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>Certain lexical rules regarding qualified identifiers
30 are slightly different in GHC compared to the Haskell
32 <replaceable>module</replaceable><literal>.</literal><replaceable>reservedop</replaceable>,
33 such as <literal>M.\</literal>, GHC will interpret it as a
34 single qualified operator rather than the two lexemes
35 <literal>M</literal> and <literal>.\</literal>.</para>
40 <sect3 id="infelicities-syntax">
41 <title>Context-free syntax</title>
45 <para>GHC is a little less strict about the layout rule when used
46 in <literal>do</literal> expressions. Specifically, the
47 restriction that "a nested context must be indented further to
48 the right than the enclosing context" is relaxed to allow the
49 nested context to be at the same level as the enclosing context,
50 if the enclosing context is a <literal>do</literal>
53 <para>For example, the following code is accepted by GHC:
56 main = do args <- getArgs
57 if null args then return [] else do
58 ps <- mapM process args
59 mapM print ps</programlisting>
65 <para>GHC doesn't do fixity resolution in expressions during
66 parsing. For example, according to the Haskell report, the
67 following expression is legal Haskell:
69 let x = 42 in x == 42 == True</programlisting>
72 (let x = 42 in x == 42) == True</programlisting>
74 because according to the report, the <literal>let</literal>
75 expression <quote>extends as far to the right as
76 possible</quote>. Since it can't extend past the second
77 equals sign without causing a parse error
78 (<literal>==</literal> is non-fix), the
79 <literal>let</literal>-expression must terminate there. GHC
80 simply gobbles up the whole expression, parsing like this:
82 (let x = 42 in x == 42 == True)</programlisting>
84 The Haskell report is arguably wrong here, but nevertheless
85 it's a difference between GHC & Haskell 98.</para>
90 <sect3 id="infelicities-exprs-pats">
91 <title>Expressions and patterns</title>
93 <para>None known.</para>
96 <sect3 id="infelicities-decls">
97 <title>Declarations and bindings</title>
99 <para>GHC's typechecker makes all pattern bindings monomorphic
100 by default; this behaviour can be disabled with
101 <option>-fno-mono-pat-binds</option>. See <xref
102 linkend="options-language" />.</para>
105 <sect3 id="infelicities-Modules">
106 <title>Module system and interface files</title>
108 <para>None known.</para>
111 <sect3 id="infelicities-numbers">
112 <title>Numbers, basic types, and built-in classes</title>
116 <term>Multiply-defined array elements—not checked:</term>
118 <para>This code fragment should
119 elicit a fatal error, but it does not:
122 main = print (array (1,1) [(1,2), (1,3)])</programlisting>
123 GHC's implementation of <literal>array</literal> takes the value of an
124 array slot from the last (index,value) pair in the list, and does no
125 checking for duplicates. The reason for this is efficiency, pure and simple.
133 <sect3 id="infelicities-Prelude">
134 <title>In <literal>Prelude</literal> support</title>
138 <term>Arbitrary-sized tuples</term>
140 <para>Tuples are currently limited to size 100. HOWEVER:
141 standard instances for tuples (<literal>Eq</literal>,
142 <literal>Ord</literal>, <literal>Bounded</literal>,
143 <literal>Ix</literal> <literal>Read</literal>, and
144 <literal>Show</literal>) are available
145 <emphasis>only</emphasis> up to 16-tuples.</para>
147 <para>This limitation is easily subvertible, so please ask
148 if you get stuck on it.</para>
153 <term><literal>Read</literal>ing integers</term>
155 <para>GHC's implementation of the
156 <literal>Read</literal> class for integral types accepts
157 hexadecimal and octal literals (the code in the Haskell
158 98 report doesn't). So, for example,
159 <programlisting>read "0xf00" :: Int</programlisting>
161 <para>A possible reason for this is that <literal>readLitChar</literal> accepts hex and
162 octal escapes, so it seems inconsistent not to do so for integers too.</para>
167 <term><literal>isAlpha</literal></term>
169 <para>The Haskell 98 definition of <literal>isAlpha</literal>
172 <programlisting>isAlpha c = isUpper c || isLower c</programlisting>
174 <para>GHC's implementation diverges from the Haskell 98
175 definition in the sense that Unicode alphabetic characters which
176 are neither upper nor lower case will still be identified as
177 alphabetic by <literal>isAlpha</literal>.</para>
184 <sect2 id="haskell98-undefined">
185 <title>GHC's interpretation of undefined behaviour in
186 Haskell 98</title>
188 <para>This section documents GHC's take on various issues that are
189 left undefined or implementation specific in Haskell 98.</para>
194 The <literal>Char</literal> type
195 <indexterm><primary><literal>Char</literal></primary><secondary>size of</secondary></indexterm>
198 <para>Following the ISO-10646 standard,
199 <literal>maxBound :: Char</literal> in GHC is
200 <literal>0x10FFFF</literal>.</para>
207 <indexterm><primary><literal>Int</literal></primary><secondary>size of</secondary></indexterm>
210 <para>In GHC the <literal>Int</literal> type follows the
211 size of an address on the host architecture; in other words
212 it holds 32 bits on a 32-bit machine, and 64-bits on a
213 64-bit machine.</para>
215 <para>Arithmetic on <literal>Int</literal> is unchecked for
216 overflow<indexterm><primary>overflow</primary><secondary><literal>Int</literal></secondary>
217 </indexterm>, so all operations on <literal>Int</literal> happen
219 2<superscript><replaceable>n</replaceable></superscript>
220 where <replaceable>n</replaceable> is the size in bits of
221 the <literal>Int</literal> type.</para>
223 <para>The <literal>fromInteger</literal><indexterm><primary><literal>fromInteger</literal></primary>
224 </indexterm>function (and hence
225 also <literal>fromIntegral</literal><indexterm><primary><literal>fromIntegral</literal></primary>
226 </indexterm>) is a special case when
227 converting to <literal>Int</literal>. The value of
228 <literal>fromIntegral x :: Int</literal> is given by taking
229 the lower <replaceable>n</replaceable> bits of <literal>(abs
230 x)</literal>, multiplied by the sign of <literal>x</literal>
231 (in 2's complement <replaceable>n</replaceable>-bit
232 arithmetic). This behaviour was chosen so that for example
233 writing <literal>0xffffffff :: Int</literal> preserves the
234 bit-pattern in the resulting <literal>Int</literal>.</para>
237 <para>Negative literals, such as <literal>-3</literal>, are
238 specified by (a careful reading of) the Haskell Report as
239 meaning <literal>Prelude.negate (Prelude.fromInteger 3)</literal>.
