    den: int = 1

var r = Rational()
assert $r == "(num: 0, den: 1)"

These default values are used when the field is not initialized explicitly. See also default values for object fields for details.

Definite assignment analysis

We found Nim’s default initialization rule to be one major source of bugs. There is a new experimental switch called strictDefs that protects against these bugs. When enabled, it is enforced that a variable has been given a value explicitly before the variable can be used:

{.experimental: "strictDefs".}

proc main =
  var r: Rational
  echo r # Warning: use explicit initialization of 'r' for clarity [Uninit]

main()

To turn the warning into an error, use --warningAsError:Uninit:on on the command line.

The analysis understands basic control flow so the following works because every possible code path assigns a value to r before it is used:

{.experimental: "strictDefs".}

proc main(cond: bool) =
  var r: Rational
  if cond:
    r = Rational(num: 3, den: 3)
  else:
    r = Rational()
  echo r

main(false)

Even better, this feature works with let variables too:

{.experimental: "strictDefs".}

proc main(cond: bool) =
  let r: Rational
  if cond:
    r = Rational(num: 3, den: 3)
  else:
    r = Rational()
  echo r

main(false)

It is checked that every let variable is assigned a value exactly once.

Strict effects

--experimental:strictEffects are now always enabled. Strict effects require callback parameters to be annotated with effectsOf:

func sort*[T](a: var openArray[T],
              cmp: proc (x, y: T): int {.closure.},
              order = SortOrder.Ascending) {.effectsOf: cmp.}

The meaning here is that sort has the effects of cmp: sort can raise the exceptions of cmp.

Improved error message for type mismatch

proc foo(s: string) = discard
proc foo(x, y: int) = discard
proc foo(c: char) = discard

foo 4

produces:

temp3.nim(11, 1) Error: type mismatch
Expression: foo 4
  [1] 4: int literal(4)

Expected one of (first mismatch at [position]):
[1] proc foo(c: char)
[1] proc foo(s: string)
[2] proc foo(x, y: int)

Consistent underscore handling

The underscore identifier (_) is now generally not added to a scope when used as the name of a definition. While this was already the case for variables, it is now also the case for routine parameters, generic parameters, routine declarations, type declarations, etc. This means that the following code now does not compile:

proc foo(_: int): int = _ + 1
echo foo(1)

proc foo[_](t: typedesc[_]): seq[_] = @[default(_)]
echo foo[int]()

proc _() = echo "_"
_()

type _ = int
let x: _ = 3

Whereas the following code now compiles:

proc foo(_, _: int): int = 123
echo foo(1, 2)

proc foo[_, _](): int = 123
echo foo[int, bool]()

proc foo[T, U](_: typedesc[T], _: typedesc[U]): (T, U) = (default(T), default(U))
echo foo(int, bool)

proc _() = echo "one"
proc _() = echo "two"

type _ = int
type _ = float

JavaScript codegen improvement

The JavaScript backend now uses BigInt for 64-bit integer types (int64 and uint64) by default. As this affects JS code generation, code using these types to interface with the JS backend may need to be updated. Note that int and uint are not affected.

For compatibility with platforms that do not support BigInt and in the case of potential bugs with the new implementation, the old behavior is currently still supported with the command line option --jsbigint64:off.

Docgen improvements

Markdown is now the default markup language of doc comments (instead of the legacy RstMarkdown mode). In this release we begin to separate RST and Markdown features to better follow the specification of each language, with the focus on Markdown development. See also the docs.

  • Added a {.doctype: Markdown | RST | RstMarkdown.} pragma allowing to select the markup language mode in the doc comments of the current .nim file for processing by nim doc:

    1. Markdown (default) is basically CommonMark (standard Markdown) + some Pandoc Markdown features + some RST features that are missing in our current implementation of CommonMark and Pandoc Markdown.
    2. RST closely follows the RST spec with few additional Nim features.
    3. RstMarkdown is a maximum mix of RST and Markdown features, which is kept for the sake of compatibility and ease of migration.

  • Added separate md2html and rst2html commands for processing standalone .md and .rst files respectively (and also md2tex/rst2tex).

  • Added Pandoc Markdown bracket syntax [...] for making anchor-less links.

  • Docgen now supports concise syntax for referencing Nim symbols: instead of specifying HTML anchors directly one can use original Nim symbol declarations (adding the aforementioned link brackets [...] around them).
    • To use this feature across modules, a new importdoc directive was added. Using this feature for referencing also helps to ensure that links (inside one module or the whole project) are not broken.

