Sweden-Number/documentation/winelib-toolkit.sgml

  <chapter id="winelib-toolkit">
    <title id="winelib-toolkit.title">The Winelib development toolkit</title>

    <sect1 id="winemaker">
      <title id="winemaker.title">Winemaker</title>

      <sect2 id="vc-projects">
        <title id="vc-projects.title">Support for Visual C++ projects</title>
        <para>
          Unfortunately Winemaker does not support the Visual C++ project 
          files, ...yet. Supporting Visual C++ project files (the 
          <filename>.dsp</filename> and some <filename>.mak</filename> files 
          for older versions of Visual C++) is definitely on 
          the list of important Winemaker improvements as it will allow it to 
          properly detect the defines to be used, any custom include path, the 
          list of libraries to link with, and exactly which source files to use
          to build a specific target. All things that the current version of 
          Winemaker has to guess or that you have to tell it as will become 
          clear in the next section.
        </para>
        <para>
          When the time comes Winemaker, and its associated build system, will 
          need some extensions to support:
        </para>
        <itemizedlist>
          <listitem>
            <para>
              per file defines and include paths. Visual C++ projects allow 
              the user to specify compiler options for each individual file 
              being compiled. But this is probably not very frequent so it 
              might not be that important.
            </para>
          </listitem>
          <listitem>
            <para>
              multiple configurations. Visual C++ projects usually have at 
              least a 'Debug' and a 'Release' configuration which are compiled 
              with different compiler options. How exactly we deal with these 
              configurations remains to be determined.
            </para>
          </listitem>
        </itemizedlist>
      </sect2>

