1 <Chapter id="bugs-and-infelicities">
2 <title>Known bugs and infelicities</title>
4 <sect1 id="vs-Haskell-defn">
5 <title>Haskell 98 vs. Glasgow Haskell: language non-compliance
8 <indexterm><primary>GHC vs the Haskell 98 language</primary></indexterm>
9 <indexterm><primary>Haskell 98 language vs GHC</primary></indexterm>
11 <para>This section lists Glasgow Haskell infelicities in its
12 implementation of Haskell 98. See also the “when things
13 go wrong” section (<XRef LinkEnd="wrong">) for information
14 about crashes, space leaks, and other undesirable phenomena.</para>
16 <para>The limitations here are listed in Haskell Report order
19 <sect2 id="haskell98-divergence">
20 <title>Divergence from Haskell 98</title>
23 <sect3 id="infelicities-lexical">
24 <title>Lexical syntax</title>
28 <para>The Haskell report specifies that programs may be
29 written using Unicode. GHC only accepts the ISO-8859-1
30 character set at the moment.</para>
34 <para>Certain lexical rules regarding qualified identifiers
35 are slightly different in GHC compared to the Haskell
37 <replaceable>module</replaceable><literal>.</literal><replaceable>reservedop</replaceable>,
38 such as <literal>M.\</literal>, GHC will interpret it as a
39 single qualified operator rather than the two lexemes
40 <literal>M</literal> and <literal>.\</literal>.</para>
45 <sect3 id="infelicities-syntax">
46 <title>Context-free syntax</title>
50 <para>GHC doesn't do fixity resolution in expressions during
51 parsing. For example, according to the Haskell report, the
52 following expression is legal Haskell:
54 let x = 42 in x == 42 == True</programlisting>
57 (let x = 42 in x == 42) == True</programlisting>
59 because according to the report, the <literal>let</literal>
60 expression <quote>extends as far to the right as
61 possible</quote>. Since it can't extend past the second
62 equals sign without causing a parse error
63 (<literal>==</literal> is non-fix), the
64 <literal>let</literal>-expression must terminate there. GHC
65 simply gobbles up the whole expression, parsing like this:
67 (let x = 42 in x == 42 == True)</programlisting>
69 The Haskell report is arguably wrong here, but nevertheless
70 it's a difference between GHC & Haskell 98.</para>
75 <sect3 id="infelicities-exprs-pats">
76 <title>Expressions and patterns</title>
78 <para>None known.</para>
81 <sect3 id="infelicities-decls">
82 <title>Declarations and bindings</title>
84 <para>None known.</para>
87 <sect3 id="infelicities-Modules">
88 <title>Module system and interface files</title>
90 <para>None known.</para>
93 <sect3 id="infelicities-numbers">
94 <title>Numbers, basic types, and built-in classes</title>
98 <term>Multiply-defined array elements—not checked:</term>
100 <para>This code fragment <emphasis>should</emphasis>
101 elicit a fatal error, but it does not:
104 main = print (array (1,1) [(1,2), (1,3)])</programlisting>
113 <sect3 id="infelicities-Prelude">
114 <title>In <literal>Prelude</literal> support</title>
118 <term>The <literal>Char</literal> type</term>
119 <indexterm><primary><literal>Char</literal></primary><secondary>size
120 of</secondary></indexterm>
122 <para>The Haskell report says that the
123 <literal>Char</literal> type holds 16 bits. GHC follows
124 the ISO-10646 standard a little more closely:
125 <literal>maxBound :: Char</literal> in GHC is
126 <literal>0x10FFFF</literal>.</para>
131 <term>Arbitrary-sized tuples</term>
133 <para>Tuples are currently limited to size 100. HOWEVER:
134 standard instances for tuples (<literal>Eq</literal>,
135 <literal>Ord</literal>, <literal>Bounded</literal>,
136 <literal>Ix</literal> <literal>Read</literal>, and
137 <literal>Show</literal>) are available
138 <emphasis>only</emphasis> up to 16-tuples.</para>
140 <para>This limitation is easily subvertible, so please ask
141 if you get stuck on it.</para>
146 <term><literal>Read</literal>ing integers</term>
148 <para>GHC's implementation of the
149 <literal>Read</literal> class for integral types accepts
150 hexadeciaml and octal literals (the code in the Haskell
151 98 report doesn't). So, for example,
152 <programlisting>read "0xf00" :: Int</programlisting>
154 <para>A possible reason for this is that <literal>readLitChar</literal> accepts hex and
155 octal escapes, so it seems inconsistent not to do so for integers too.</para>
162 <sect2 id="haskell98-undefined">
163 <title>GHC's interpretation of undefined behaviour in
164 Haskell 98</title>
166 <para>This section documents GHC's take on various issues that are
167 left undefined or implementation specific in Haskell 98.</para>
171 <term>Sized integral types</term>
172 <indexterm><primary><literal>Int</literal></primary><secondary>size of</secondary>
176 <para>In GHC the <literal>Int</literal> type follows the
177 size of an address on the host architecture; in other words
178 it holds 32 bits on a 32-bit machine, and 64-bits on a
179 64-bit machine.</para>
181 <para>Arithmetic on <literal>Int</literal> is unchecked for
182 overflow<indexterm><primary>overflow</primary><secondary><literal>Int</literal></secondary>
183 </indexterm>, so all operations on <literal>Int</literal> happen
185 2<superscript><replaceable>n</replaceable></superscript>
186 where <replaceable>n</replaceable> is the size in bits of
187 the <literal>Int</literal> type.</para>
189 <para>The <literal>fromInteger</literal><indexterm><primary><literal>fromInteger</literal></primary>
190 </indexterm>function (and hence
191 also <literal>fromIntegral</literal><indexterm><primary><literal>fromIntegral</literal></primary>
192 </indexterm>) is a special case when
193 converting to <literal>Int</literal>. The value of
194 <literal>fromIntegral x :: Int</literal> is given by taking
195 the lower <replaceable>n</replaceable> bits of <literal>(abs
196 x)</literal>, multiplied by the sign of <literal>x</literal>
197 (in 2's complement <replaceable>n</replaceable>-bit
198 arithmetic). This behaviour was chosen so that for example
199 writing <literal>0xffffffff :: Int</literal> preserves the
200 bit-pattern in the resulting <literal>Int</literal>.</para>
203 <para>Negative literals, such as <literal>-3</literal>, are
204 specified by (a careful reading of) the Haskell Report as
205 meaning <literal>Prelude.negate (Prelude.fromInteger 3)</literal>.
