<chapter id="haskell98">
<title>Hugs <foreignphrase>vs</foreignphrase> Haskell 98 and addenda</title>

<para>
In <option>+98</option> mode, Hugs supports
<ulink url="http://www.haskell.org/definition/">Haskell 98</ulink>
and some standardized extensions
(described by addenda to the Haskell 98 report).
</para>

<sect1 id="bugs-haskell98">
<title>Haskell 98 non-compliance</title>

<para>
Hugs deviates from Haskell 98 in a few minor ways,
listed here corresponding to the relevant sections of the Report.
</para>

<sect2 id="bugs-lexical">
<title>Lexical structure</title>

<variablelist>

<varlistentry>
<term>Restricted character set</term>
<listitem><para>
The Haskell report specifies that programs may be written using Unicode.
Hugs only accepts the ISO8859-1 (Latin-1) subset at the moment.
</para></listitem>
</varlistentry>

<varlistentry>
<term>Floating point literals</term>
<listitem><para>
Hugs is confused by such things as
<quote><literal>0xy</literal></quote>,
<quote><literal>0oy</literal></quote>,
<quote><literal>9e+y</literal></quote> and
<quote><literal>9.0e+y</literal></quote>,
because it doesn't look far enough ahead.
</para></listitem>
</varlistentry>

</variablelist>

</sect2>

<sect2 id="bugs-syntax">
<title>Expressions</title>

<variablelist>

<varlistentry>
<term>Interaction of fixities with the
<literal>let</literal>/lambda meta-rule</term>
<listitem><para>
Hugs doesn't use the fixity of operators until after parsing, and so
fails to accept legal (but weird) Haskell 98 expressions like
<programlisting>
  let x = True in x == x == True
</programlisting>
</para></listitem>
</varlistentry>

<varlistentry>
<term>Restricted syntax for left sections</term>
<listitem><para>
In Hugs, the expression must be an fexp
(or <literal>case</literal> or <literal>do</literal>).
Legal expressions like <literal>(a+b+)</literal>) and
<literal>(a*b+)</literal>) are rejected.
</para></listitem>
</varlistentry>

</variablelist>

</sect2>

<sect2 id="bugs-types">
<title>Declarations and bindings</title>

<variablelist>

<varlistentry>
<term>Slight relaxation of polymorphic recursion</term>
<listitem><para>
Hugs's treatment of polymorphic recursion is less restrictive than Haskell 98
when the functions involved are mutually recursive.
Consider the following example:
<programlisting>
  data BalancedTree a = Zero a | Succ (BalancedTree (a,a))

  zig :: BalancedTree a -> a
  zig (Zero a) = a
  zig (Succ t) = fst (zag t)

  zag (Zero a) = a
  zag (Succ t) = snd (zig t)
</programlisting>
As with many operations on non-regular (or nested) types,
<literal>zig</literal> and <literal>zag</literal>
need to be polymorphic in the element type.
In Haskell 98, the bindings of the two functions are interdependent,
and thus constitute a single binding group.
When type inference is performed on this group,
<literal>zig</literal> may be used at different types,
because it has a user-supplied polymorphic signature.
However, <literal>zag</literal> may not, and the example is rejected,
unless we add an explicit type signature for <literal>zag</literal>.
(It could be argued that this is a bug in Haskell 98.)
</para>

<para>
In Hugs, the binding of <literal>zig</literal> depends on that of
<literal>zag</literal>, but not vice versa.
(The binding of <literal>zag</literal> is considered to depend only on
the explicit signature of <literal>zig</literal>.)
It is possible to infer a polymorphic type for <literal>zag</literal>,
and from that for <literal>zig</literal>.
This type matches the declared signature, so Hugs accepts this example.
</para></listitem>
</varlistentry>

<varlistentry>
<term>Relaxation of type classes</term>
<listitem><para>
Contrary to the the Report (4.3.1), Hugs allows the types of the member
functions of a class <replaceable>C</replaceable> <replaceable>a</replaceable>
to impose further constraints on <replaceable>a</replaceable>, as in
<programlisting>
  class Foo a where
    op :: Num a => a -> a -> a
</programlisting>
</para></listitem>
</varlistentry>

<varlistentry>
<term>Different implementation of the monomorphism restriction for top-level bindings</term>
<listitem><para>
For example, Hugs rejects the following example from the
Haskell 98 Report, 4.5.5:
<programlisting>
  module M where
    import List
    len1 = genericLength "Hello"
    len2 = (2*len1) :: Rational
</programlisting>
This module consists of two binding groups, containing <literal>len1</literal>
and <literal>len2</literal> respectively.
Type inference on the first (<literal>len1</literal>) triggers the monomorphism
restriction, so that <literal>len1</literal> is assigned the monomorphic type
<literal>(Num a => a)</literal>.
The next step differs between Haskell 98 and Hugs:
<itemizedlist>
<listitem><para>
In Haskell 98,
type inference is then performed on <literal>len2</literal>,
resolving the type variable <literal>a</literal>
to <literal>Rational</literal>, and the module is legal.
</para></listitem>

<listitem><para>
In Hugs, the defaulting rule is applied to <literal>len1</literal>,
instantiating the type variable <literal>a</literal> to
<literal>Integer</literal>.
Then type inference on <literal>len2</literal> fails.
</para></listitem>
</itemizedlist>
<!--
<http://www.mail-archive.com/haskell@haskell.org/msg05160.html>
-->
</para></listitem>
</varlistentry>

