Change --shared to -shared in Win32 DLL docs

[ghc-hetmet.git] / docs / users_guide / win32-dlls.xml

<?xml version="1.0" encoding="iso-8859-1"?>
<chapter id="win32">
<title>Running GHC on Win32 systems</title>

<sect1 id="ghc-windows">
<title>
Starting GHC on Windows platforms</title>

<para>
The installer that installs GHC on Win32 also sets up the file-suffix associations
for ".hs" and ".lhs" files so that double-clicking them starts <command>ghci</command>.
</para>
<para>
Be aware of that <command>ghc</command> and <command>ghci</command> do
require filenames containing spaces to be escaped using quotes:
<programlisting>
  c:\ghc\bin\ghci "c:\\Program Files\\Haskell\\Project.hs"
</programlisting>
If the quotes are left off in the above command, <command>ghci</command> will
interpret the filename as two, "c:\\Program" and "Files\\Haskell\\Project.hs".
</para>

<!-- not clear whether there are current editions of Win32 OSes that
     doesn't do this by default.

<para> Solution: don't use "Open With...", avoid spaces in file names, 
or fiddle with the appropriate registry setting:
<programlisting>
  HKEY_CLASSES_ROOT\Unknown\shell\openas\command
</programlisting>
Notice how the "%1" argument is quoted (or not).
</para>
<para> This problem doesn't occur when double-clicking.
</para>
-->

</sect1>

<sect1 id="ghci-windows">
<title>Running GHCi on Windows</title>

  <para>We recommend running GHCi in a standard Windows console:
  select the <literal>GHCi</literal> option from the start menu item
  added by the GHC installer, or use
  <literal>Start->Run->cmd</literal> to get a Windows console and
  invoke <literal>ghci</literal> from there (as long as it's in your
  <literal>PATH</literal>).</para>

  <para>If you run GHCi in a Cygwin or MSYS shell, then the Control-C
  behaviour is adversely affected.  In one of these environments you
  should use the <literal>ghcii.sh</literal> script to start GHCi,
  otherwise when you hit Control-C you'll be returned to the shell
  prompt but the GHCi process will still be running.  However, even
  using the <literal>ghcii.sh</literal> script, if you hit Control-C
  then the GHCi process will be killed immediately, rather than
  letting you interrupt a running program inside GHCi as it should.
  This problem is caused by the fact that the Cygwin and MSYS shell
  environments don't pass Control-C events to non-Cygwin child
  processes, because in order to do that there needs to be a Windows
  console.</para>

  <para>There's an exception: you can use a Cygwin shell if the
  <literal>CYGWIN</literal> environment variable does
  <emphasis>not</emphasis> contain <literal>tty</literal>.  In this
  mode, the Cygwin shell behaves like a Windows console shell and
  console events are propagated to child processes.  Note that the
  <literal>CYGWIN</literal> environment variable must be set
  <emphasis>before</emphasis> starting the Cygwin shell; changing it
  afterwards has no effect on the shell.</para>

  <para>This problem doesn't just affect GHCi, it affects any
  GHC-compiled program that wants to catch console events.  See the
  <ulink
  url="../libraries/base/GHC-ConsoleHandler.html">GHC.ConsoleHandler</ulink>
  module.</para>
</sect1>

<sect1 id="terminal-interaction">
<title>
Interacting with the terminal</title>

<para>By default GHC builds applications that open a console window when they start.
If you want to build a GUI-only application, with no console window, use the flag
<literal>-optl-mwindows</literal> in the link step.
</para>

<para>       <emphasis>Warning:</emphasis> Windows GUI-only programs have no
        stdin, stdout or stderr so using the ordinary Haskell
        input/output functions will cause your program to fail with an
        IO exception, such as:
<screen>
      Fail: &lt;stdout&gt;: hPutChar: failed (Bad file descriptor)
</screen>
        However using Debug.Trace.trace is alright because it uses
        Windows debugging output support rather than stderr.</para>

<para>For some reason, Mingw ships with the <literal>readline</literal> library,
but not with the <literal>readline</literal> headers. As a result, GHC (like Hugs) does not
use <literal>readline</literal> for interactive input on Windows.
You can get a close simulation by using an emacs shell buffer!
</para>

</sect1>

<sect1 id="library-differences">
<title>
Differences in library behaviour </title>

<para>
Some of the standard Haskell libraries behave slightly differently on Windows.