206 So <literal>-2147483648</literal> means <literal>negate (fromInteger 2147483648)</literal>.
207 Since <literal>fromInteger</literal> takes the lower 32 bits of the representation,
208 <literal>fromInteger (2147483648::Integer)</literal>, computed at type <literal>Int</literal> is
209 <literal>-2147483648::Int</literal>. The <literal>negate</literal> operation then
210 overflows, but it is unchecked, so <literal>negate (-2147483648::Int)</literal> is just
211 <literal>-2147483648</literal>. In short, one can write <literal>minBound::Int</literal> as
212 a literal with the expected meaning (but that is not in general guaranteed.
215 <para>The <literal>fromIntegral</literal> function also
216 preserves bit-patterns when converting between the sized
217 integral types (<literal>Int8</literal>,
218 <literal>Int16</literal>, <literal>Int32</literal>,
219 <literal>Int64</literal> and the unsigned
220 <literal>Word</literal> variants), see the modules
221 <literal>Data.Int</literal> and <literal>Data.Word</literal>
222 in the library documentation.</para>
227 <term>Unchecked float arithmetic</term>
229 <para>Operations on <literal>Float</literal> and
230 <literal>Double</literal> numbers are
231 <emphasis>unchecked</emphasis> for overflow, underflow, and
232 other sad occurrences. (note, however that some
233 architectures trap floating-point overflow and
234 loss-of-precision and report a floating-point exception,
235 probably terminating the
236 program)<indexterm><primary>floating-point
237 exceptions</primary></indexterm>.</para>
247 <title>Known bugs or infelicities</title>
249 <para>In addition to the divergences from the Haskell 98 standard
250 listed above, GHC has the following known bugs or
255 <para> GHC can warn about non-exhaustive or overlapping
256 patterns (see <xref linkend="options-sanity">, and usually
257 does so correctly. But not always. It gets confused by
258 string patterns, and by guards, and can then emit bogus
259 warnings. The entire overlap-check code needs an overhaul
264 <para>Dangers with multiple <literal>Main</literal>
267 <para>GHC does not insist that module <literal>Main</literal>
268 lives in a file called <filename>Main.hs</filename>. This is
269 useful if you want multiple versions of
270 <literal>Main</literal>. But there's a danger: when compiling
271 module <literal>Main</literal> (regardless of what file it
272 comes from), GHC looks for the interface
273 <filename>Main.hi</filename>; it uses this to get version
274 information from the last time it recompiled
275 <literal>Main</literal>. The trouble is that this
276 <filename>Main.hi</filename> may not correspond to the source
277 file being compiled.</para>
279 <para>Solution: remove <filename>Main.hi</filename> first. A
280 better solution would be for GHC to record the source-file
281 filename in the interface file, or even an MD5 checksum.
286 <para>GHC does not allow you to have a data type with a context
287 that mentions type variables that are not data type parameters.
290 data C a b => T a = MkT a
292 so that <literal>MkT</literal>'s type is
294 MkT :: forall a b. C a b => a -> T a
296 In principle, with a suitable class declaration with a functional dependency,
297 it's possible that this type is not ambiguous; but GHC nevertheless rejects
298 it. The type variables mentioned in the context of the data type declaration must
299 be among the type parameters of the data type.</para>
303 <para>GHCi does not respect the <literal>default</literal>
304 declaration in the module whose scope you are in. Instead,
305 for expressions typed at the command line, you always get the
306 default default-type behaviour; that is,
307 <literal>default(Int,Double)</literal>.</para>
309 <para>It would be better for GHCi to record what the default
310 settings in each module are, and use those of the 'current'
311 module (whatever that is).</para>
315 <para>GHCi does not keep careful track of what instance
316 declarations are 'in scope' if they come from other packages.
317 Instead, all instance declarations that GHC has seen in other
318 packages are all in scope everywhere, whether or not the
319 module from that package is used by the command-line
324 <para>GHC's inliner can be persuaded into non-termination
325 using the standard way to encode recursion via a data type:</para>
327 data U = MkU (U -> Bool)
330 russel u@(MkU p) = not $ p u
333 x = russel (MkU russel)
336 <para>We have never found another class of programs, other
337 than this contrived one, that makes GHC diverge, and fixing
338 the problem would impose an extra overhead on every
339 compilation. So the bug remains un-fixed. There is more
341 url="http://research.microsoft.com/~simonpj/Papers/inlining">
342 Secrets of the GHC inliner</ulink>.</para>
350 ;;; Local Variables: ***
352 ;;; sgml-parent-document: ("users_guide.sgml" "book" "chapter" "sect1") ***