Haskell's semantics are historically based on those of the Miranda programming language, which served to focus the efforts of the initial Haskell working group. The last formal specification of the language was made in July 2010, while the development of GHC's implementation has continued to extend Haskell via language extensions. The next formal specification is planned for 2020.
Haskell is used in academia and industry. As of September 2019, Haskell was the 23rd most popular programming language in terms of Google searches for tutorials and made up less than 1% of active users on the GitHub source code repository.
The first version of Haskell ("Haskell 1.0") was defined in 1990. The committee's efforts resulted in a series of language definitions (1.0, 1.1, 1.2, 1.3, 1.4).
In late 1997, the series culminated in Haskell 98, intended to specify a stable, minimal, portable version of the language and an accompanying standard library for teaching, and as a base for future extensions. The committee expressly welcomed creating extensions and variants of Haskell 98 via adding and incorporating experimental features.
In February 1999, the Haskell 98 language standard was originally published as The Haskell 98 Report. In January 2003, a revised version was published as Haskell 98 Language and Libraries: The Revised Report. The language continues to evolve rapidly, with the Glasgow Haskell Compiler (GHC) implementation representing the current de facto standard.
In early 2006, the process of defining a successor to the Haskell 98 standard, informally named Haskell Prime, began. This was intended to be an ongoing incremental process to revise the language definition, producing a new revision up to once per year. The first revision, named Haskell 2010, was announced in November 2009 and published in July 2010.
Haskell 2010 is an incremental update to the language, mostly incorporating several well-used and uncontroversial features previously enabled via compiler-specific flags.
Hierarchical module names. Module names are allowed to consist of dot-separated sequences of capitalised identifiers, rather than only one such identifier. This lets modules be named in a hierarchical manner (e.g., Data.List instead of List), although technically modules are still in a single monolithic namespace. This extension was specified in an addendum to Haskell 98 and was in practice universally used.
The foreign function interface (FFI) allows bindings to other programming languages. Only bindings to C are specified in the Report, but the design allows for other language bindings. To support this, data type declarations were permitted to contain no constructors, enabling robust nonce types for foreign data that could not be constructed in Haskell. This extension was also previously specified in an Addendum to the Haskell 98 Report and widely used.
So-called n+k patterns (definitions of the form fact (n+1) = (n+1) * fact n) were no longer allowed. This syntactic sugar had misleading semantics, in which the code looked like it used the (+) operator, but in fact desugared to code using (-) and (>=).
The rules of type inference were relaxed to allow more programs to type check.
Some syntax issues (changes in the formal grammar) were fixed: pattern guards were added, allowing pattern matching within guards; resolution of operator fixity was specified in a simpler way that reflected actual practice; an edge case in the interaction of the language's lexical syntax of operators and comments was addressed; and the interaction of do-notation and if-then-else was tweaked to eliminate unexpected syntax errors.
The LANGUAGEpragma was specified. By 2010 dozens of extensions to the language were in wide use, and GHC (among other compilers) provided the LANGUAGE pragma to specify individual extensions with a list of identifiers. Haskell 2010 compilers are required to support the Haskell2010 extension, and are encouraged to support several others which correspond to extensions added in Haskell 2010.
moduleMain(main)where-- not needed in interpreter, is the default in a module filemain::IO()-- the compiler can infer this type definitionmain=putStrLn"Hello, World!"
The factorial function in Haskell, defined in a few different ways:
-- Type annotation (optional, same for each implementation)factorial::(Integrala)=>a->a-- Using recursion (with the "ifthenelse" expression)factorialn=ifn<2then1elsen*factorial(n-1)-- Using recursion (with pattern matching)factorial0=1factorialn=n*factorial(n-1)-- Using recursion (with guards)factorialn|n<2=1|otherwise=n*factorial(n-1)-- Using a list and the "product" functionfactorialn=product[1..n]-- Using fold (implements "product")factorialn=foldl(*)1[1..n]-- Point-free stylefactorial=foldr(*)1.enumFromTo1
As the Integer type has arbitrary-precision, this code will compute values such as factorial 100000 (a 456,574-digit number), with no loss of precision.
An implementation of an algorithm similar to quick sort over lists, where the first element is taken as the pivot:
-- Type annotation (optional, same for each implementation)quickSort::Orda=>[a]->[a]-- Using list comprehensionsquickSort=-- The empty list is already sortedquickSort(x:xs)=quickSort[a|a<-xs,a<x]-- Sort the left part of the list++[x]++-- Insert pivot between two sorted partsquickSort[a|a<-xs,a>=x]-- Sort the right part of the list-- Using filterquickSort=quickSort(x:xs)=quickSort(filter(<x)xs)++[x]++quickSort(filter(>=x)xs)
Jhc, a Haskell compiler written by John Meacham, emphasizes speed and efficiency of generated programs and exploring new program transformations.
Ajhc is a fork of Jhc.
The Utrecht Haskell Compiler (UHC) is a Haskell implementation from Utrecht University. It supports almost all Haskell 98 features plus many experimental extensions. It is implemented using attribute grammars and is currently used mostly for research on generated type systems and language extensions.
Implementations no longer actively maintained include:
The Haskell User's Gofer System (Hugs) is a bytecode interpreter. It was once one of the implementations used most widely, alongside the GHC compiler, but has now been mostly replaced by GHCi. It also comes with a graphics library.
HBC is an early implementation supporting Haskell 1.4. It was implemented by Lennart Augustsson in, and based on, Lazy ML. It has not been actively developed for some time.
nhc98 is a bytecode compiler focusing on minimizing memory use.