<itemizedlist>
<listitem> <para>
On Windows, the '<literal>^Z</literal>' character is interpreted as an
end-of-file character, so if you read a file containing this character
the file will appear to end just before it. To avoid this,
use <literal>IOExts.openFileEx</literal> to open a file in binary
(untranslated) mode or change an already opened file handle into
binary mode using <literal>IOExts.hSetBinaryMode</literal>. The
<literal>IOExts</literal> module is part of the
<literal>lang</literal> package.
</para>
</listitem>
</itemizedlist>
</para>
</sect1>

<sect1 id="ghci-cygwin">
<title>
Using GHC (and other GHC-compiled executables) with cygwin</title>

<sect2>
<title>Background</title> <para>The cygwin tools aim to provide a
unix-style API on top of the windows libraries, to facilitate ports of
unix software to windows. To this end, they introduce a unix-style
directory hierarchy under some root directory (typically
<filename>/</filename> is <filename>C:\cygwin\</filename>). Moreover,
everything built against the cygwin API (including the cygwin tools
and programs compiled with cygwin's ghc) will see / as the root of
their file system, happily pretending to work in a typical unix
environment, and finding things like <filename>/bin</filename> and <filename>/usr/include</filename> without
ever explicitly bothering with their actual location on the windows
system (probably <filename>C:\cygwin\bin</filename> and <filename>C:\cygwin\usr\include</filename>).
</para>
</sect2>

<sect2><title>The problem</title>
<para>GHC, by default, no longer depends on cygwin, but is a native
windows program. It is built using mingw, and it uses mingw's ghc
while compiling your Haskell sources (even if you call it from
cygwin's bash), but what matters here is that - just like any other
normal windows program - neither GHC nor the executables it produces
are aware of cygwin's pretended unix hierarchy. GHC will happily
accept either '/' or '\' as path separators, but it won't know where
to find <filename>/home/joe/Main.hs</filename> or <filename>/bin/bash</filename> 
or the like. This causes all
kinds of fun when GHC is used from within cygwin's bash, or in
make-sessions running under cygwin.
</para>
</sect2>

<sect2><title>Things to do</title>
<itemizedlist>
<listitem>
<para> Don't use absolute paths in make, configure &amp; co if there is any chance 
  that those might be passed to GHC (or to GHC-compiled programs). Relative
  paths are fine because cygwin tools are happy with them and GHC accepts 
  '/' as path-separator. And relative paths don't depend on where cygwin's
  root directory is located, or on which partition or network drive your source
  tree happens to reside, as long as you 'cd' there first.
</para></listitem>

<listitem>
<para> If you have to use absolute paths (beware of the innocent-looking
  <literal>ROOT=`pwd`</literal> in makefile hierarchies or configure scripts), cygwin provides
  a tool called <command>cygpath</command> that can convert cygwin's unix-style paths to their
  actual windows-style counterparts. Many cygwin tools actually accept
  absolute windows-style paths (remember, though, that you either need 
  to escape '\' or convert '\' to '/'), so you should be fine just using those 
  everywhere. If you need to use tools that do some kind of path-mangling 
  that depends on unix-style paths (one fun example is trying to interpret ':' 
  as a separator in path lists..), you can still try to convert paths using 
  <command>cygpath</command> just before they are passed to GHC and friends.
</para></listitem>
  
<listitem>
<para> If you don't have <command>cygpath</command>, you probably don't have cygwin and hence
  no problems with it... unless you want to write one build process for several
  platforms. Again, relative paths are your friend, but if you have to use
  absolute paths, and don't want to use different tools on different platforms,
  you can simply write a short Haskell program to print the current directory
   (thanks to George Russell for this idea): compiled with GHC, this will give 
  you the view of the file system that GHC depends on (which will differ 
  depending on whether GHC is compiled with cygwin's gcc or mingw's
  gcc or on a real unix system..) - that little program can also deal with 
  escaping '\' in paths. Apart from the banner and the startup time, 
  something like this would also do:
<programlisting>
  $ echo "Directory.getCurrentDirectory >>= putStrLn . init . tail . show " | ghci
</programlisting>
</para></listitem>
</itemizedlist>
</sect2>
</sect1>


