path: root/doc/hackers.guide.txt

KVIrc hackers guide - Szymon Stefanek - 2004.05.26
-------------------------------------------------------------------------------

This is an always-work-in-progress guide for KVIrc source code hackers.

-------------------------------------------------------------------------------
The source tree
-------------------------------------------------------------------------------

/                         Root directory.
|                         This is almost completely automake and autoconf stuff.
|
|-- admin                 Administrative files used during the compilation
|                         automake, autoconf and document generation scripts.
|                      
|-- data                  Data for the KVIrc program. Most of this stuff is
|   |                     installed in $(prefix)/share/kvirc/$VERSION/
|   |
|   |-- applnk            *.desktop and menu entries for KDE
|   |
|   |-- config            Default configuration files
|   |
|   |-- defscript         The default script
|   |
|   |-- deftheme          The default themes
|   |
|   |-- doctemplates      Some document templates that get parsed by gendoc.pl
|   |                     when the html documentation is generated
|   |
|   |-- helppics          Data pictures for the html documentation
|   |
|   |-- icons             Icons in various sizes
|   |
|   |-- man               The manual pages
|   |
|   |-- mimelnk           *.desktop entries for the *.kvs and *.kvc file types
|   |                     that are respectively kvirc scripts and kvirc
|   |                     configuration files. This is all stuff for KDE
|   |
|   |-- msgcolors         Default sets of message colors
|   |
|   |-- pics              Most of the pictures that KVIrc uses
|   |
|   |-- protocols         irc:// and irc6:// protocol definitions for konqueror
|   |
|   `-- resources         Resources for the windows compilation.
|
|-- debian                Debian mantainer's stuff
|-- debian.robin
|
|-- doc                   Any kind of documentation
|   |
|   `-- scriptexamples    Various script examples
|
|-- no-dist               Stuff that does NOT end in the final distribution