Implementations not fully Haskell 98 compliant, and using a variant Haskell language, include:
Gofer was an educational dialect of Haskell, with a feature called constructor classes, developed by Mark Jones. It was supplanted by Hugs (Haskell User's Gofer System).
Helium, a newer dialect of Haskell. The focus is on making learning easier via clearer error messages. It currently lacks full support for type classes, rendering it incompatible with many Haskell programs.
Ampersand, an open source language and toolchain for building information systems with reactive behaviour, is being used by Ordina and TNO (both in the Netherlands) for purposes such as conceptual analysis, analysis of archimate-models, legal analysis and prototyping large information systems. The compiler is written in Haskell. Web applications realised with Ampersand comprise the Semantic Treehouse and RAP3.
Bluespec SystemVerilog (BSV) is a language for semiconductor design that is an extension of Haskell. Also, Bluespec, Inc.'s tools are implemented in Haskell.
Cryptol, a language and toolchain for developing and verifying cryptography algorithms, is implemented in Haskell.
GitHub implemented Semantic, an open-source library for analysis, diffing, and interpretation of untrusted source code, in Haskell.
Keera Studios writes mobile games and apps for both Android and iOS using Haskell.
seL4, the first formally verifiedmicrokernel, used Haskell as a prototyping language for the OS developer.:p.2 At the same time, the Haskell code defined an executable specification with which to reason, for automatic translation by the theorem-proving tool.:p.3 The Haskell code thus served as an intermediate prototype before final C refinement.:p.3
Swift Navigation, a high precision GPS manufacturer, implements significant portions of its product in Haskell, providing some open source software.
Jan-Willem Maessen, in 2002, and Simon Peyton Jones, in 2003, discussed problems associated with lazy evaluation while also acknowledging the theoretical motives for it. In addition to purely practical considerations such as improved performance, they note that, in addition to adding some performance overhead, lazy evaluation makes it more difficult for programmers to reason about the performance of their code (particularly its space use).
Bastiaan Heeren, Daan Leijen, and Arjan van IJzendoorn in 2003 also observed some stumbling blocks for Haskell learners: "The subtle syntax and sophisticated type system of Haskell are a double edged sword – highly appreciated by experienced programmers but also a source of frustration among beginners, since the generality of Haskell often leads to cryptic error messages." To address these, researchers from Utrecht University developed an advanced interpreter called Helium, which improved the user-friendliness of error messages by limiting the generality of some Haskell features, and in particular removing support for type classes.
Ben Lippmeier designed Disciple as a strict-by-default (lazy by explicit annotation) dialect of Haskell with a type-and-effect system, to address Haskell's difficulties in reasoning about lazy evaluation and in using traditional data structures such as mutable arrays. He argues (p. 20) that "destructive update furnishes the programmer with two important and powerful tools ... a set of efficient array-like data structures for managing collections of objects, and ... the ability to broadcast a new value to all parts of a program with minimal burden on the programmer."
Robert Harper, one of the authors of Standard ML, has given his reasons for not using Haskell to teach introductory programming. Among these are the difficulty of reasoning about resource use with non-strict evaluation, that lazy evaluation complicates the definition of data types and inductive reasoning, and the "inferiority" of Haskell's (old) class system compared to ML's module system.
Haskell's build tool, Cabal, has historically been criticised for poorly handling multiple versions of the same library, a problem known as "Cabal hell". The Stackage server and Stack build tool were made in response to these criticisms. Cabal itself now has a much more sophisticated build system, heavily inspired by Nix, which became the default with version 3.0.
Clean is a close, slightly older relative of Haskell. Its biggest deviation from Haskell is in the use of uniqueness types instead of monads for I/O and side-effects.
A series of languages inspired by Haskell, but with different type systems, have been developed, including:
Haskell has served as a testbed for many new ideas in language design. There have been many Haskell variants produced, exploring new language ideas, including:
Disciple, a strict-by-default (laziness available by annotation) dialect of Haskell that supports destructive update, computational effects, type directed field projections and allied functional aspects.
Hume, a strict functional language for embedded systems based on processes as stateless automata over a sort of tuples of one element mailbox channels where the state is kept by feedback into the mailboxes, and a mapping description from outputs to channels as box wiring, with a Haskell-like expression language and syntax.
^Syme, Don; Granicz, Adam; Cisternino, Antonio (2007). Expert F#. Apress. p. 2. F# also draws from Haskell particularly with regard to two advanced language features called sequence expressions and workflows.
^Lattner, Chris (3 June 2014). "Chris Lattner's Homepage". Chris Lattner. Retrieved 3 June 2014. The Swift language is the product of tireless effort from a team of language experts, documentation gurus, compiler optimization ninjas, and an incredibly important internal dogfooding group who provided feedback to help refine and battle-test ideas. Of course, it also greatly benefited from the experiences hard-won by many other languages in the field, drawing ideas from Objective-C, Rust, Haskell, Ruby, Python, C#, CLU, and far too many others to list.
^"Enpuzzled". Enpuzzled - Keera Studios. 21 March 2020. Retrieved 21 March 2020.
^"Magic Cookies". Magic Cookies! - Keera Studios. 21 March 2020. Retrieved 21 March 2020.
A formal proof of functional correctness was completed in 2009.
Klein, Gerwin; Elphinstone, Kevin; Heiser, Gernot; Andronick, June; Cock, David; Derrin, Philip; Elkaduwe, Dhammika; Engelhardt, Kai; Kolanski, Rafal; Norrish, Michael; Sewell, Thomas; Tuch, Harvey; Winwood, Simon (October 2009). "seL4: Formal verification of an OS kernel"(PDF). 22nd ACM Symposium on Operating System Principles. Big Sky, MT, USA.