<sect1 id="win32-dlls">
<title>Building and using Win32 DLLs
</title>

<para>
<emphasis>Making Haskell libraries into DLLs doesn't work on Windows at the
moment; we hope to re-instate this facility in the future.  Note that
building an entire Haskell application as a single DLL is still supported: it's
        just multi-DLL Haskell programs that don't work.  The Windows
        distribution of GHC contains static libraries only.</emphasis></para>

<!--
<para>
<indexterm><primary>Dynamic link libraries, Win32</primary></indexterm>
<indexterm><primary>DLLs, Win32</primary></indexterm>
On Win32 platforms, the compiler is capable of both producing and using
dynamic link libraries (DLLs) containing ghc-compiled code. This
section shows you how to make use of this facility.
</para>

<para>
Until recently, <command>strip</command> didn't work reliably on DLLs, so you
should test your version with care, or make sure you have the latest
binutils. Unfortunately, we don't know exactly which version of binutils
cured the problem (it was supposedly fixed some years ago).
</para>


<sect2 id="win32-dlls-link">
<title>Linking with DLLs</title>

<para>
The default on Win32 platforms is to link applications in such a way
that the executables will use the Prelude and system libraries DLLs,
rather than contain (large chunks of) them. This is transparent at the
command-line, so 
</para>

<para>
<screen>
sh$ cat main.hs
module Main where
main = putStrLn "hello, world!"
sh$ ghc -o main main.hs
ghc: module version changed to 1; reason: no old .hi file
sh$ strip main.exe
sh$ ls -l main.exe
-rwxr-xr-x   1 544      everyone     4608 May  3 17:11 main.exe*
sh$ ./main
hello, world!
sh$ 
</screen>
</para>

<para>
will give you a binary as before, but the <filename>main.exe</filename>
generated will use the Prelude and RTS DLLs instead of linking them in
statically.
</para>

<para>
4K for a <literal>"hello, world"</literal> application&mdash;not bad, huh? :-)
</para>

</sect2>

<sect2 id="win32-dlls-linking-static">
<title>Not linking with DLLs
<indexterm><primary>-static option (Win32)</primary></indexterm></title>

<para>
If you want to build an executable that doesn't depend on any
ghc-compiled DLLs, use the <option>-static</option> option to link in
the code statically.
</para>

<para>
Notice that you cannot mix code that has been compiled with
<option>-static</option> and not, so you have to use the <option>-static</option>
option on all the Haskell modules that make up your application.
</para>

</sect2>
-->

<sect2 id="win32-dlls-create">
<title>Creating a DLL</title>

<para>
<indexterm><primary>Creating a Win32 DLL</primary></indexterm>
<indexterm><primary>&ndash;shared</primary></indexterm>
Sealing up your Haskell library inside a DLL is straightforward;
compile up the object files that make up the library, and then build
the DLL by issuing a command of the form:
</para>

<para>
<screen>
ghc &ndash;shared -o foo.dll bar.o baz.o wibble.a -lfooble
</screen>
</para>

<para>
By feeding the ghc compiler driver the option <option>&ndash;shared</option>, it
will build a DLL rather than produce an executable. The DLL will
consist of all the object files and archives given on the command
line.
</para>

<!--
<para>
To create a `static' DLL, i.e. one that does not depend on the GHC DLLs,
use the <option>-static</option> when compiling up your Haskell code and
building the DLL.
</para>
-->

<para>
A couple of things to notice:
</para>

<para>

<itemizedlist>
<!--
<listitem>
<para>
Since DLLs correspond to packages (see <xref linkend="packages"/>) you need
to use <option>-package-name dll-name</option> when compiling modules that
belong to a DLL if you're going to call them from Haskell. Otherwise, Haskell
code that calls entry points in that DLL will do so incorrectly, and crash.
For similar reasons, you can only compile a single module tree into a DLL,
as <function>startupHaskell</function> needs to be able to call its
initialisation function, and only takes one such argument (see <xref
linkend="win32-dlls-foreign"/>). Hence the modules
you compile into a DLL must have a common root.
</para>
</listitem>
-->

