This is intended to be a document to help new developers get started.
Existing developers should feel free to add their comments.

SOURCE TREE STRUCTURE
=====================

Source tree is loosely based on the original Windows modules.  Most
directories are shared between the binary emulator and the library.

Shared directories:
-------------------

KERNEL:

	files/			- file I/O
	loader/			- Win16-, Win32-binary loader
	memory/			- memory management
	msdos/			- DOS and BIOS emulation
	scheduler/		- process and thread management

GDI:

	graphics/		- graphics drivers
	graphics/x11drv/	- X11 display driver
	graphics/metafiledrv/	- metafile driver
	objects/		- logical objects

USER:

	controls/		- built-in widgets
	resources/		- built-in dialog resources
	windows/		- window management

Miscellaneous:

	misc/			- shell, registry, winsock, etc...
	multimedia/		- multimedia driver
	ipc/			- SysV IPC management
	win32/			- misc Win32 functions

Tools:

	rc/			- resource compiler
	tools/			- relay code builder + misc tools
	documentation/		- some documentation


Emulator-specific directories:
------------------------------

	debugger/		- built-in debugger
	if1632/			- relay code
	miscemu/		- hardware instruction emulation
	graphics/win16drv/	- Win16 printer driver

Winelib-specific directories:
-----------------------------

	library/		- Winelib-specific routines (should disappear)
	programs/		- utilities (Progman, WinHelp)
	libtest/		- Winelib test samples


MEMORY AND SEGMENTS
===================

NE (Win16) executables consist of multiple segments.  The Wine loader
loads each segment into a unique location in the Wine processes memory
and assigns a selector to that segment.  Because of this, it's not
possible to exchange addresses freely between 16-bit and 32-bit code.
Addresses used by 16-bit code are segmented addresses (16:16), formed
by a 16-bit selector and a 16-bit offset.  Those used by the Wine code
are regular 32-bit linear addresses.

There are four ways to obtain a segmented pointer:
  - Use the SEGPTR_* macros in include/heap.h (recommended).
  - Allocate a block of memory from the global heap and use
    WIN16_GlobalLock to get its segmented address.
  - Allocate a block of memory from a local heap, and build the
    segmented address from the local heap selector (see the
    USER_HEAP_* macros for an example of this).
  - Declare the argument as 'segptr' instead of 'ptr' in the spec file
    for a given API function.

Once you have a segmented pointer, it must be converted to a linear
pointer before you can use it from 32-bit code.  This can be done with
the PTR_SEG_TO_LIN() and PTR_SEG_OFF_TO_LIN() macros.  The linear
pointer can then be used freely with standard Unix functions like
memcpy() etc. without worrying about 64k boundaries.  Note: there's no
easy way to convert back from a linear to a segmented address.

In most cases, you don't need to worry about segmented address, as the
conversion is made automatically by the callback code and the API
functions only see linear addresses. However, in some cases it is
necessary to manipulate segmented addresses; the most frequent cases
are:
  - API functions that return a pointer
  - lParam of Windows messages that point to a structure
  - Pointers contained inside structures accessed by 16-bit code.

It is usually a good practice to used the type 'SEGPTR' for segmented
pointers, instead of something like 'LPSTR' or 'char *'.  As SEGPTR is
defined as a DWORD, you'll get a compilation warning if you mistakenly
use it as a regular 32-bit pointer.


STRUCTURE PACKING
=================

Under Windows, data structures are tightly packed, i.e. there is no
padding between structure members. On the other hand, by default gcc
aligns structure members (e.g. WORDs are on a WORD boundary, etc.).
This means that a structure like

struct { BYTE x; WORD y; };

will take 3 bytes under Windows, but 4 with gcc, because gcc will add a
dummy byte between x and y. To have the correct layout for structures
used by Windows code, you need to use the WINE_PACKED attribute; so you
would declare the above structure like this:

struct { BYTE x; WORD y WINE_PACKED; };

You have to do this every time a structure member is not aligned
correctly under Windows (i.e. a WORD not on an even address, or a
DWORD on a address that is not a multiple of 4).


NAMING CONVENTIONS FOR API FUNCTIONS AND TYPES
==============================================

In order to support both Win16 and Win32 APIs within the same source
code, as well as share the include files between the emulator and the
library, the following convention must be used in naming all API
functions and types. If the Windows API uses the name 'xxx', the Wine
code must use:

 - 'xxx16' for the 16-bit version,
 - 'xxx32' for the 32-bit version when no ASCII/Unicode strings are
   involved,
 - 'xxx32A' for the 32-bit version with ASCII strings,
 - 'xxx32W' for the 32-bit version with Unicode strings.

