1 <?xml version="1.0" encoding="iso-8859-1"?>
2 <chapter id="bugs-and-infelicities">
3 <title>Known bugs and infelicities</title>
5 <sect1 id="vs-Haskell-defn">
6 <title>Haskell 98 vs. Glasgow Haskell: language non-compliance
9 <indexterm><primary>GHC vs the Haskell 98 language</primary></indexterm>
10 <indexterm><primary>Haskell 98 language vs GHC</primary></indexterm>
12 <para>This section lists Glasgow Haskell infelicities in its
13 implementation of Haskell 98. See also the “when things
14 go wrong” section (<xref linkend="wrong"/>) for information
15 about crashes, space leaks, and other undesirable phenomena.</para>
17 <para>The limitations here are listed in Haskell Report order
20 <sect2 id="haskell98-divergence">
21 <title>Divergence from Haskell 98</title>
24 <sect3 id="infelicities-lexical">
25 <title>Lexical syntax</title>
29 <para>The Haskell report specifies that programs may be
30 written using Unicode. GHC only accepts the ISO-8859-1
31 character set at the moment.</para>
35 <para>Certain lexical rules regarding qualified identifiers
36 are slightly different in GHC compared to the Haskell
38 <replaceable>module</replaceable><literal>.</literal><replaceable>reservedop</replaceable>,
39 such as <literal>M.\</literal>, GHC will interpret it as a
40 single qualified operator rather than the two lexemes
41 <literal>M</literal> and <literal>.\</literal>.</para>
46 <sect3 id="infelicities-syntax">
47 <title>Context-free syntax</title>
51 <para>GHC is a little less strict about the layout rule when used
52 in <literal>do</literal> expressions. Specifically, the
53 restriction that "a nested context must be indented further to
54 the right than the enclosing context" is relaxed to allow the
55 nested context to be at the same level as the enclosing context,
56 if the enclosing context is a <literal>do</literal>
59 <para>For example, the following code is accepted by GHC:
62 main = do args <- getArgs
63 if null args then return [] else do
64 ps <- mapM process args
65 mapM print ps</programlisting>
71 <para>GHC doesn't do fixity resolution in expressions during
72 parsing. For example, according to the Haskell report, the
73 following expression is legal Haskell:
75 let x = 42 in x == 42 == True</programlisting>
78 (let x = 42 in x == 42) == True</programlisting>
80 because according to the report, the <literal>let</literal>
81 expression <quote>extends as far to the right as
82 possible</quote>. Since it can't extend past the second
83 equals sign without causing a parse error
84 (<literal>==</literal> is non-fix), the
85 <literal>let</literal>-expression must terminate there. GHC
86 simply gobbles up the whole expression, parsing like this:
88 (let x = 42 in x == 42 == True)</programlisting>
90 The Haskell report is arguably wrong here, but nevertheless
91 it's a difference between GHC & Haskell 98.</para>
96 <sect3 id="infelicities-exprs-pats">
97 <title>Expressions and patterns</title>
99 <para>None known.</para>
102 <sect3 id="infelicities-decls">
103 <title>Declarations and bindings</title>
105 <para>None known.</para>
108 <sect3 id="infelicities-Modules">
109 <title>Module system and interface files</title>
111 <para>None known.</para>
114 <sect3 id="infelicities-numbers">
115 <title>Numbers, basic types, and built-in classes</title>
119 <term>Multiply-defined array elements—not checked:</term>
121 <para>This code fragment should
122 elicit a fatal error, but it does not:
125 main = print (array (1,1) [(1,2), (1,3)])</programlisting>
126 GHC's implementation of <literal>array</literal> takes the value of an
127 array slot from the last (index,value) pair in the list, and does no
128 checking for duplicates. The reason for this is efficiency, pure and simple.
