4 <para>This section has the answers to questions that get asked
5 regularly on the GHC mailing lists, in no particular order. Please
6 let us know if you think there's a question/answer that should be
11 <term>How do I port GHC to platform X?</term>
13 <para>There are two distinct possibilities: either</para>
16 <para>The hardware architecture for your system is already
17 supported by GHC, but you're running an OS that isn't
18 supported (or perhaps has been supported in the past, but
19 currently isn't). This is the easiest type of porting
20 job, but it still requires some careful
25 <para>Your system's hardware architecture isn't supported
26 by GHC. This will be a more difficult port (though by
27 comparison perhaps not as difficult as porting
32 <para>Both ways require you to bootrap from intermediate
33 <literal>HC</literal> files: these are the stylised C files
34 generated by GHC when it compiles Haskell source. Basically
35 the idea is to take the HC files for GHC itself to the target
36 machine and compile them with <literal>gcc</literal> to get a
37 working GHC, and go from there.</para>
40 url="http://www.haskell.org/ghc/latest/building/building-guide.html">Building
41 Guide</ulink> has all the details on how to bootstrap GHC on a
48 <term>Do I have to recompile all my code if I upgrade
51 <para>Yes. There are two reasons for this:</para>
54 <para>GHC does a lot of cross-module optimisation, so
55 compiled code will include parts of the libraries it was
56 compiled against (including the Prelude), so will be
57 deeply tied to the actual version of those libraries it
58 was compiled against. When you upgrade GHC, the libraries
59 may change; even if the external interface of the
60 libraries doesn't change, sometimes internal details may
61 change because GHC optimised the code in the library
65 <para>We sometimes change the ABI (application binary
66 interface) between versions of GHC. Code compiled with
67 one version of GHC is not necessarily compatible with code
68 compiled by a different version, even if you arrange to
69 keep the same libraries.</para>
76 <term>Why doesn't GHC use shared libraries?</term>
78 <para>The subject of shared libraries has come up several
79 times in the past — take a look through the mailing-list
80 archives for some of the previous discussions. The upshot is
81 that shared libraries wouldn't really buy much unless you
82 really need to save the disk space: in all other
83 considerations, static linking comes out better.</para>
85 <para>Unfortunately GHC-compiled libraries are very tightly
86 coupled, which means it's unlikely you'd be able to swap out a
87 shared library for a newer version unless it was compiled with
88 <emphasis>exactly</emphasis> the same compiler and set of
89 libraries as the old version.</para>
94 <term>I can't get string gaps to work</term>
96 <para>If you're also using CPP, beware of the known pitfall
97 with string gaps mentioned in <xref
98 linkend="cpp-string-gaps"/>.</para>
103 <term>GHCi complains about missing symbols like
104 <literal>CC_LIST</literal> when loading a previously compiled .o
107 <para> This probably means the .o files in question were
108 compiled for profiling (with <option>-prof</option>). Workaround:
109 recompile them without profiling. We really ought to detect
110 this situation and give a proper error message.</para>
115 <term>Linking a program causes the following error on Linux:
116 <literal>/usr/bin/ld: cannot open -lgmp: No such file or
117 directory</literal></term>
119 <para>The problem is that your system doesn't have the GMP
120 library installed. If this is a RedHat distribution, install
121 the RedHat-supplied <literal>gmp-devel</literal> package, and
122 the <literal>gmp</literal> package if you don't already have
123 it. There have been reports that installing the RedHat
124 packages also works for SuSE (SuSE don't supply a shared gmp
130 <term>I Can't run GHCi on Linux, because it complains about a
131 missing <literal>libreadline.so.3</literal>.</term>
133 <para>The "correct" fix for this problem is to install the
134 correct RPM for the particular flavour of Linux on your
135 machine. If this isn't an option, however, there is a hack
136 that might work: make a symbolic link from
137 <filename>libreadline.so.4</filename> to
138 <filename>libreadline.so.3</filename> in
139 <literal>/usr/lib</literal>. We tried this on a SuSE 7.1 box
140 and it seemed to work, but YMMV.</para>
145 <term>Solaris users may sometimes get link errors due to
146 libraries needed by GNU Readline.</term>
148 <para>We suggest you try linking in some combination of the
149 <literal>termcap</literal>, <literal>curses</literal> and
150 <literal>ncurses</literal> libraries, by giving
151 <literal>-ltermcap</literal>, <literal>-lcurses</literal> and
152 <literal>-lncurses</literal> respectively. If you encounter
153 this problem, we would appreciate feedback on it, since we
154 don't fully understand what's going on here.</para>
159 <term>The build fails in readline.</term>
161 <para>It has been reported that if you have multiple versions
162 of the readline library installed on Linux, then this may
163 cause the build to fail. If you have multiple versions of
164 readline, try uninstalling all except the most recent
170 <term>When I try to start ghci (probably one I compiled myself)
171 it says <literal>ghc-5.02: not built for interactive
174 <para>To build a working ghci, you need to build GHC 5.02 with
175 itself; the above message appears if you build it with 4.08.X,
176 for example. It'll still work fine for batch-mode
177 compilation, though. Note that you really must build with
178 exactly the same version of the compiler. Building 5.02 with
179 5.00.2, for example, may or may not give a working interactive
180 system; it probably won't, and certainly isn't supported.