</variablelist>

</sect2>

<sect2 id="bugs-modules">
<title>Modules</title>

<variablelist>

<varlistentry>
<term>Implicit module header</term>
<listitem><para>
In Haskell 98, if the module header is omitted, it defaults to
<quote><literal>module Main(main) where</literal></quote>.
In Hugs it defaults to <quote><literal>module Main where</literal></quote>,
because many people test small modules without module headers.
</para></listitem>
</varlistentry>

<varlistentry>
<term>Implicit export list</term>
<listitem><para>
In Haskell 98, a missing export list means all names defined in the
current module.
In Hugs, it is treated as
<quote><literal>(module <replaceable>M</replaceable>)</literal></quote>,
where <replaceable>M</replaceable> is the current module.
This is almost the same, differing only when an imported module is aliased as
<replaceable>M</replaceable>.
</para></listitem>
</varlistentry>

<varlistentry>
<term>Type synonyms in export and import lists</term>
<listitem><para>
Hugs allows the <replaceable>T</replaceable><literal>(..)</literal>
syntax for type synonyms in export and import lists.
It also allows the form <replaceable>T</replaceable><literal>()</literal>
for type synonyms in import lists.
</para></listitem>
</varlistentry>

<varlistentry>
<term>Mutually recursive modules are not supported</term>
<listitem><para>
Note that although the Haskell 98 specification of the
<literal>Prelude</literal> and library modules is recursive,
Hugs achieves the same effect by putting most of these definitions
in a module <literal>Hugs.Prelude</literal> that these modules import.
</para></listitem>
</varlistentry>

<varlistentry>
<term>Weird treatment of <literal>(:)</literal></term>
<listitem><para>
The Hugs prelude exports <literal>(:)</literal> as if it were an identifier,
even though this is not permitted in user-defined modules.
This means that Hugs incorrectly rejects the following:
<programlisting>
  module Foo where
  import Prelude()
  cs = 'a':cs
</programlisting>
</para></listitem>
</varlistentry>

</variablelist>

</sect2>

<sect2 id="bugs-primitives">
<title>Predefined types and classes</title>

<variablelist>

<varlistentry>
<term>Unicode is not supported</term>
<listitem><para>
The type <literal>Char</literal> is limited to the ISO8859-1 subset of Unicode.
</para></listitem>
</varlistentry>

<varlistentry>
<term>Rational literals lose precision</term>
<listitem><para>
In Haskell 98, a floating point literal like <literal>1.234e-5</literal>
stands for <quote><literal>fromRational (1234 % 100000000)</literal></quote>.
In particular, if the literal is of <literal>Rational</literal> type,
the fraction is exact.
In Hugs such literals are stored as double precision floating point numbers.
If the literal is of <literal>Rational</literal> type,
it usually denotes the same number, but some precision may be lost.
</para></listitem>
</varlistentry>

<varlistentry>
<term>
Floating point values are printed differently
</term>
<listitem><para>
Haskell 98 specifies that <literal>show</literal> for floating point numbers
is the function <literal>Numeric.showFloat</literal>,
but Hugs uses an internal function with slightly different semantics.
</para></listitem>
</varlistentry>

<varlistentry>
<term>
Derived <literal>Read</literal> instances do not work for some infix
constructors.
</term>
<listitem><para>
</para></listitem>
</varlistentry>

<varlistentry>
<term>Derived instances for large tuples are not supplied</term>
<listitem><para>
In Haskell 98, all tuple types are instances of
<literal>Eq</literal>, <literal>Ord</literal>, <literal>Bounded</literal>,
<literal>Read</literal>, and <literal>Show</literal>
if all their component types are.
Hugs defines these instances only for tuple types of size 5 or less
(3 or less in the small Hugs configuration).
</para></listitem>
</varlistentry>

</variablelist>

</sect2>

</sect1>

<sect1 id="addenda">
<title>Addenda to Haskell 98</title>

<para>
These addenda describe extensions that have been standardized
across haskell implementations.
</para>

<sect2 id="ffi">
<title>Foreign Function Interface</title>

<para>
The Haskell Foreign Function Interface, as described in the
<ulink url="http://www.cse.unsw.edu.au/~chak/haskell/ffi/">FFI addendum</ulink>
is implemented except for the following limitations:
</para>
<itemizedlist>
<listitem><para>
Only the <literal>ccall</literal> calling convention is supported.
All others are flagged as errors.
</para></listitem>

<listitem><para>
<literal>foreign export</literal> is not implemented.
</para></listitem>

<listitem><para>
<literal>foreign import wrapper</literal> are only implemented for the
x86, PowerPC and Sparc architectures and has been most thoroughly tested on
Windows and Linux using gcc.
</para></listitem>
</itemizedlist>
<para>
Modules containing <literal>foreign</literal> declarations must be compiled
with <command>ffihugs</command> before use (see <xref linkend="ffihugs"/>).
</para>

</sect2>

<sect2 id="hierarchical-libraries">
<title>Hierarchical Namespace Extension</title>

<para>
The
<ulink url="http://www.haskell.org/hierarchical-modules/">Haskell Hierarchical Namespace Extension</ulink>
allows dots in module names, e.g. <literal>System.IO.Error</literal>,
creating a hierarchical module namespace.
Hugs has supported this since the December 2001 release.
When searching for the source file corresponding to a hierarchical name,
Hugs replaces the dots with slashes.
</para>

</sect2>

</sect1>

</chapter>