136 <sect3 id="infelicities-Prelude">
137 <title>In <literal>Prelude</literal> support</title>
141 <term>Arbitrary-sized tuples</term>
143 <para>Tuples are currently limited to size 100. HOWEVER:
144 standard instances for tuples (<literal>Eq</literal>,
145 <literal>Ord</literal>, <literal>Bounded</literal>,
146 <literal>Ix</literal> <literal>Read</literal>, and
147 <literal>Show</literal>) are available
148 <emphasis>only</emphasis> up to 16-tuples.</para>
150 <para>This limitation is easily subvertible, so please ask
151 if you get stuck on it.</para>
156 <term><literal>Read</literal>ing integers</term>
158 <para>GHC's implementation of the
159 <literal>Read</literal> class for integral types accepts
160 hexadecimal and octal literals (the code in the Haskell
161 98 report doesn't). So, for example,
162 <programlisting>read "0xf00" :: Int</programlisting>
164 <para>A possible reason for this is that <literal>readLitChar</literal> accepts hex and
165 octal escapes, so it seems inconsistent not to do so for integers too.</para>
170 <term><literal>isAlpha</literal></term>
172 <para>The Haskell 98 definition of <literal>isAlpha</literal>
175 <programlisting>isAlpha c = isUpper c || isLower c</programlisting>
177 <para>GHC's implementation diverges from the Haskell 98
178 definition in the sense that Unicode alphabetic characters which
179 are neither upper nor lower case will still be identified as
180 alphabetic by <literal>isAlpha</literal>.</para>
187 <sect2 id="haskell98-undefined">
188 <title>GHC's interpretation of undefined behaviour in
189 Haskell 98</title>
191 <para>This section documents GHC's take on various issues that are
192 left undefined or implementation specific in Haskell 98.</para>
197 The <literal>Char</literal> type
198 <indexterm><primary><literal>Char</literal></primary><secondary>size of</secondary></indexterm>
201 <para>Following the ISO-10646 standard,
202 <literal>maxBound :: Char</literal> in GHC is
203 <literal>0x10FFFF</literal>.</para>
210 <indexterm><primary><literal>Int</literal></primary><secondary>size of</secondary></indexterm>
213 <para>In GHC the <literal>Int</literal> type follows the
214 size of an address on the host architecture; in other words
215 it holds 32 bits on a 32-bit machine, and 64-bits on a
216 64-bit machine.</para>
218 <para>Arithmetic on <literal>Int</literal> is unchecked for
219 overflow<indexterm><primary>overflow</primary><secondary><literal>Int</literal></secondary>
220 </indexterm>, so all operations on <literal>Int</literal> happen
222 2<superscript><replaceable>n</replaceable></superscript>
223 where <replaceable>n</replaceable> is the size in bits of
224 the <literal>Int</literal> type.</para>
226 <para>The <literal>fromInteger</literal><indexterm><primary><literal>fromInteger</literal></primary>
227 </indexterm>function (and hence
228 also <literal>fromIntegral</literal><indexterm><primary><literal>fromIntegral</literal></primary>
229 </indexterm>) is a special case when
230 converting to <literal>Int</literal>. The value of
231 <literal>fromIntegral x :: Int</literal> is given by taking
232 the lower <replaceable>n</replaceable> bits of <literal>(abs
233 x)</literal>, multiplied by the sign of <literal>x</literal>
234 (in 2's complement <replaceable>n</replaceable>-bit
235 arithmetic). This behaviour was chosen so that for example
236 writing <literal>0xffffffff :: Int</literal> preserves the
237 bit-pattern in the resulting <literal>Int</literal>.</para>
240 <para>Negative literals, such as <literal>-3</literal>, are
241 specified by (a careful reading of) the Haskell Report as
242 meaning <literal>Prelude.negate (Prelude.fromInteger 3)</literal>.
243 So <literal>-2147483648</literal> means <literal>negate (fromInteger 2147483648)</literal>.
244 Since <literal>fromInteger</literal> takes the lower 32 bits of the representation,
245 <literal>fromInteger (2147483648::Integer)</literal>, computed at type <literal>Int</literal> is
246 <literal>-2147483648::Int</literal>. The <literal>negate</literal> operation then
247 overflows, but it is unchecked, so <literal>negate (-2147483648::Int)</literal> is just
248 <literal>-2147483648</literal>. In short, one can write <literal>minBound::Int</literal> as
249 a literal with the expected meaning (but that is not in general guaranteed.