      <sect2 id="source-analysis">
        <title id="source-analysis.title">Winemaker's source analysis</title>
        <para>
          Winemaker can do its work even without a Windows makefile or a 
          Visual Studio project to start from (it would not know what to do 
          with a windows makefile anyway). This involves doing many educated 
          guesses which may be wrong. But by and large it works. The purpose 
          of this section is to describe in more details how winemaker 
          proceeds so that you can better understand why it gets things 
          wrong and how to fix it/avoid it.
        </para>
        <para>
          At the core winemaker does a recursive traversal of 
          your source tree looking for targets (things to build) and source 
          files. Let's start with the targets.
        </para>
        <para>
          First are executables and DLLs. Each time it finds one of these in 
          a directory, winemaker puts it in the list of things to build and 
          will later generate a <filename>Makefile.in</filename> file in this 
          directory. Note that Winemaker also knows about the commonly used 
          <filename>Release</filename> and <filename>Debug</filename> 
          directories, so it will attribute the executables and libraries 
          found in these to their parent directory. When it finds an 
          executable or a DLL winemaker is happy because these give it more 
          information than the other cases described below.
        </para>
        <para>
          If it does not find any executable or DLL winemaker will look for 
          files with a <filename>.mak</filename> extension. If they are not 
          disguised Visual C++ projects (and currently even if they are), 
          winemaker will assume that a target by that name should be built 
          in this directory. But it will not know whether this target is an 
          executable or a library. So it will assume it is of the default 
          type, i.e. a graphical application, which you can override by using 
          the <option>--cuiexe</option> and <option>--dll</option> options.
        </para>
        <para>
          Finally winemaker will check to see if there is a file called 
          <filename>makefile</filename>. If there is, then it will assume 
          that there is exactly one target to build for this directory. But 
          it will not know the name or type of this target. For the type it 
          will do as in the above case. And for the name it will use the 
          directory's name. Actually, if the directory starts with 
          <filename>src</filename> winemaker will try to make use of the name 
          of the parent directory instead.
        </para>
        <para>
          Once the target list for a directory has been established, 
          winemaker will check whether it contains a mix of executables and 
          libraries. If it is so, then winemaker will make it so that each 
          executable is linked with all the libraries of that directory.
        </para>
        <para>
          If the previous two steps don't produce the expected results (or 
          you think they will not) then you should put winemaker in 
          interactive mode (see <xref linkend="interactive" 
          endterm="interactive.title">). This will allow you to specify the 
          target list (and more) for each directory.
        </para>
        <para>
          In each directory winemaker also looks for source files: C, C++ 
          or resource files. If it also found targets to build in this 
          directory it will then try to assign each source file to one of 
          these targets based on their names. Source files that do not seem 
          to match any specific target are put in a global list for this 
          directory, see the <literal>EXTRA_xxx</literal> variables in the 
          <filename>Makefile.in</filename>, and linked with each of the 
          targets. The assumption here is that these source files contain 
          common code which is shared by all the targets.
          If no targets were found in the directory where these files are 
          located, then they are assigned to the parent's directory. So if a 
          target is found in the parent directory it will also 'inherit' the 
          source files found in its subdirectories.
        </para>
        <para>
          Finally winemaker also looks for more exotic files like 
          <filename>.h</filename> headers, <filename>.inl</filename> files 
          containing inline functions and a few others. These are not put in 
          the regular source file lists since they are not compiled directly. 
          But winemaker will still remember them so that they are processed 
          when the time comes to fix the source files.
        </para>
        <para>
          Fixing the source files is done as soon as winemaker has finished 
          its recursive directory traversal. The two main tasks in this step 
          are fixing the CRLF issues and verifying the case of the include 
          statements.
        </para>
        <para>
          Winemaker makes a backup of each source file (in such a way that 
          symbolic links are preserved), then reads it fixing the CRLF 
          issues and the other issues as it goes. Once it has finished 
          working on a file it checks whether it has done any non 
          CRLF-related modification and deletes the backup file if it did 
          not (or if you used <option>--nobackup</option>).
        </para>
        <para>
          Checking the case of the include statements (of any form, 
          including files referenced by resource files), is done in the 
          context of that source file's project. This way winemaker can use 
          the proper include path when looking for the file that is included. 
          If winemaker fails to find a file in any of the directories of the 
          include path, it will rename it to lowercase on the basis that it 
          is most likely a system header and that all system headers names 
          are lowercase (this can be overridden by using 
          <option>--nolower-include</option>).
        </para>
        <para>
          Finally winemaker generates the <filename>Makefile.in</filename> 
          files and other support files (wrapper files, spec files, 
          <filename>configure.in</filename>, 
          <filename>Make.rules.in</filename>). From the above description 
          you can guess at the items that winemaker may get wrong in 
          this phase: macro definitions, include path, DLL path, DLLs to
          import, library path, libraries to link with. You can deal with
          these issues by using winemaker's <option>-D</>, <option>-P</>,
          <option>-i</>, <option>-I</>, <option>-L</> and <option>-l</>
          options if they are
          homogeneous enough between all your targets. Otherwise you may 
          want to use winemaker's <link linkend="interactive">interactive 
          mode</link> so that you can specify different settings for each 
          project / target.
        </para>
        <para>
          For instance, one of the problems you are likely to encounter is 
          that of the <varname>STRICT</varname> macro. Some programs will 
          not compile if <varname>STRICT</varname> is not turned on, and 
          others will not compile if it is. Fortunately all the files in a 
          given source tree use the same setting so that all you have to do 
          is add <literal>-DSTRICT</literal> on winemaker's command line 
          or in the <filename>Makefile.in</filename> file(s).
        </para>
        <para>
          Finally the most likely reasons for missing or duplicate symbols 
          are:
        </para>
        <itemizedlist>
          <listitem>
            <para>
              The target is not importing the right set of DLLs, or is not
              being linked with the right set of libraries. You can avoid
              this by using winemaker's <option>-P</>, <option>-i</>,
              <option>-L</option> and <option>-l</> options or adding these
              DLLs and libraries to the <filename>Makefile.in</> file.
            </para>
          </listitem>
          <listitem>
            <para>
              Maybe you have multiple targets in a single directory and 
              winemaker guessed wrong when trying to match the source files 
              with the targets. The only way to fix this kind of problem is 
              to edit the <filename>Makefile.in</filename> file manually.
            </para>
          </listitem>
          <listitem>
            <para>
              Winemaker assumes you have organized your source files 
              hierarchically. If a target uses source files that are in a 
              sibling directory, e.g. if you link with 
              <filename>../hello/world.o</filename> then you will get missing 
              symbols. Again the only solution is to manually edit the 
              <filename>Makefile.in</filename> file.
            </para>
          </listitem>
        </itemizedlist>
      </sect2>

      <sect2 id="interactive">
        <title id="interactive.title">The interactive mode</title>
        <para>
          what is it, 
          when to use it,
          how to use it
        </para>
      </sect2>