You should then use the macros WINELIB_NAME[_AW](xxx) or
DECL_WINELIB_TYPE[_AW](xxx) (defined in include/wintypes.h) to define
the correct 'xxx' function or type for Winelib. When compiling the
emulator, 'xxx' is _not_ defined, meaning that you must always specify
explicitly whether you want the 16-bit or 32-bit version.

Note: if 'xxx' is the same in Win16 and Win32, you can simply use the
same name as Windows.

Examples:

typedef short INT16;
typedef int INT32;
DECL_WINELIB_TYPE(INT);

typedef struct { /* Win32 ASCII data structure */ } WNDCLASS32A;
typedef struct { /* Win32 Unicode data structure */ } WNDCLASS32W;
typedef struct { /* Win16 data structure */ } WNDCLASS16;
DECL_WINELIB_TYPE_AW(WNDCLASS);

ATOM RegisterClass16( WNDCLASS16 * );
ATOM RegisterClass32A( WNDCLASS32A * );
ATOM RegisterClass32W( WNDCLASS32W * );
#define RegisterClass WINELIB_NAME_AW(RegisterClass)

The Winelib user can then say:

    INT i;
    WNDCLASS wc = { ... };
    RegisterClass( &wc );

and this will use the correct declaration depending on the definition
of the symbols WINELIB16, WINELIB32 and UNICODE.


API ENTRY POINTS
================

Because Win16 programs use a 16-bit stack and because they can only
call 16:16 addressed functions, all API entry points must be at low
address offsets and must have the arguments translated and moved to
Wines 32-bit stack.  This task is handled by the code in the "if1632"
directory.  To define a new API entry point handler you must place a
new entry in the appropriate API specification file.  These files are
named *.spec.  For example, the API specification file for the USER
DLL is contained in the file user.spec.  These entries are processed
by the "build" program to create an assembly file containing the entry
point code for each API call.  The format of the *.spec files is
documented in the file "tools/build-spec.txt".


DEBUG MESSAGES
==============

To display a message only during debugging, you normally write something
like this:

#ifdef DEBUG_WIN
	printf("abc...");
#endif

You can write this shorter (and better) in this way:

	dprintf_win(stddeb,"abc...");

All symbols of the form dprintf_xxxx are macros defined in include/debug.h .
The macro-definitions are generated by the shell-script tools/make_debug. It
scans the source code for symbols of this forms and puts the necessary
macro definitions in include/debug.h and include/stddebug.h . These macros
test for the symbol DEBUG_XXXX (e.g. dprintf_win refers to DEBUG_WIN) being 
defined and thus decided whether to actually display the text. If you want
to enable specific types of messages, simply put the corresponding
#define DEBUG_XXXX in include/stddebug.h . If you want to enable or disable
a specific type of message in just one c-source-file, put the corresponding 
#define DEBUG_XXXX or #undefine DEBUG_XXXX between #include<stddebug.h> and
#include <debug.h> in that specific file. In addition you can change the 
types of displayed messages by supplying the "-debugmsg" option to Wine. 
If your debugging code is more complex than just printf, you can use the
symbols debugging_XXX as well. These are true when XXX is enabled, either
permanent or in the command line. So instead of writing

#ifdef DEBUG_WIN
	DumpSomeStructure(&str);
#endif

write
	if(debugging_win)DumpSomeStructure(&str);
Don't worry about the inefficiency of the test. If it is permanently 
disabled (thus debugging_win is 0 at compile time), the compiler will 
eliminate the dead code.

The file handle "stddeb" is intended for displaying standard informational
messages, whereas "stdnimp" is intended for displaying messages concerning
not yet implemented functions.

You have to start tools/make_debug only if you introduced a new macro,
e.g.  dprintf_win32s - not if you just changed one of the #define
DEBUG_XXX's in include/stddebug.h or in a specific file.

MORE INFO
=========

1. There is a FREE online version of the MSDN library (including
   documentation for the Win32 API) on http://www.microsoft.com/msdn/

2. http://www.sonic.net/~undoc/bookstore.html

3. In 1993 Dr. Dobbs Journal published a column called "Undocumented Corner".

4. You might want to check out BYTE from December 1983 as well :-)