252 <para>The <literal>fromIntegral</literal> function also
253 preserves bit-patterns when converting between the sized
254 integral types (<literal>Int8</literal>,
255 <literal>Int16</literal>, <literal>Int32</literal>,
256 <literal>Int64</literal> and the unsigned
257 <literal>Word</literal> variants), see the modules
258 <literal>Data.Int</literal> and <literal>Data.Word</literal>
259 in the library documentation.</para>
264 <term>Unchecked float arithmetic</term>
266 <para>Operations on <literal>Float</literal> and
267 <literal>Double</literal> numbers are
268 <emphasis>unchecked</emphasis> for overflow, underflow, and
269 other sad occurrences. (note, however that some
270 architectures trap floating-point overflow and
271 loss-of-precision and report a floating-point exception,
272 probably terminating the
273 program)<indexterm><primary>floating-point
274 exceptions</primary></indexterm>.</para>
284 <title>Known bugs or infelicities</title>
286 <para>In addition to the divergences from the Haskell 98 standard
287 listed above, GHC has the following known bugs or
290 <sect2 id="bugs-ghc">
291 <title>Bugs in GHC</title>
295 <para> GHC can warn about non-exhaustive or overlapping
296 patterns (see <xref linkend="options-sanity"/>), and usually
297 does so correctly. But not always. It gets confused by
298 string patterns, and by guards, and can then emit bogus
299 warnings. The entire overlap-check code needs an overhaul
304 <para>GHC does not allow you to have a data type with a context
305 that mentions type variables that are not data type parameters.
308 data C a b => T a = MkT a
310 so that <literal>MkT</literal>'s type is
312 MkT :: forall a b. C a b => a -> T a
314 In principle, with a suitable class declaration with a functional dependency,
315 it's possible that this type is not ambiguous; but GHC nevertheless rejects
316 it. The type variables mentioned in the context of the data type declaration must
317 be among the type parameters of the data type.</para>
321 <para>GHC's inliner can be persuaded into non-termination
322 using the standard way to encode recursion via a data type:</para>
324 data U = MkU (U -> Bool)
327 russel u@(MkU p) = not $ p u
330 x = russel (MkU russel)
333 <para>We have never found another class of programs, other
334 than this contrived one, that makes GHC diverge, and fixing
335 the problem would impose an extra overhead on every
336 compilation. So the bug remains un-fixed. There is more
338 url="http://research.microsoft.com/~simonpj/Papers/inlining">
339 Secrets of the GHC inliner</ulink>.</para>
344 <sect2 id="bugs-ghci">
345 <title>Bugs in GHCi (the interactive GHC)</title>
348 <para>GHCi does not respect the <literal>default</literal>
349 declaration in the module whose scope you are in. Instead,
350 for expressions typed at the command line, you always get the
351 default default-type behaviour; that is,
352 <literal>default(Int,Double)</literal>.</para>
354 <para>It would be better for GHCi to record what the default
355 settings in each module are, and use those of the 'current'
356 module (whatever that is).</para>
360 <para>GHCi does not keep careful track of what instance
361 declarations are 'in scope' if they come from other packages.
362 Instead, all instance declarations that GHC has seen in other
363 packages are all in scope everywhere, whether or not the
364 module from that package is used by the command-line
369 <para>On Windows, there's a GNU ld/BFD bug
370 whereby it emits bogus PE object files that have more than
371 0xffff relocations. When GHCi tries to load a package affected by this
372 bug, you get an error message of the form
374 Loading package javavm ... linking ... WARNING: Overflown relocation field (# relocs found: 30765)
376 The last time we looked, this bug still
377 wasn't fixed in the BFD codebase, and there wasn't any
378 noticeable interest in fixing it when we reported the bug
381 <para>The workaround is to split up the .o files that make up
382 your package into two or more .o's, along the lines of
383 how the "base" package does it.</para>
392 ;;; Local Variables: ***
394 ;;; sgml-parent-document: ("users_guide.xml" "book" "chapter") ***