      <sect2 id="Makefile.in">
        <title id="Makefile.in.title">The Makefile.in files</title>
        <para>
          The <filename>Makefile.in</filename> is your makefile. More 
          precisely it is the template from which the actual makefile will 
          be generated by the <filename>configure</filename> script. It also 
          relies on the <filename>Make.rules</filename> file for most of 
          the actual logic. This way it only contains a relatively simple 
          description of what needs to be built, not the complex logic of 
          how things are actually built.
        </para>
        <para>
          So this is the file to modify if you want to customize things. 
          Here's a detailed description of its content:
        </para>
        <programlisting>
### Generic autoconf variables

TOPSRCDIR             = @top_srcdir@
TOPOBJDIR             = .
SRCDIR                = @srcdir@
VPATH                 = @srcdir@
        </programlisting>
        <para>
          The above is part of the standard autoconf boiler-plate. These 
          variables make it possible to have per-architecture directories for 
          compiled files and other similar goodies (But note that this kind 
          of functionality has not been tested with winemaker generated 
          <filename>Makefile.in</filename> files yet).
        </para>
        <programlisting>
SUBDIRS               =
DLLS                  =
EXES                  = hello.exe
        </programlisting>
        <para>
          This is where the targets for this directory are listed. The names 
          are pretty self-explanatory. <varname>SUBDIRS</varname> is usually 
          only present in the top-level makefile. For libraries and
          executables, specify the full name, including the '.dll' or '.exe'
          extension. Note that these names must be in all lowercase.
        </para>
        <programlisting>
### Global settings

DEFINES               = -DSTRICT
INCLUDE_PATH          =
DLL_PATH              =
LIBRARY_PATH          =
LIBRARIES             =
        </programlisting>
        <para>
          This section contains the global compilation settings: they apply
          to all the targets in this makefile. The <varname>LIBRARIES</>
          variable allows you to specify additional Unix libraries to link
          with. Note that you would normally not specify Winelib libraries
          there. To link with a Winelib library, one uses the <varname>DLLS</>
          variables of the Makefile. The exception is for C++ libraries where
          you currently don't have  a choice but to link with them in the
          Unix sense. One library you are likely to find here is 
          <literal>mfc</> (note, the '-l' is omitted).
        </para>
        <para>
          The other variable 
          names should be self-explanatory. You can also use three additional 
          variables that are usually not present in the file: 
          <varname>CEXTRA</varname>, <varname>CXXEXTRA</varname> and 
          <varname>WRCEXTRA</varname> which allow you to specify additional 
          flags for, respectively, the C compiler, the C++ compiler and the 
          resource compiler. Finally note that all these variable contain 
          the option's name.
        </para>
        <para>
          Then come one section per target, each describing the various 
          components that target is made of.
        </para>
        <programlisting>
### hello.exe sources and settings

hello_exe_C_SRCS          = hello.c
hello_exe_CXX_SRCS        =
hello_exe_RC_SRCS         =
hello_exe_SPEC_SRCS       =
        </programlisting>
        <para>
          Each section will start with a comment indicating the name of the 
          target. Then come a series of variables prefixed with the name of 
          that target. Note that the name of the prefix may be slightly 
          different from that of the target because of restrictions on the 
          variable names.
        </para>
        <para>
          The above variables list the sources that are used togenerate the 
          target. Note that there should only be one resource file in 
          <varname>RC_SRCS</varname>, and that <varname>SPEC_SRCS</varname> 
          will usually be empty for executables, and will contain a single
          spec file for libraries.
        </para>
        <programlisting>
hello_exe_DLL_PATH        =
hello_exe_DLLS            =
hello_exe_LIBRARY_PATH    =
hello_exe_LIBRARIES       =
hello_exe_DEPENDS         =
        </programlisting>
        <para>
          The above variables specify how to link the target. Note that they 
          add to the global settings we saw at the beginning of this file.
        </para>
        <para>
          The <varname>DLLS</> field is where you would enumerate the list of
          DLLs that executable imports. It should contain the full DLL name
          including the '.dll' extension, but not the '-l' option.
        </para>
        <para>
          <varname>DEPENDS</>, when present, specifies a list of other 
          targets that this target depends on. Winemaker will automatically 
          fill this field when an executable and a library are built in the
          same directory.
        </para>
        <programlisting>
hello_exe_OBJS        = $(hello_exe_C_SRCS:.c=.o) \
                        $(hello_exe_CXX_SRCS:.cpp=.o) \
                        $(EXTRA_OBJS)
        </programlisting>
        <para>
          The above just builds a list of all the object files that 
          correspond to this target. This list is later used for the link 
          command.
        </para>
        <programlisting>
### Global source lists