|-- po                    Internationalisation (i18n :)
|   |
|   |-- kvirc             Translations for kvilib and the main KVIrc executable
|   |
|   `-- modules           Translations for the modules
|
|-- scripts               Some SHELL scripts. Probably only config is used atm.
|
|-- src                   The sources
|   |
|   |-- kvilib            KVIrc library. Any source code snippet that can be
|   |   |                 abstracted enough to not depend on the KVIrc core
|   |   |                 ends up here. kvilib depends only on external stuff.
|   |   |
|   |   |-- build         This is the build directory for automake.
|   |   |                 The main Makefile.am is here.
|   |   |
|   |   |-- config        The headers that control the compile-time
|   |   |                 configuration.
|   |   |
|   |   |-- core          The really basic classes: strings, memory management,
|   |   |                 error code defines etc..
|   |   |
|   |   |-- ext           Here ends everything that has no other specific place
|   |   |                 in kvilib.
|   |   |
|   |   |-- file          File management and file utilities
|   |   |
|   |   |-- include       Compile-time generated include files. these are
|   |   |                 only links.
|   |   |
|   |   |-- irc           IRC protocol related classes.
|   |   |
|   |   |-- net           Networking related stuff: sockets, ssl, http ...
|   |   |
|   |   |-- system        System function wrappers or stuff that depends
|   |   |                 on the strict operating system support.
|   |   |                 Threads, localisation, env, time, shared library..
|   |   |
|   |   `-- tal           Toolkit Abstraction Layer: wrapper classes that
|   |                     inherit from KDE* or QT classes, depending on the
|   |                     compilation type.
|   |
|   |-- kvirc             The KVIrc executable sources
|   |   |
|   |   |-- build         Again the build directory for automake.
|   |   |                 Makefile.am is here.
|   |   |
|   |   |-- include       Again compile-time include files. Only links.
|   |   |
|   |   |-- kernel        The core of the executable. The main function is
|   |   |                 here. Here is also the KviApp object and the options
|   |   |                 core management.
|   |   |
|   |   |-- kvs           The NEW shiny scripting engine.
|   |   |                 At the time of writing this two-stage UNICODE
|   |   |                 KVS interpreter is not finished yet and thus almost
|   |   |                 nothing here is really hardwired to the rest
|   |   |                 of the core. 
|   |   |
|   |   |-- module        The module management stuff: the loader, the module
|   |   |                 interface definitions etc..
|   |   |
|   |   |-- sparser       The IRC server parser
|   |   |
|   |   |-- ui            User interface. 99% of the core GUI is here.
|   |   |                 Here you can find KviFrame (the main window)
|   |   |                 KviMdiManager, KviWindow, KviChannel, KviQuery,
|   |   |                 KviConsole, KviInput, KviIrcView and KviUserListView
|   |   |                 which are the most common widgets in kvirc.
|   |   |
|   |   `-- uparser       The currently used scripting engine (user parser).
|   |                     If you want to implement just some simple features
|   |                     (like new commands or functions) then this is the
|   |                     place to look at. If you want to make some long
|   |                     term hacks then it's probably better to look at the
|   |                     kvs directory instead.
|   |
|   `-- modules           Yes, the modules :D
|       |
|       |-- about         The about dialog
|       |
|       |-- aliaseditor   The alias editor window
|       |
|       |-- avatar        Avatar manipulation stuff
|       |
|       |-- chan          $chan.* scripting stuff
|       |
|       |-- channelsjoin  The channelsjoin dialog
|       |
|       |-- clock         This was a clock applet but actually it is not
|       |                 working and thus not compiled.
|       |
|       |-- codetester    The codetester window
|       |
|       |-- config        config.* scripting stuff
|       |
|       |-- dcc           This module implements the whole DCC protocol.
|       |                 Windows, transfer threads etc: everyting is here.
|       |
|       |-- dialog        dialog.* scripting stuff
|       |
|       |-- dockwidget    The dock widget for KDE and windows.
|       |
|       |-- editor        The scripting editor core widget.
|       |
|       |-- eventeditor   The event editor window
|       |
|       |-- file          file.* scripting stuff
|       |
|       |-- filetransferwindow The file transfers window
|       |
|       |-- help          The help browser
|       |
|       |-- http          http.* scripting stuff
|       |
|       |-- ident         A small ident daemon
|       |
|       |-- iograph       Another applet that is not compiled for now.
|       |
|       |-- lag           Lag meter
|       |
|       |-- lamerizer     A crypt/text-transformation engine
|       |
|       |-- links         The links window
|       |
|       |-- list          The channel list window
|       |
|       |-- log           log.* scripting stuff
|       |
|       |-- logview       The logviewer window
|       |
|       |-- mask          mask.* scripting stuff
|       |
|       |-- mircimport    A server entry importer from the mirc's servers.ini
|       |
|       |-- mp3player     mp3player.* scripting stuff
|       |                 This is an interface to the xmms program on unix
|       |                 and to winamp on windows. On unix libxmms.so is
|       |                 loaded at runtime. On windows there is also
|       |                 a gen_kvirc.dll plugin for winamp that needs
|       |                 to be loaded by the winamp program in order to make
|       |                 communications with KVIrc possible.
|       |
|       |-- my            my.* scripting stuff
|       |
|       |-- objects       All the object oriented scripting stuff
|       |
|       |-- options       The options dialog
|       |
|       |-- popupeditor   The popup editor window
|       |
|       |-- raweditor     The raw events editor window
|       |
|       |-- regchan       regchan.* scripting stuff
|       |
|       |-- reguser       reguser.* scripting stuff
|       |
|       |-- rijndael      A crypting engine
|       |
|       |-- setup         The module that is loaded when KVIrc is started
|       |                 for the first time. It contains the initial
|       |                 configuration wizard.
|       |
|       |-- sharedfile    sharedfile.* scripting stuff
|       |
|       |-- sharedfileswindow The shared files window
|       |
|       |-- snd           snd.* scripting stuff
|       |
|       |-- socketspy     The socketspy window
|       |
|       |-- spaste        spaste.* scripting stuff
|       |
|       |-- str           str.* scripting stuff
|       |
|       |-- system        system.* scripting stuff
|       |
|       |-- tb_options    The options toolbar
|       |
|       |-- tb_scripting  The scripting toolbar
|       |
|       |-- tb_winops     The window operations toolbar
|       |
|       |-- term          The embedded terminal emulator (needs KDE)
|       |
|       |-- tip           The tip of the day
|       |
|       |-- tmphighlight  tmphighlight.* scripting stuff
|       |
|       |-- toolbar       toolbar.* scripting stuff
|       |
|       |-- toolbareditor The toolbar editor window
|       |
|       |-- url           The url window
|       |
|       `-- window        window.* scripting stuff
|
`-- win32build            The directory for Windows builds