<listitem>
<para>
By default, the entry points of all the object files will be exported from
the DLL when using <option>&ndash;shared</option>. Should you want to constrain
this, you can specify the <emphasis>module definition file</emphasis> to use
on the command line as follows:

<screen>
ghc &ndash;shared -o .... MyDef.def
</screen>

See Microsoft documentation for details, but a module definition file
simply lists what entry points you want to export. Here's one that's
suitable when building a Haskell COM server DLL:

<programlisting>
EXPORTS
 DllCanUnloadNow     = DllCanUnloadNow@0
 DllGetClassObject   = DllGetClassObject@12
 DllRegisterServer   = DllRegisterServer@0
 DllUnregisterServer = DllUnregisterServer@0
</programlisting>
</para>
</listitem>

<listitem>
<para>
In addition to creating a DLL, the <option>&ndash;shared</option> option also
creates an import library. The import library name is derived from the
name of the DLL, as follows:

<programlisting>
DLL: HScool.dll  ==&#62; import lib: libHScool.dll.a
</programlisting>

The naming scheme may look a bit weird, but it has the purpose of allowing
the co-existence of import libraries with ordinary static libraries (e.g.,
<filename>libHSfoo.a</filename> and
<filename>libHSfoo.dll.a</filename>.

Additionally, when the compiler driver is linking in non-static mode, it
will rewrite occurrence of <option>-lHSfoo</option> on the command line to
<option>-lHSfoo.dll</option>. By doing this for you, switching from
non-static to static linking is simply a question of adding
<option>-static</option> to your command line.

</para>
</listitem>
</itemizedlist>
</para>

</sect2>


<sect2 id="win32-dlls-foreign">
<title>Making DLLs to be called from other languages</title>

<para>
If you want to package up Haskell code to be called from other languages,
such as Visual Basic or C++, there are some extra things it is useful to
know.  This is a special case of <xref linkend="ffi-library" />; we'll deal with
          the DLL-specific issues that arise below.  Here's an example:
</para>

<itemizedlist>

<listitem>
<para>
Use <literal>foreign export</literal> declarations to export the Haskell
functions you want to call from the outside. For example,

<programlisting>
module Adder where

adder :: Int -> Int -> IO Int  &ndash;&ndash; gratuitous use of IO
adder x y = return (x+y)

foreign export stdcall adder :: Int -> Int -> IO Int
</programlisting>
</para>
</listitem>

<listitem>
<para>
Compile it up:

<screen>
ghc -c adder.hs -fglasgow-exts
</screen>
  
This will produce two files, adder.o and adder_stub.o
</para>
</listitem>

<listitem>
<para>
compile up a <function>DllMain()</function> that starts up the Haskell
RTS-&ndash;&ndash;a possible implementation is:

<programlisting>
#include &lt;windows.h&gt;
#include &lt;Rts.h&gt;

extern void __stginit_Adder(void);

static char* args[] = { "ghcDll", NULL };
                       /* N.B. argv arrays must end with NULL */
BOOL
STDCALL
DllMain
   ( HANDLE hModule
   , DWORD reason
   , void* reserved
   )
{
  if (reason == DLL_PROCESS_ATTACH) {
      /* By now, the RTS DLL should have been hoisted in, but we need to start it up. */
      startupHaskell(1, args, __stginit_Adder);
      return TRUE;
  }
  return TRUE;
}
</programlisting>

Here, <literal>Adder</literal> is the name of the root module in the module
tree (as mentioned above, there must be a single root module, and hence a
single module tree in the DLL).