C_SRCS                = $(hello_exe_C_SRCS)
CXX_SRCS              = $(hello_exe_CXX_SRCS)
RC_SRCS               = $(hello_exe_RC_SRCS)
SPEC_SRCS             = $(hello_exe_SPEC_SRCS)
        </programlisting>
        <para>
          This section builds 'summary' lists of source files. These lists are 
          used by the <filename>Make.rules</filename> file.
        </para>
        <note><para>
          FIXME:The following is not up-to-date.
        </para></note>
        <programlisting>
### Generic autoconf targets

all: $(DLLS:%=%.so) $(EXES:%=%.so)

@MAKE_RULES@

install::
        for i in $(EXES); do $(INSTALL_PROGRAM) $$i $(bindir); done
        for i in $(EXES:%=%.so) $(DLLS); do $(INSTALL_LIBRARY) $$i $(libdir); done

uninstall::
        for i in $(EXES); do $(RM) $(bindir)/$$i;done
        for i in $(EXES:%=%.so) $(DLLS); do $(RM) $(libdir)/$$i;done
        </programlisting>
        <para>
          The above first defines the default target for this makefile. Here 
          it consists in trying to build all the targets. Then it includes 
          the <filename>Make.rules</filename> file which contains the build 
          logic, and provides a few more standard targets to install / 
          uninstall the targets. 
        </para>
        <programlisting>
### Target specific build rules

$(hello_SPEC_SRCS:.spec=.tmp.o): $(hello_OBJS)
        $(LDCOMBINE) $(hello_OBJS) -o $@
        -$(STRIP) $(STRIPFLAGS) $@

$(hello_SPEC_SRCS:.spec=.spec.c): $(hello_SPEC_SRCS:.spec) $(hello_SPEC_SRCS:.spec=.tmp.o) $(hello_RC_SRCS:.rc=.res)
        $(WINEBUILD) -fPIC $(hello_LIBRARY_PATH) $(WINE_LIBRARY_PATH) -sym $(hello_SPEC_SRCS:.spec=.tmp.o) -o $@ -spec $(hello_SPEC_SRCS)

hello.so: $(hello_SPEC_SRCS:.spec=.spec.o) $(hello_OBJS) $(hello_DEP
ENDS) 
        $(LDSHARED) $(LDDLLFLAGS) -o $@ $(hello_OBJS) $(hello_SPEC_SRCS:.spec=.spec.o) $(hello_LIBRARY_PATH) $(hello_LIBRARIES:%=-l%) $(DLL_LINK) $(LIBS)
        test -f hello || $(LN_S) $(WINE) hello
        </programlisting>
        <para>
          Then come additional directives to link the executables and 
          libraries. These are pretty much standard and you should not need 
          to modify them.
        </para>
      </sect2>

      <sect2 id="Make.rules.in">
        <title id="Make.rules.in.title">The Make.rules.in file</title>
        <para>
          What's in the Make.rules.in...
        </para>
      </sect2>

      <sect2 id="configure.in">
        <title id="configure.in.title">The configure.in file</title>
        <para>
          What's in the configure.in...
        </para>
      </sect2>
    </sect1>

    <sect1 id="wrc">
      <title id="wrc.title">Compiling resource files: WRC</title>
      <para>
        To compile resources you should use the Wine Resource Compiler, 
        wrc for short, which produces a binary <filename>.res</filename> 
        file. This resource file is then used by winebuild when compiling 
        the spec file (see <xref linkend="spec-file" 
        endterm="spec-file.title">).
      </para>
      <para>
        Again the makefiles generated by winemaker take care of this for you. 
        But if you were to write your own makefile you would put something 
        like the following:
      </para>
      <programlisting>
WRC=$(WINE_DIR)/tools/wrc/wrc