[pragma@phoenix src]# cat $(find ./ -name \*.h) | wc -l
  60189
[pragma@phoenix src]# cat $(find ./ -name \*.cpp) | wc -l
 164400


-------------------------------------------------------------------------------
The coding style
-------------------------------------------------------------------------------

The coding style helps the reader a lot. In a large project you tend
to forget the exact meaning of some functions or variables.
A good naming convention makes the code "auto commenting": by looking
at the name of a variable or function you can understand its type
and guess its meaning and usage.
Following these rules is not strictly mandatory (maybe with the
exception of the first one) but it is highly appreciated.

- INDENT WITH TABS (the only MANDATORY rule)

	Go back to the line above and read it again.
	
	INDENT, TABS.

	Tabs can be assigned any number of spaces in any decent source code
	editor.
	
	Actually 95% of the KVIrc code is indented in BSD/Allman style
	but the K&R style is also tolerated.

- Try to use the following variable naming conventions

	g_*                : global variables
	m_*                : member variables
	no prefix          : any other scope
	
	[prefix]pName      : pointer to something named Name
	[prefix]iName      : integer (signed) variable named Name
	[prefix]uName      : unsigned integer
	[prefix]szName     : string named Name
	[prefix]dName      : floating point vars
	[prefix]eName      : enumerated value variables
	[prefix]tName      : kvi_time_t values

	i,j,k,tmp,aux,p    : short names are used for short term variables
	                     like the temporaries used in functions.
	                     Do NOT name a member or global variable i.

	So finally:
	
	g_pApp is a global pointer to the application object
	m_pData is a member variable pointer to some data object
	m_szName is a string member variable named Name
	szPippo is a string variable named Pippo
	tmp is a short term temporary variable
	i,j,k are probably some short term iteration variables
	...


- Function names

	C++ class member functions
		Function names start with lower case letters. Each word except the first
		one should start with an upper case letter. Try to use descriptive names
		and not acronyms or shortcuts (unless they are really obvious).
		For example:
			fillUserList, setAutoDelete, joinChannel, markQueryAsDead ...
	Standalone C/C++ functions
		For standalone functions you can follow the C++ rule but the
		kernel-like syntax is also acceptable (all low case letters with
		underscore separators).
		If you're defining a widely used C function (maybe in kvilib)
		then adding a kvi_ prefix is also a good idea.

- Class names

	The class names start with an upper case letter. In most cases
	there is a Kvi prefix and the rest follows the rule for function names.
		KviApp, KviConsole, KviWindow, KviStr, KviConfig, KviUserParser ...
	
	If possible, do not use "shortcut" names.
	Actually KviCommand is preferred over KviCmd unless the KviCmd class
	is REALLY widely used across the source (like KviStr for example).
	
	This helps a lot in remembering the class names: with the shortcuts
	you're often forced to open the corresponding header file to look up
	which letters have been left off...

	Structure names usually follow the same conventions.

- Simple data types

	If you need to define a simple data type then something like kvi_typename_t
	is a good choice.

- Preprocessor

	Preprocessor macros should be all uppercase with undescores separating
	words.

- Comment the code

	You don't need to write poems: two lines describing what a function
	does will be enough.
	If a function is simple and its meaning is clear from its name
	then comments are not needed (this is why we're using expressive
	variable and function names).
	Single line C++ comments are preferred over the C style comments.

-------------------------------------------------------------------------------
Coding tips
-------------------------------------------------------------------------------

- Don't use C++ exceptions: they make the code unmanteinable in the long term

- If you need to access some system function then first look if there is
	an existing kvi_* wrapper and use that one instead. The wrapper is there
	because of portability issues.

- Don't use the STL features: anything that you need IS either in the Qt library
	or in kvilib.