Compile this up:

<screen>
ghc -c dllMain.c
</screen>
</para>
</listitem>

<listitem>
<para>
Construct the DLL:

<screen>
ghc &ndash;shared -o adder.dll adder.o adder_stub.o dllMain.o
</screen>

</para>
</listitem>

<listitem>
<para>
Start using <function>adder</function> from VBA-&ndash;&ndash;here's how I would
<constant>Declare</constant> it:

<programlisting>
Private Declare Function adder Lib "adder.dll" Alias "adder@8"
      (ByVal x As Long, ByVal y As Long) As Long
</programlisting>

Since this Haskell DLL depends on a couple of the DLLs that come with GHC,
make sure that they are in scope/visible.
</para>

<para>
Building statically linked DLLs is the same as in the previous section: it
suffices to add <option>-static</option> to the commands used to compile up
the Haskell source and build the DLL.
</para>

</listitem>

</itemizedlist>

</sect2>

<sect2>
<title>Beware of DllMain()!</title>

<para>The body of a <literal>DllMain()</literal> function is an
extremely dangerous place! This is because the order in which DLLs are
unloaded when a process is terminating is unspecified. This means that
the <literal>DllMain()</literal> for your DLL may be called when other DLLs containing
functions that you call when de-initializing your DLL have already
been unloaded. In other words, you can't put shutdown code inside
<literal>DllMain()</literal>, unless your shutdown code only requires use of certain
functions which are guaranteed to be available (see the Platform SDK
docs for more info).</para>

<para>In particular, if you are writing a DLL that's statically
linked with Haskell, it is not safe to call
<literal>hs_exit()</literal> from <literal>DllMain()</literal>, since
<literal>hs_exit()</literal> may make use of other DLLs (see also <xref
          linkend="hs-exit" />).  What's more, if you
wait until program shutdown to execute your deinitialisation code, Windows will have
terminated all the threads in your program except the one calling
<literal>DllMain()</literal>, which can cause even more
problems.</para>

<para>A solution is to always export <literal>Begin()</literal> and <literal>End()</literal> functions from your
DLL, and call these from the application that uses the DLL, so that
you can be sure that all DLLs needed by any shutdown code in your
End() function are available when it is called.</para>

<para>The following example is untested but illustrates the idea (please let us
        know if you find problems with this example or have a better one).  Suppose we have a DLL called Lewis which makes use of 2
Haskell modules <literal>Bar</literal> and <literal>Zap</literal>,
where <literal>Bar</literal> imports <literal>Zap</literal> and is
therefore the root module in the sense of <xref
linkend="using-own-main" />. Then the main C++ unit for the DLL would
look something like:</para>

<programlisting>
 // Lewis.cpp -- compiled using GCC
 #include &lt;Windows.h&gt;
 #include "HsFFI.h"

 #define __LEWIS_DLL_EXPORT
 #include "Lewis.h"

 #include "Bar_stub.h"  // generated by GHC
 #include "Zap_stub.h"

 BOOL APIENTRY DllMain( HANDLE hModule, 
                        DWORD  ul_reason_for_call, 
                        LPVOID lpReserved
                       ){
   return TRUE;
 }

 extern "C"{

 LEWIS_API HsBool lewis_Begin(){
   int argc = ...
   char *argv[] = ...

   // Initialize Haskell runtime
   hs_init(&amp;argc, &amp;argv);

   // Tell Haskell about all root modules
   hs_add_root(__stginit_Bar);

   // do any other initialization here and
   // return false if there was a problem
   return HS_BOOL_TRUE;
 }

 LEWIS_API void lewis_End(){
   hs_exit();
 }

 LEWIS_API HsInt lewis_Test(HsInt x){
   // use Haskell functions exported by
   // modules Bar and/or Zap

   return ...
 }

 } // extern "C"

and some application which used the functions in the DLL would have a main() function like:

 // MyApp.cpp
 #include "stdafx.h"
 #include "Lewis.h"

 int main(int argc, char *argv[]){
   if (lewis_Begin()){
      // can now safely call other functions
      // exported by Lewis DLL

   }
   lewis_End();
   return 0;
 }
</programlisting>

<para><literal>Lewis.h</literal> would have to have some appropriate <literal>#ifndef</literal> to ensure that the
Haskell FFI types were defined for external users of the DLL (who
wouldn't necessarily have GHC installed and therefore wouldn't have
the include files like <literal>HsFFI.h</literal> etc). 
</para>
</sect2>

</sect1>
</chapter>

<!-- Emacs stuff:
     ;;; Local Variables: ***
     ;;; mode: xml ***
     ;;; sgml-parent-document: ("users_guide.xml" "book" "chapter") ***
     ;;; End: ***
 -->