WINELIB_FLAGS = -I$(WINE_DIR)/include -DWINELIB -D_REENTRANT
WRCFLAGS      = -r -L

.SUFFIXES: .rc .res

.rc.res:
	$(WRC) $(WRCFLAGS) $(WINELIB_FLAGS) -o $@ $<
      </programlisting>
      <para>
        There are two issues you are likely to encounter with resource files.
      </para>
      <para>
        The first problem is with the C library headers. WRC does not know 
        where these headers are located. So if an RC file, of a file it 
        includes, references such a header you will get a 'file not found' 
        error from wrc. Here are a few ways to deal with this:
      </para>
      <itemizedlist>
        <listitem>
          <para>
            The solution traditionally used by the Winelib headers is to 
            enclose the offending include statement in an 
            <literal>#ifndef RC_INVOKED</literal> statement where 
            <varname>RC_INVOKED</varname> is a macro name which is 
            automatically defined by wrc.
          </para>
        </listitem>
        <listitem>
          <para>
            Alternately you can add one or more <option>-I</option> directive 
            to your wrc command so that it finds you system files. For 
            instance you may add <literal>-I/usr/include 
            -I/usr/lib/gcc-lib/i386-linux/2.95.2/include</literal> to cater 
            to both C and C++ headers. But this supposes that you know where 
            these header files reside which decreases the portability of your 
            makefiles to other platforms (unless you automatically detect all 
            the necessary directories in the autoconf script).
          </para>
          <para>
            Or you could use the C/C++ compiler to perform the preprocessing. 
            To do so, simply modify your makefile as follows:
          </para>
          <programlisting>
.rc.res:
	$(CC) $(CC_OPTS) -DRC_INVOKED -E -x c $< | $(WRC) -N $(WRCFLAGS) $(WINELIB_FLAGS) -o $@

          </programlisting>
          <!-- FIXME: does this still cause problems for the line numbers? -->
        </listitem>
      </itemizedlist>
      <para>
        The second problem is that the headers may contain constructs that 
        WRC fails to understand. A typical example is a function which return 
        a 'const' type. WRC expects a function to be two identifiers followed 
        by an opening parenthesis. With the const this is three identifiers 
        followed by a parenthesis and thus WRC is confused (note: WRC should 
        in fact ignore all this like the windows resource compiler does). 
        The current work-around is to enclose offending statement(s) in an 
        <literal>#ifndef RC_INVOKED</literal>.
      </para>

      <para>
        Using GIF files in resources is problematic. For best results,
        convert them to BMP and change your <filename>.res</filename>
        file.
      </para>
      <para>
        If you use common controls/dialogs in your resource files, you
        will need to add <function>#include &lt;commctrl.h></function>
        after the <function>#include &lt;windows.h></function> line,
        so that <command>wrc</command> knows the values of control
        specific flags.
      </para>
    </sect1>

    <sect1 id="wmc">
      <title id="wmc.title">Compiling message files: WMC</title>
      <para>
        how does one use it???
      </para>
    </sect1>

    <sect1 id="spec-file">
      <title id="spec-file.title">The Spec file</title>

      <sect2 id="spec-intro">
        <title id="spec-intro.title">Introduction</title>
        <para>
          In Windows the program's life starts either when its 
          <function>main</function> is called,  for console applications, or 
          when its <function>WinMain</function> is called, for windows 
          applications in the 'windows' subsystem. On Unix it is always 
          <function>main</function> that is called. Furthermore in Winelib it 
          has some special tasks to accomplish, such as initializing Winelib, 
          that a normal <function>main</function> does not have to do.
        </para>
        <para>
          Furthermore windows applications and libraries contain some 
          information which are necessary to make APIs such as 
          <function>GetProcAddress</function> work. So it is necessary to 
          duplicate these data structures in the Unix world to make these 
          same APIs work with Winelib applications and libraries.
        </para>
        <para>
          The spec file is there to solve the semantic gap described above. 
          It provides the <function>main</function> function that initializes 
          Winelib and calls the module's <function>WinMain</function> / 
          <function>DllMain</function>, and it contains information about 
          each API exported from a Dll so that the appropriate tables can be 
          generated.
        </para>
        <para>
          A typical spec file will look something like this:
        </para>
        <screen>
init    WinMain
rsrc    resource.res
        </screen>
        <para>
          And here are the entries you will probably want to change: 
        </para>
        <variablelist>
          <varlistentry>
            <term>init</term>
            <listitem>
              <para>
                <literal>init</literal> defines what is the entry point of 
                that module. For a library this is customarily set to 
                <literal>DllMain</literal>, for a console application this 
                is <literal>main</literal> and for a graphical application 
                this is <literal>WinMain</literal>.
              </para>
            </listitem>
          </varlistentry>
          <varlistentry>
            <term>rsrc</term>
            <listitem>
              <para>
                This item specifies the name of the compiled resource file 
                to link with your module. If your resource file is called 
                <filename>hello.rc</filename> then the wrc compilation step
                (see <xref linkend="wrc" endterm="wrc.title">) will generate 
                a file called <filename>hello.res</filename>. This is the 
                name you must provide here. Note that because of this you 
                cannot compile the spec file before you have compiled the 
                resource file. So you should put a rule like the following 
                in your makefile:
              </para>
              <programlisting>
hello.spec.c: hello.res
              </programlisting>
              <para>
                If your project does not have a resource file then you must 
                omit this entry altogether.
              </para>
            </listitem>
          </varlistentry>
          <varlistentry>
            <term>@</term>
            <listitem>
              <note><para>
                FIXME: You must now export functions from DLLs.
              </para></note>
              <para>
                This entry is not shown above because it is not always 
                necessary. In fact it is only necessary to export functions 
                when you plan to dynamically load the library with 
                <function>LoadLibrary</function> and then do a 
                <function>GetProcAddress</function> on these functions. 
                This is not necessary if you just plan on linking with the 
                library and calling the functions normally. For more details 
                about this see: <xref linkend="spec-reference" 
                endterm="spec-reference.title">.
              </para>
            </listitem>
          </varlistentry>
        </variablelist>
      </sect2>