- Windows compilation has COMPILE_ON_WINDOWS #defined and a KDE compilation
	has COMPILE_KDE_SUPPORT #defined.

- Modularize, abstract, modularize, abstract ...

- When your objects need to be allocated with new in a module and destroyed
	in the kvirc core or kvilib (or viceversa) then derive the class from
	KviHeapObject that will provide the new and delete operators.
	This is a workaround for Windows that uses a separate
	heap for each executable module (*.exe or *.dll). Data allocated on one
	heap must be freed on the same heap.

-------------------------------------------------------------------------------
The strings
-------------------------------------------------------------------------------

This is the list of the various string types used in KVIrc.

(const) char *
	The classic C null terminated string.
	
KviStr
	The basic KVIrc string class. It has been first implemented as a hack
	around various bugs of the original QString class (the NOT unicode one
	that now has been renamed to QCString). It has the property of being
	always non null and it has no reference counting.
	Actually many occurences of this string are replaced by QString
	(especially in GUI modules) to handle correctly the UNICODE character set.

QString
	The Qt UNICODE string. See the Qt documentation for details.
	This is the string that should be mostly used in KVIrc in the near future.
	Take care: in general it is NOT null terminated.
	There is a KviQString wrapper namespace (#include "kvi_qstring.h") that
	adds some missing functionality. For example,
		KviQString::sprintf(qstring_buffer,qstring_format,...)
	allows formatting a QString with a format string that is a QString itself
	(and thus it is UNICODE).
	QString uses reference counting. An assigment of a QString to another
	QString does NOT make an immediate copy, it just increases the reference
	count instead. The copy is made at the first modification of one of the
	two strings (the operation is called "detaching"). While generally this
	is not an issue, you must take care when passing QString objects between
	concurrent threads.

(const) QChar * 
    The array of Qt chars. This is usually obtained by callling
    KviQString::nullTerminatedArray() which itself is a hack...
    This array is used in some functions that were written for
    const char * strings and haven't been ported completely.

QCString
	The Qt non UNICODE string. See the Qt documentation for details.

The Goal:
	- Use KviStr only where it is strictly needed (for protocol or performance
		related issues). One of such places is the IRC server parser (but there
		are more).
	- Use QString everywhere in the user interface and in any other
		place where KviStr is not strictly needed. Save and restore
		strings in the UTF8 format.
	- Get rid of ALL occurences of KviWStr and kvi_wchar_t * : DONE on 2004.11.02

-------------------------------------------------------------------------------
Strings and localisation
-------------------------------------------------------------------------------

Any string that is shown to the user should be translated to the user's local
language. To make a string translaetable use one of the __tr* macros.
The most common one across the sources is __tr("string") that returns
a const char * translation of "string".
Actually __tr() is being phased out in favor of __tr2qs() that returns
a QString instead of a const char * pointer.
The arguments of these macros are extracted from the sources by the
gettext program and are used to build the translation hashes loaded at runtime.
Remember that the arguments must be string constants and not variables.

The list that follows describes briefly the localisation macros
defined in kvi_locale.h

CSTRING is an US-ASCII null terminated C string.

__tr2qs(CSTRING) : translates CSTRING to a QString &
__tr(CSTRING) : translates CSTRING to another CSTRING
	This should disappear in favor of __tr2qs

These macros are NOT THREAD SAFE: you can't call them from non GUI threads.
If you need to translate some string in a slave thread (probably when
sending a message event to the main GUI thread) then you need to use the
__tr_no_lookup() (on the slave side) and __tr_no_xgettext() (on the master side).


-------------------------------------------------------------------------------
Anatomy of an IRC context
-------------------------------------------------------------------------------