181 Note also that you can build 5.02 with any older compiler,
182 back to 4.08.1, if you don't want a working interactive
183 system; that's OK, and supported.</para>
188 <term>When I use a foreign function that takes or returns a
189 float, it gives the wrong answer, or crashes.</term>
191 <para>You should use the <option>-#include</option> option to
192 bring the correct prototype into scope (see <xref
193 linkend="options-C-compiler"/>).</para>
198 <term>My program that uses a really large heap crashes on
201 <para>For utterly horrible reasons, programs that use more
202 than 128Mb of heap won't work when compiled dynamically on
203 Windows (they should be fine statically compiled).</para>
208 <term>GHC doesn't like filenames containing
209 <literal>+</literal>.</term>
211 <para>Indeed not. You could change <literal>+</literal> to
212 <literal>p</literal> or <literal>plus</literal>.</para>
217 <term>When I open a FIFO (named pipe) and try to read from it, I
218 get EOF immediately.</term>
220 <para>This is a consequence of the fact that GHC opens the
221 FIFO in non-blocking mode. The behaviour varies from OS to
222 OS: on Linux and Solaris you can wait for a writer by doing an
223 explicit <literal>threadWaitRead</literal> on the file
224 descriptor (gotten from <literal>Posix.handleToFd</literal>)
225 before the first read, but this doesn't work on FreeBSD
226 (although rumour has it that recent versions of FreeBSD
227 changed the behavour to match other OSs). A workaround for
228 all systems is to open the FIFO for writing yourself, before
229 (or at the same time as) opening it for reading.</para>
234 <term>When I <literal>foreign import</literal> a function that
235 returns <literal>char</literal> or <literal>short</literal>, I
236 get garbage back.</term>
238 <para>This is a known bug in GHC versions prior to 5.02.2.
239 GHC doesn't mask out the more significant bits of the result.
240 It doesn't manifest with gcc 2.95, but apparently shows up
241 with g++ and gcc 3.0.</para>
246 <term>My program is failing with <literal>head []</literal>, or
247 an array bounds error, or some other random error, and I have no
248 idea how to find the bug. Can you help?</term>
251 <para>Compile your program with <literal>-prof
252 -auto-all</literal> (make sure you have the profiling libraries
253 installed), and run it with <literal>+RTS -xc -RTS</literal> to get a
254 “stack trace” at the point at which the exception was
255 raised. See <xref linkend="rts-options-debugging"/> for more
261 <term>How do I increase the heap size permanently for a given
264 <para>See <xref linkend="rts-hooks"/>.</para>
269 <term>I'm trying to compile my program for parallel execution
270 with the <option>-parallel</option>, and GHC complains with an
271 error like “failed to load interface file for
272 Prelude”.</term>
274 <para>GHC doesn't ship with support for parallel execution,
275 that support is provided separately by the <ulink
276 url="http://www.macs.hw.ac.uk/~dsg/gph/">GPH</ulink> project.</para>
281 <term>When is it safe to use
282 <literal>unsafePerformIO</literal>?</term>
284 <para>We'll give two answers to this question, each of which
285 may be helpful. These criteria are not rigorous in any real
286 sense (you'd need a formal semantics for Haskell in order to
287 give a proper answer to this question), but should give you a
288 feel for the kind of things you can and cannot do with
289 <literal>unsafePerformIO</literal>.</para>
293 <para>It is safe to implement a function or API using
294 <literal>unsafePerformIO</literal> if you could imagine
295 also implementing the same function or API in Haskell
296 without using <literal>unsafePerformIO</literal> (forget
297 about efficiency, just consider the semantics).</para>
301 <para>In pure Haskell, the value of a function depends
302 only on the values of its arguments (and free variables,
303 if it has any). If you can implement the function using
304 <literal>unsafePerformIO</literal> and still retain this
305 invariant, then you're probably using
306 <literal>unsafePerformIO</literal> in a safe way. Note
307 that you need only consider the
308 <emphasis>observable</emphasis> values of the arguments
313 <para>For more information, see <ulink
314 url="http://www.haskell.org/pipermail/glasgow-haskell-users/2002-July/003681.html">this
315 thread</ulink>.</para>
320 <term>Why does linking take so long?</term>
322 <para>Linking a small program should take no more than a few