      <sect2 id="spec-compiling">
        <title id="spec-compiling.title">Compiling it</title>
          <note><para>
            FIXME: This section is very outdated and does not correctly
            describe the current use of winebuild and spec files. In
            particular, with recent versions of winebuild most of the
            information that used to be in the spec files is now specified on
            the command line.
          </para></note>
          <para>
            Compiling a spec file is a two step process. It is first 
            converted into a C file by winebuild, and then compiled into an 
            object file using your regular C compiler. This is all taken 
            care of by the winemaker generated makefiles of course. But 
            here's what it would like if you had to do it by hand:
          </para>
          <screen>
WINEBUILD=$(WINE_DIR)/tools/winebuild

.SUFFIXES: .spec .spec.c .spec.o

.spec.spec.c:
	$(WINEBUILD) -fPIC -o $@ -spec $<

.spec.c.spec.o:
	$(CC) -c -o $*.spec.o $<
          </screen>
          <para>
            Nothing really complex there. Just don't forget the 
            <literal>.SUFFIXES</literal> statement, and beware of the tab if 
            you copy this straight to your Makefile.
          </para>
      </sect2>

      <sect2 id="spec-reference">
        <title id="spec-reference.title">More details</title>

        <para>
          Here is a more detailed description of the spec file's format.
        </para>

        <programlisting>
# comment text
        </programlisting>
        <para>
          Anything after a '#' will be ignored as comments.
        </para>

        <programlisting>
init    FUNCTION
        </programlisting>
        <para>
          This field is optional and specific to Win32 modules. It 
          specifies a function which will be called when the DLL is loaded 
          or the executable started.
        </para>

        <programlisting>
rsrc    RES_FILE
        </programlisting>
        <para>
          This field is optional. If present it specifies the name of the 
          .res file containing the compiled resources. See <xref 
          linkend="wrc" endterm="wrc.title"> for details on compiling a 
          resource file.
        </para>

        <programlisting>
ORDINAL VARTYPE EXPORTNAME (DATA [DATA [DATA [...]]])
2 byte Variable(-1 0xff 0 0)
        </programlisting>
        <para>
          This field can be present zero or more times.
          Each instance defines data storage at the ordinal specified. You 
          may store items as bytes, 16-bit words, or 32-bit words.
          <literal>ORDINAL</literal> is replaced by the ordinal number 
          corresponding to the variable. <literal>VARTYPE</literal> should 
          be <literal>byte</literal>, <literal>word</literal> or 
          <literal>long</literal> for 8, 16, or 32 bits respectively. 
          <literal>EXPORTNAME</literal> will be the name available for
          dynamic linking. <literal>DATA</literal> can be a decimal number 
          or a hex number preceded by "0x". The example defines the 
          variable <literal>Variable</literal> at ordinal 2 and containing 
          4 bytes.
        </para>