KviIrcContext [persistent set of resources]
  |
  +--KviIrcConnection [changed at every connection, with (almost) all the children]
  |     |
  |     +--KviIrcConnectionTarget [target server, proxy to use and address to bind]
  |     |     |
  |     |     +--KviIrcServer
  |     |     |
  |     |     +--KviProxy [null if not using a proxy]
  |     |
  |     +--KviIrcLink [high level network link: trasmits and receives IRC messages]
  |     |     |
  |     |     +--(KviIrcConnectionTargetResolver) [kickstarts the connection]
  |     |     |
  |     |     +--KviIrcSocket [low level network link: transmits packets of bytes]
  |     |
  |     +--KviPtrList<KviChannel> [active channels]
  |     |
  |     +--KviPtrList<KviQuery> [active queries]
  |     |
  |     +--KviIrcConnectionUserInfo [nick, user, host, local ip...]
  |     |
  |     +--KviIrcConnectionServerInfo [name, supported modes, supported flags...]
  |     |
  |     +--KviNotifyListManager [kvi_notifylist.h]
  |     |
  |     +--...
  |
  +--KviConsole [persistent]
  |
  +--(KviLinksWindow), (KviListWindow) [other may-be-persistent context windows]
  |
  +--KviPtrList<KviChannel> [dead channels]
  |
  +--KviPtrList<KviQuery> [dead queries]
  |
  +--...

KviIrcContext is the set of resources used to deal with a single irc
connection. An irc context is persistent and reusable until the user decides to
destroy it. The irc context owns the console window (KviConsole) that is
strictly tied to the lifetime of the context itself. The console is created
when the IRC context is created and when the user closes the console then
the IRC context is destroyed too.
In earlier KVIrc versions there was only KviConsole that did the role of both
KviConsole and KviIrcContext, but since the class has grown in complexity
to a point where it started to be unmantainable the splitting has been unavoidable.

KviIrcConnection rappresents an IRC connection: it is the highest protocol
implementation on the KVIrc's networking stack. A KviIrcConnection
owns a KviIrcLink (that is the lower level). KviIrcConnection is NOT reusable:
it lives only for the lifetime of a single IRC connection inside the parent
irc context. KviIrcConnection talks to the parent's KviConsole.
The connection target is a KviIrcConnectionTarget class and it contains
the KviIrcServer, KviProxy and the eventual bind address. The owned target
is passed down the networking stack to the lower level classes.
The connection contains also the lists of queries and channels currently opened.
When a channel or query is marked as dead then its ownership is passed to
the KviIrcContext (it becomes permanent between two connections).
The connection owns a lot of other interesting classes to take a look at:
KviIrcConnectionUserInfo, KviIrcConnectionServerInfo, KviNotifyListManager...

KviIrcLink is the middle level of the KVIrc's networking stack.
It handles host lookups, the connection startup and data stream input and output.
This is meant to be a "pluggable" class: it should be flexible enough to allow
inheritance and protocol overriding. KviIrcLink owns and manages the KviIrcSocket.
It takes care of extracting IRC protocol messages from the KviIrcSocket raw data
stream and of formatting the outgoing messages by adding the trailing CRLF.
The host lookups are done by the means of KviIrcConnectionTargetResolver.

KviIrcSocket is the lowest level of the KVIrc's networking stack.
It manages the connection through proxies and accesses the system level
socket directly. The incoming data stream is passed to KviIrcLink::processData()
and the outgoing data stream is received through KviIrcSocket::sendPacket()
KviIrcSocket also manages the outgoing send queue and implements the
"anti-server-flood" algorithm.
This class doesn't know anything about the IRC protocol: it just receives
and sends out raw data packets!


.......

kvirc (KviApp)
  |
  +-frame window (KviFrame)
     |
     +-irc_context 1 (KviIrcContext)
     |  |
     |  +-irc_connection (KviIrcConnection)
     |  |  |
     |  |  +-list of channels
     |  |  |
     |  |  +-list of queries
     |  |  |
     |  |  +-irc_link
     |  |
     |  +-console (KviConsole)
     |
     +-irc_context 2
     | ...


KviConsole <-> KviIrcContext

-------------------------------------------------------------------------------
Important global variables
-------------------------------------------------------------------------------

All these variables are almost alwas set (and point to a real alive object).
The only critical moments where these variables must be double checked
are the startup phase and the shutdown phase.
Do not attempt to change the values of these variables unless you REALLY know
what you're doing.


KviApp * g_pApp;
	The one and only application object
	Declared in "kvi_app.h"
	Always set.