240 So <literal>-2147483648</literal> means <literal>negate (fromInteger 2147483648)</literal>.
241 Since <literal>fromInteger</literal> takes the lower 32 bits of the representation,
242 <literal>fromInteger (2147483648::Integer)</literal>, computed at type <literal>Int</literal> is
243 <literal>-2147483648::Int</literal>. The <literal>negate</literal> operation then
244 overflows, but it is unchecked, so <literal>negate (-2147483648::Int)</literal> is just
245 <literal>-2147483648</literal>. In short, one can write <literal>minBound::Int</literal> as
246 a literal with the expected meaning (but that is not in general guaranteed.
249 <para>The <literal>fromIntegral</literal> function also
250 preserves bit-patterns when converting between the sized
251 integral types (<literal>Int8</literal>,
252 <literal>Int16</literal>, <literal>Int32</literal>,
253 <literal>Int64</literal> and the unsigned
254 <literal>Word</literal> variants), see the modules
255 <literal>Data.Int</literal> and <literal>Data.Word</literal>
256 in the library documentation.</para>
261 <term>Unchecked float arithmetic</term>
263 <para>Operations on <literal>Float</literal> and
264 <literal>Double</literal> numbers are
265 <emphasis>unchecked</emphasis> for overflow, underflow, and
266 other sad occurrences. (note, however that some
267 architectures trap floating-point overflow and
268 loss-of-precision and report a floating-point exception,
269 probably terminating the
270 program)<indexterm><primary>floating-point
271 exceptions</primary></indexterm>.</para>
281 <title>Known bugs or infelicities</title>
283 <para>The bug tracker lists bugs that have been reported in GHC but not
284 yet fixed: see the <ulink url="http://sourceforge.net/projects/ghc/">SourceForge GHC
285 page</ulink>. In addition to those, GHC also has the following known bugs
286 or infelicities. These bugs are more permanent; it is unlikely that
287 any of them will be fixed in the short term.</para>
289 <sect2 id="bugs-ghc">
290 <title>Bugs in GHC</title>
294 <para> GHC can warn about non-exhaustive or overlapping
295 patterns (see <xref linkend="options-sanity"/>), and usually
296 does so correctly. But not always. It gets confused by
297 string patterns, and by guards, and can then emit bogus
298 warnings. The entire overlap-check code needs an overhaul
303 <para>GHC does not allow you to have a data type with a context
304 that mentions type variables that are not data type parameters.
307 data C a b => T a = MkT a
309 so that <literal>MkT</literal>'s type is
311 MkT :: forall a b. C a b => a -> T a
313 In principle, with a suitable class declaration with a functional dependency,
314 it's possible that this type is not ambiguous; but GHC nevertheless rejects
315 it. The type variables mentioned in the context of the data type declaration must
316 be among the type parameters of the data type.</para>
320 <para>GHC's inliner can be persuaded into non-termination
321 using the standard way to encode recursion via a data type:</para>
323 data U = MkU (U -> Bool)
326 russel u@(MkU p) = not $ p u
329 x = russel (MkU russel)
332 <para>We have never found another class of programs, other
333 than this contrived one, that makes GHC diverge, and fixing
334 the problem would impose an extra overhead on every
335 compilation. So the bug remains un-fixed. There is more
337 url="http://research.microsoft.com/~simonpj/Papers/inlining">
338 Secrets of the GHC inliner</ulink>.</para>
342 <para>GHC does not keep careful track of
343 what instance declarations are 'in scope' if they come from other packages.
344 Instead, all instance declarations that GHC has seen in other
345 packages are all in scope everywhere, whether or not the
346 module from that package is used by the command-line
347 expression. This bug affects only the <option>--make</option> mode and
354 <sect2 id="bugs-ghci">
355 <title>Bugs in GHCi (the interactive GHC)</title>
358 <para>GHCi does not respect the <literal>default</literal>
359 declaration in the module whose scope you are in. Instead,
360 for expressions typed at the command line, you always get the
361 default default-type behaviour; that is,
362 <literal>default(Int,Double)</literal>.</para>
364 <para>It would be better for GHCi to record what the default
365 settings in each module are, and use those of the 'current'
366 module (whatever that is).</para>
370 <para>On Windows, there's a GNU ld/BFD bug
371 whereby it emits bogus PE object files that have more than
372 0xffff relocations. When GHCi tries to load a package affected by this
373 bug, you get an error message of the form
375 Loading package javavm ... linking ... WARNING: Overflown relocation field (# relocs found: 30765)
377 The last time we looked, this bug still
378 wasn't fixed in the BFD codebase, and there wasn't any
379 noticeable interest in fixing it when we reported the bug
382 <para>The workaround is to split up the .o files that make up
383 your package into two or more .o's, along the lines of
384 how the "base" package does it.</para>
393 ;;; Local Variables: ***
395 ;;; sgml-parent-document: ("users_guide.xml" "book" "chapter") ***