        <programlisting>
ORDINAL equate EXPORTNAME DATA
        </programlisting>
        <para>
          This field can be present zero or more times.
          Each instance defines an ordinal as an absolute value.
          <literal>ORDINAL</literal> is replaced by the ordinal number 
          corresponding to the variable. <literal>EXPORTNAME</literal> will 
          be the name available for dynamic linking. 
          <literal>DATA</literal> can be a decimal number or a hex number 
          preceded by "0x".
        </para>

        <programlisting>
ORDINAL FUNCTYPE EXPORTNAME([ARGTYPE [ARGTYPE [...]]]) HANDLERNAME
100 pascal CreateWindow(ptr ptr long s_word s_word s_word s_word
                        word word word ptr)
           WIN_CreateWindow
101 pascal GetFocus() WIN_GetFocus()
        </programlisting>
        <para>
          This field can be present zero or more times.
          Each instance defines a function entry point. The prototype 
          defined by <literal>EXPORTNAME ([ARGTYPE [ARGTYPE [...]]])</literal> 
          specifies the name available for dynamic linking and the format 
          of the arguments. <literal>ORDINAL</literal> is replaced
          by the ordinal number corresponding to the function, or 
          <literal>@</literal> for automatic ordinal allocation (Win32 only).
        </para>
        <para>
          <literal>FUNCTYPE</literal> should be one of:
        </para>
        <variablelist>
          <varlistentry>
            <term>pascal16</term>
            <listitem>
              <para>
                for a Win16 function returning a 16-bit value
              </para>
            </listitem>
          </varlistentry>
          <varlistentry>
            <term>pascal</term>
            <listitem>
              <para>
                for a Win16 function returning a 32-bit value
              </para>
            </listitem>
          </varlistentry>
          <varlistentry>
            <term>register</term>
            <listitem>
              <para>
                for a function using CPU register to pass arguments
              </para>
            </listitem>
          </varlistentry>
          <varlistentry>
            <term>interrupt</term>
            <listitem>
              <para>
                for a Win16 interrupt handler routine
              </para>
            </listitem>
          </varlistentry>
          <varlistentry>
            <term>stdcall</term>
            <listitem>
              <para>
                for a normal Win32 function
              </para>
            </listitem>
          </varlistentry>
          <varlistentry>
            <term>cdecl</term>
            <listitem>
              <para>
                for a Win32 function using the C calling convention
              </para>
            </listitem>
          </varlistentry>
          <varlistentry>
            <term>varargs</term>
            <listitem>
              <para>
                for a Win32 function taking a variable number of arguments
              </para>
            </listitem>
          </varlistentry>
        </variablelist>

        <para>
          <literal>ARGTYPE</literal> should be one of:
        </para>
        <variablelist>
          <varlistentry>
            <term>word</term>
            <listitem>
              <para>
                for a 16 bit word
              </para>
            </listitem>
          </varlistentry>
          <varlistentry>
            <term>long</term>
            <listitem>
              <para>
                a 32 bit value
              </para>
            </listitem>
          </varlistentry>
          <varlistentry>
            <term>ptr</term>
            <listitem>
              <para>
                for a linear pointer
              </para>
            </listitem>
          </varlistentry>
          <varlistentry>
            <term>str</term>
            <listitem>
              <para>
                for a linear pointer to a null-terminated string
              </para>
            </listitem>
          </varlistentry>
          <varlistentry>
            <term>s_word</term>
            <listitem>
              <para>
                for a 16 bit signed word
              </para>
            </listitem>
          </varlistentry>
          <varlistentry>
            <term>segptr</term>
            <listitem>
              <para>
                for a segmented pointer
              </para>
            </listitem>
          </varlistentry>
          <varlistentry>
            <term>segstr</term>
            <listitem>
              <para>
                for a segmented pointer to a null-terminated string
              </para>
            </listitem>
          </varlistentry>
          <varlistentry>
            <term>wstr</term>
            <listitem>
              <para>
                for a linear pointer to a null-terminated wide
                (16-bit Unicode) string
              </para>
            </listitem>
          </varlistentry>
        </variablelist>