KviServerParser * g_pServerParser;
	The one and only server parser
	Declared in "kvi_sparser.h"
	Almost always set (critical phases at early startup and late shutdown)
	
KviFrame * g_pFrame;
	The one and only main window
	Declared in "kvi_frame.h"
	Almost always set (critical phases at early startup and late shutdown)
	
KviWindow * g_pActiveWindow;
	The one and only active window
	Declared in "kvi_window.h"
	Almost always set (critical phases at early startup and late shutdown)

Note for C++ purists: In fact we could be using the protected singleton pattern
on most of these variables and access it by the means of Class::instance().
The global var names save some typing and can be written by any other class
without having to worry about friends or write-access functions.

-------------------------------------------------------------------------------
The charset mess
-------------------------------------------------------------------------------

IRC is not UNICODE :/ ... sigh ...
The fact is that every user wants his local encoding to be used.
KVIrc tries to be even smarter and allow a different encoding for each window.
This is a difficult task since we simply can't translate the strings that
come from and go to the server just at the socket level.

User -> Server

We need to allow the local user to write UNICODE data, encode it to the
proper charset (again depending on the window the text was typed in) and
send it down to the server.
When the user writes commands this is going to become a little mess since
nicknames, channel names or usernames may or may not be encoded in the
encoding of the current window.

Server -> User

We need to carry the plain 8bit data (in whatever encoding it is) from the
server up to the GUI level, then convert to UNICODE by choosing the proper
decode routine just when we know in which window the text is going to be
displayed. In (non RFC) servers that allow encoded characters in nicknames
this is going to become a real mess since the same 8bit nick may result in
a different UNICODE string depending on the window it was "decoded" on.

(Partial) Solution:
- Each server has an encoding set. If empty then the network encoding is used.
- Each network han and encoding set. If empty then the default system encoding
  is used.
- The system encoding is set by the user. If empty then the encoding is guessed
  from the user's locale.
- Each window (with the exception of the console) has its own encoding used
  ONLY for private messages and notices. This allows one to join
  a channel with a "special" encoding and still see what's being written in.
  The real utility of this last feature still needs to be evaluated.

-------------------------------------------------------------------------------
Output levels
-------------------------------------------------------------------------------

There are few macros that specify the output level that the user desires.
These marcors are defined in "kvi_options.h"

_OUTPUT_MUTE: returns true if the user wants KVIrc to spit less useless output
      possible. The goal of the user is to chat on IRC so print only data
      relevant to this. If stuff goes wrong then print the errors in
      short forms (one liners) and do it only in case of serious ones.
      Don't print any transient error or warning.
      Usage: if(!_OUTPUT_MUTE)output...

_OUTPUT_QUIET: returns true if the uses wants KVIrc to spit less output than
      normal. The goal of the user is to chat visually on IRC so print
      only data relevant to this.
      Usage: if(!_OUTPUT_QUIET)output...

<normal level>: Reference output level: here stuff is printed unconditionally.
      Usage: output...

_OUTPUT_VERBOSE: returns true if the users allows KVIrc to print some
      additional output. This is intended mainly for scripters and
      curious pepole that want detailed informations about what is going
      on around them.
      Usage: if(_OUTPUT_VERBOSE)output...

_OUTPUT_PARANOIC: returns true if the users allows KVIrc to print anything
      including debug info. This is intended mainly for developers.
      Usage: if(_OUTPUT_PARANOIC)output...

-------------------------------------------------------------------------------
Rule for safe text output
-------------------------------------------------------------------------------

If the format string you're going to output is not constant (i.e. it
comes from the server) you MUST use KviWindow::outputNoFmt() instead 
of KviWindow::output().

BAD:
	QString szText = pConnection->decodeText(msg.safeTrailing());
	pWindow->output(KVI_OUT_SOMETHING,szText); <--potential crash/security hole

GOOD:
	QString szText = pConnection->decodeText(msg.safeTrailing());
	pWindow->outputNoFmt(KVI_OUT_SOMETHING,szText); <--faster and no crashes

-------------------------------------------------------------------------------
KVIrc (and script) versioning
-------------------------------------------------------------------------------

- Standard definition

The KVIrc versioning follows a really common standard: we use a
string of numbers separated by dots with decreasing weight from left to right.