  • Added support for RST & Markdown quote blocks (blocks starting with >).

  • Added a popular Markdown definition lists extension.

  • Added Markdown indented code blocks (blocks indented by >= 4 spaces).

  • Added syntax for additional parameters to Markdown code blocks:

    nim test="nim c $1" ...

C++ interop enhancements

Nim 2.0 takes C++ interop to the next level with the new virtual pragma and the extended constructor pragma. Now one can define constructors and virtual procs that map to C++ constructors and virtual methods, allowing one to further customize the interoperability. There is also extended support for the codeGenDecl pragma, so that it works on types.

It’s a common pattern in C++ to use inheritance to extend a library. Some even use multiple inheritance as a mechanism to make interfaces.

Consider the following example:

struct Base {
  int someValue;
  Base(int inValue)  {
    someValue = inValue;
  };
};

class IPrinter {
public:
  virtual void print() = 0;
};

type
  Base* {.importcpp, inheritable.} = object
    someValue*: int32
  IPrinter* {.importcpp.} = object

const objTemplate = """
  struct $1 : public $3, public IPrinter {
    $2
  };
""";

type NimChild {.codegenDecl: objTemplate.} = object of Base

proc makeNimChild(val: int32): NimChild {.constructor: "NimClass('1 #1) : Base(#1)".} =
  echo "It calls the base constructor passing " & $this.someValue
  this.someValue = val * 2 # Notice how we can access `this` inside the constructor. It's of the type `ptr NimChild`.

proc print*(self: NimChild) {.virtual.} =
  echo "Some value is " & $self.someValue

let child = makeNimChild(10)
child.print()

It outputs:

It calls the base constructor passing 10
Some value is 20

ARC/ORC refinements

With the 2.0 release, the ARC/ORC model got refined once again and is now finally complete:

  1. Programmers now have control over the “item was moved from” state as =wasMoved is overridable.
  2. There is a new =dup hook which is more efficient than the old combination of =wasMoved(tmp); =copy(tmp, x) operations.
  3. Destructors now take a parameter of the attached object type T directly and don’t have to take a var T parameter.

With these important optimizations we improved the runtime of the compiler and important benchmarks by 0%! Wait … what? Yes, unfortunately it turns out that for a modern optimizer like in GCC or LLVM there is no difference.

But! This refined model is more efficient once separate compilation enters the picture. In other words, as we think of providing a stable ABI it is important not to lose any efficiency in the calling conventions.

Tool changes

  • Nim now ships Nimble version 0.14 which added support for lock-files. Libraries are stored in $nimbleDir/pkgs2 (it was $nimbleDir/pkgs before). Use nimble develop --global to create an old style link file in the special links directory documented here.

  • nimgrep now offers the option --inContext (and --notInContext), which allows to filter only matches with the context block containing a given pattern.

  • nimgrep: names of options containing “include/exclude” are deprecated, e.g. instead of --includeFile and --excludeFile we have --filename and --notFilename respectively. Also, the semantics are now consistent for such positive/negative filters.

  • Nim now ships with an alternative package manager called Atlas. More on this in upcoming versions.

Porting guide

Block and Break

Using an unnamed break in a block is deprecated. This warning will become an error in future versions! Use a named block with a named break instead. In other words, turn:

block:
  a()
  if cond:
    break
  b()

Into:

block maybePerformB:
  a()
  if cond:
    break maybePerformB
  b()

Strict funcs

The definition of "strictFuncs" was changed. The old definition was roughly: “A store to a ref/ptr deref is forbidden unless it’s coming from a var T parameter”. The new definition is: “A store to a ref/ptr deref is forbidden”.

This new definition is much easier to understand, but the price is some expressibility. The following code used to be accepted:

{.experimental: "strictFuncs".}

type Node = ref object
  s: string

func create(s: string): Node =
  result = Node()
  result.s = s # store to result[]

Now it has to be rewritten to:

{.experimental: "strictFuncs".}

type Node = ref object
  s: string

func create(s: string): Node =
  result = Node(s: s)

Standard library

Several standard library modules have been moved to nimble packages, use nimble or atlas to install them:

  • std/punycode => punycode
  • std/asyncftpclient => asyncftpclient
  • std/smtp => smtp
  • std/db_common => db_connector/db_common
  • std/db_sqlite => db_connector/db_sqlite
  • std/db_mysql => db_connector/db_mysql
  • std/db_postgres => db_connector/db_postgres
  • std/db_odbc => db_connector/db_odbc
  • std/md5 => checksums/md5
  • std/sha1 => checksums/sha1
  • std/sums => sums