        <para>
          Only <literal>ptr</literal>, <literal>str</literal>,
          <literal>wstr</literal> and
          <literal>long</literal> are valid for Win32 functions.
          <literal>HANDLERNAME</literal> is the name of the actual Wine 
          function that will process the request in 32-bit mode.
        </para>
        <para>
          Strings should almost always map to str, 
          wide strings - wstr.
          As the general rule it depends on whether the
          parameter is IN, OUT or IN/OUT.
        </para>
        <itemizedlist>
          <listitem>
            <para>
              IN: str/wstr
            </para>
          </listitem>
          <listitem>
            <para>
              OUT: ptr
            </para>
          </listitem>
          <listitem>
            <para>
              IN/OUT: str/wstr
            </para>
          </listitem>
        </itemizedlist>
        <para>
          It is for debug messages. If the parameter is OUT
          it might not be initialized and thus it should not
          be printed as a string.
        </para>
        <para>
          The two examples define an entry point for the 
          <function>CreateWindow</function> and <function>GetFocus</function> 
          calls respectively. The ordinals used are just examples.
        </para>
        <para>
          To declare a function using a variable number of arguments in 
          Win16, specify the function as taking no arguments. The arguments 
          are then available with CURRENT_STACK16->args. In Win32, specify 
          the function as <literal>varargs</literal> and declare it with a 
          '...' parameter in the C file.  See the wsprintf* functions in 
          <filename>user.spec</filename> and 
          <filename>user32.spec</filename> for an example.
        </para>

        <programlisting>
ORDINAL stub EXPORTNAME
        </programlisting>
        <para>
          This field can be present zero or more times.
          Each instance defines a stub function. It makes the ordinal 
          available for dynamic linking, but will terminate execution with 
          an error message if the function is ever called.
        </para>

        <programlisting>
ORDINAL extern EXPORTNAME SYMBOLNAME
        </programlisting>
        <para>
          This field can be present zero or more times.
          Each instance defines an entry that simply maps to a Wine symbol
          (variable or function); <literal>EXPORTNAME</literal> will point 
          to the symbol <literal>SYMBOLNAME</literal> that must be defined 
          in C code. This type only works with Win32.
        </para>

        <programlisting>
ORDINAL forward EXPORTNAME SYMBOLNAME
        </programlisting>
        <para>
          This field can be present zero or more times.
          Each instance defines an entry that is forwarded to another entry
          point (kind of a symbolic link). <literal>EXPORTNAME</literal>
          will forward to the entry point <literal>SYMBOLNAME</literal> 
          that must be of the form <literal>DLL.Function</literal>. This 
          type only works with Win32.
        </para>
      </sect2>
    </sect1>

    <sect1 id="linking">
      <title id="linking.title">Linking it all together</title>
      <note><para>
        FIXME:The following is not up-to-date.
      </para></note>
      <para>
        To link an executable you need to link together: your object files, 
        the spec file, any Windows libraries that your application depends
        on, gdi32 for instance, and any additional library that you use. All 
        the libraries you link with should be available as '.so' libraries. 
        If one of them is available only in '.dll' form then consult 
        <xref linkend="bindlls" endterm="bindlls.title">.
      </para>
      <para>
        It is also when attempting to link your executable that you will 
        discover whether you have missing symbols or not in your custom 
        libraries. On Windows when you build a library, the linker will 
        immediately tell you if a symbol it is supposed to export is 
        undefined. In Unix, and in Winelib, this is not the case. The symbol 
        will silently be marked as undefined and it is only when you try to 
        produce an executable that the linker will verify all the symbols are 
        accounted for.
      </para>
      <para>
        So before declaring victory when first converting a library to 
        Winelib, you should first try to link it to an executable (but you 
        would have done that to test it anyway, right?). At this point you 
        may discover some undefined symbols that you thought were implemented 
        by the library. Then, you to the library sources and fix it. But you 
        may also discover that the missing symbols are defined in, say, 
        gdi32. This is because you did not link the said library with gdi32. 
        One way to fix it is to link this executable, and any other that also 
        uses your library, with gdi32. But it is better to go back to your 
        library's makefile and explicitly link it with gdi32.
      </para>
      <para>
        As you will quickly notice, this has unfortunately not been 
        (completely) done for Winelib's own libraries. So if an application 
        must link with ole32, you will also need to link with advapi32, 
        rpcrt4 and others even if you don't use them directly. This can be 
        annoying and hopefully will be fixed soon (feel free to submit a 
        patch).
      </para>
      <!-- FIXME: try to give some sort of concrete example -->
    </sect1>
  </chapter>

<!-- Keep this comment at the end of the file
Local variables:
mode: sgml
sgml-parent-document:("winelib-user.sgml" "book" "chapter" "")
End:
-->