	<N1>.<N2>.<N3>.<N4>.....

where each <NX> is a number.

Theoretically there is no limit on the parts the version can be composed
of but in fact we use either three or four part versions. The omitted
parts on the right are implicitly assumed to be 0.

The first part is called the major release number and it is bumped
up only when really big changes occur in the source tree. A bump from N to N+1
in the major version number means that a great milestone has been achieved
and the software is really different from what it was in the moment
when the major number was bumped from N-1 to N. This usually also means
that the software might be somewhat incompatible with the previous major release.

When the major number is bumped up all the following parts are reset to 0
(and could be even temporairly omitted).

The second part is called minor release number and it is increased
more often than the major. A bump from N to N+1 in the minor version number
means that an ordinary (small) development milestone has been achieved.
Software versions with the same major and close minor numbers are likely
to be totally compatible with each other.

When the minor number is bumped up all the following parts are reset to 0.

The third part is called (public) revision number and it is increased often.
A bump from N to N+1 in the revision number usually means that a set of bugfixes
or some new features have been included in the software. Compatibility
should be assumed unless explicitly noted. Again, when the revision number
is bumped up all the following parts are reset to 0.

The fourth part is actually used only on the svn tree and it is usually not
present in the official public releases (it is assumed to be 0 for comparison
purposes). It is called the "internal revision" number and when taken
out of the version string it may assume a meaning on its own (but it's not
required in fact). KVIrc uses the ISO sources date in the format YYYYMMDD for this
number. The sources date number is defined in src/kvilib/config/kvi_sourcesdate.h
and is also displayed by kvirc --version. Some packagers prefer to use the svn
revision number instead of the sources date. This is not "official" but it's
still ok as long as it follows the "order-preserving" rule (see below).

It is unlikely that you will find a KVIrc versioned with more than
four numbers... but if you will (for some strange reason) then it will
still follow the same rules: it will be increased for yet minor changes
(two versions within a single day ?) and will be reset to (implicit)
zero when the fourth part changes.

- Comparison of version strings

The comparison of two version strings is defined as follows.
Let N1.N2.N3.N4.N5..... and M1.M2.M3.M4.M5..... be version strings.
To find out which one is greater compare each couple of numbers Ni-Mi at the same
position i (with the same weight) until Ni and Mi differ or both Ni and
Mi are omitted. If both Ni and Mi are omitted then the version strings
are equal, otherwise the greater version string is the one that
contains the greater of the Ni - Mi couple. Easy, right ?

This means that to compare 3.2.6.3.4 and 3.2.9 you first compare
3 with 3 and find that they are equal. Then you compare 2 with 2
and find that they are equal. Then compare 6 with 9 and find that
they are different and 9 is greater. This allows you to say that
the first version string is greater than the second.

To compare 3.2.6.1 with 3.2.6 you compare 3 and 3, 2 and 2, 6 and 6
and 1 with (implicit) 0, that tells you that the first version string
is greater than the second one.

This also means that 3 and 3.0.0 are assumed to be EQUAL since
the algorithm above finds that at the fourth comparison step
both numbers are omitted (thus zero from there up to infinity).

This comparison function is monotonically increasing or in
other words order-preserving. This is a *requirement* for a consistent
versioning scheme.

- Package versioning schemes with letters

It is common for packages to add letters to some of the parts
of the version string. This causes the string to lose the
advantage of being universally comparable but it might still
define a consistent scheme for some package line. In the case
that letters are added to a version string (like 3.2.6.svn10)
we say that it is comparable only to the version strings
that have the same letter pattern: 3.2.6.svn10 and 4.3.1.svn344
are comparable but 3.2.6.svn10 and 3.2.6.cvs15 are not.
For comparable strings we strip the letters in order to make the comparison.

- Stable and unstable versions

Our numbering scheme does NOT tell which versions are stable
and which are unstable. The versions are declared to be (more or less) stable
by other means (read: the mailing list and the www site).
It is true, tough, that stable versions are likely to have
more numbers omitted (read: 3.5 looks more "stable" than 3.4.5.43),
but this is not a strict requirement. It is also true that
almost all "unstable" versions come out from the svn tree and
usually contain all the four parts.