323 seconds. Larger programs can take longer, but even linking
324 GHC itself only takes 3-4 seconds on our development
327 <para>Long link times have been attributed to using Sun's
328 linker on Solaris, as compared to GNU <command>ld</command>
329 which appears to be much faster. So if you're on a Sun box,
330 try switching to GNU <command>ld</command>. <ulink
331 url="http://www.haskell.org/pipermail/glasgow-haskell-users/2002-November/004477.html">This
332 article</ulink> from the mailing list has more
338 <term>If I explicitely set the buffering on a Handle to
339 "NoBuffering" I'm not able to enter EOF by typing
343 <para>This is a consequence of Unixy terminal semantics. Unix
344 does line buffering on terminals in the kernel as part of the
345 terminal processing, unless you turn it off. However, the
346 Ctrl-D processing is also part of the terminal processing
347 which gets turned off when the kernel line buffering is
348 disabled. So GHC tries its best to get NoBuffering semantics
349 by turning off the kernel line buffering, but as a result you
350 lose Ctrl-D. C'est la vie.</para>
355 <term>If I print out a string using <literal>putStr</literal>,
356 and then attempt to read some input using
357 <literal>hGetLine</literal>, I don't see the output from the
358 <literal>putStr</literal>.</term>
361 <para>The <literal>stdout</literal> handle is line-buffered by
362 default, which means that output sent to the handle is only
363 flushed when a newline (<literal>/n</literal>) is output, the
364 buffer is full, or <literal>hFlush</literal> is called on the
365 Handle. The right way to make the text appear without sending
366 a newline is to use <literal>hFlush</literal>:</para>
371 putStr "how are you today? "
373 input &- hGetLine
376 <para>You'll probably find that the behaviour differs when
377 using GHCi: the <literal>hFlush</literal> isn't necessary to
378 make the text appear. This is because in GHCi we turn off the
379 buffering on <literal>stdout</literal>, because this is
380 normally what you want in an interpreter: output appears as it
386 <term>I can't get finalizers to work properly. My program
387 sometimes just prints
388 <literal><<loop>></literal>.</term>
391 <para>Chances are that your program is trying to write a
392 message to <literal>stdout</literal> or
393 <literal>stderr</literal> in the finalizer. Handles have
394 finalizers themselves, and since finalizers don't keep other
395 finalized values alive, the <literal>stdout</literal> and
396 <literal>stderr</literal> Handles may be finalized before your
397 finalizer runs. If this happens, your finalizer will block on
398 the handle, and probably end up receiving a
399 <literal>NonTermination</literal> exception (which is printed
400 as <literal><<loop>></literal>).</para>
406 <term>Does GHC implement any kind of extensible records?</term>
409 <para>No, extensible records are not implemented in GHC.
410 <ulink url="http://www.haskell.org/hugs/">Hugs</ulink>
411 implements TRex, one extensible record variant. The problem
412 is that the record design space is large, and seems to lack
413 local optima. And all reasonable variants break backward
414 compatibility. As a result, nothing much happens.</para>
419 <term>Why do I get errors about missing include files when
420 compiling with <option>-O</option> or
421 <option>-prof</option>?</term>
424 <para>Certain options, such as <option>-O</option>, turn on
425 via-C compilation, instead of using the native code generator.
426 Include files named by <option>-#include</option> options
427 or in <literal>foreign import</literal> declarations are only
428 used in via-C compilation mode. See <xref
429 linkend="finding-header-files"/> for more details.</para>
434 <term>How do I compile my program for profiling without
435 overwriting the object files and <literal>hi</literal> files
436 I've already built?</term>
438 <para>You can select alternative suffixes for object files and
439 interface files, so you can have several builds of the same
440 code coexisting in the same directory. For example, to
441 compile with profiling, you might do this:</para>
443 <screen>ghc --make -prof -o foo-prof -osuf p.o -hisuf p.hi Main</screen>
445 <para>See <xref linkend="options-output"/> for more details on
446 the <option>-osuf</option> and <option>-hisuf</option>
458 ;;; Local Variables: ***
460 ;;; sgml-parent-document: ("users_guide.sgml" "book" "chapter") ***