- Official, semi-official and unofficial packages

We tend to have three types of packages. The official packages are the
ones considered to be stable and released on the site in all the supported forms
(source and various kinds of binaries). The official releases are also announced
in tracker sites and spread between distributors. Since "most" stable, the
official releases are the ones likely to be included in the OS distributions.
The official packages have md5 sums and a gpg signature of one of the KVIrc
developers (with a public key available from a "trusteable" location such
as the KVIrc web site).

The semi-official packages are snapshots of the source tree made when
some important changes have occured. They are announced on the KVIrc site
only and are likely to be stable (but are not declared officially to be so).
The semi-official packages usually are at least in the source form but
there are likely to be some binaries available too.
We do not sign the semi-official packages but it's still somewhat
granted that WE (the KVIrc Development Team) make them and thus the
source can be trusted.

The unofficial packages are random snapshots of the svn source tree made
by anyone who wants to do it at any time. They are not announced on the KVIrc
web site (but might be uploaded to ftp.kvirc.net) but rather announced and available
at some other internet location. Since we always try to keep the svn
tree clean and compilable they should work fine but there is no guarantee.
There is no rule for the unofficial package format: there might be source-only
or binary only packages. We *suggest* to use the four part version string format
but in fact they might contain third party patches and even follow their own
derived version numbering scheme. The general rule is: we don't control
the unofficial packages and don't provide support if they don't compile/work
as expected the user/packager is on his own.. but we'll try to be helpful,
if possible :)

- Putting it all together

At the moment of writing the KVIrc svn tree has version 3.2.6.20070115.
Today is 15 Jan 2007 and the current svn revision number is 174.
The latest "semi-official" release was 3.2.6 (where the internal revision number
was omitted and implicitly assumed to be 0).
If tomorrow we had to emit a quick fix for the semi-official release we'd either
use 3.2.7 (increasing the public revision for a large set of bugfixes) or
(unlikely) 3.2.6.20060116 (expliciting the internal revision number for a small
set of quick, and probably dirty, bugfixes). The next "semi-official"
release is likely to be 3.2.7 while the next official stable release
might be 3.3, 3.4 or even 4.

-------------------------------------------------------------------------------
Porting to Qt 4.x (while mantaining Qt 3.x compatibility) (Work In Progress)
-------------------------------------------------------------------------------

For the moment, in random order:

- Avoid using QString::null, use KviQString::empty instead.
  This is because Qt 4.x does NOT have a null static variable. Qt 4.x in fact
  does not have the distinction between null and empty strings (Note that for
  KviStr this choice was made since the beginning).
  Do NOT replace all the uses of QString::null with QString() (as the qt 4.x
  porting documentation suggests) since for Qt 3.x this construct is SLOW.

- We're building a compatibility layer in kvilib.
  Before using ANY Qt class, look if there is an override in kvilib.
  In fact always, prefer Kvi* classes over the Q* ones and include
  the "kvi_*.h" files instead of the <qt*.h> ones.

- Widgets will be probably abstracted in kvilib/tal.

- Qt 4.x has no QCString anymore. It uses QByteArray instead. From my own
  point of view this is ugly since the name is misleading and QByteArray
  is forced to mantain the null terminator which is not needed
  in normal "byte array" operations...but well, we have to live with it.
  kvilib has kvi_qcstring.h that defines the KviQCString MACRO.
  It expands to QCString under Qt 3.x and QByteArray under Qt 4.x.

- Use:
    #ifdef COMPILE_USE_QT4
  to check if the compilation requires Qt 4.x or Qt 3.x.
  Later this *might* be automatically replaced with
    #if QT_VERSION >= 0x040000

- Use:
    ./configure --enable-debug --enable-qt4 --with-qt4-moc=/path/moc
    make clean
    make
  to switch to Qt 4.x
  Use:
    ./configure --enable-debug
    make clean
    make
  to switch back to Qt 3.x  

- Use less possible Qt3Compat features.

- Before committing to the svn always test your changes under Qt 3.x: this is
  the primary goal for now: keeping 3.x compatibility.
  Qt 3.x-only developers are encouraged (but not required) to test their
  changes agains Qt 4.x.

- Use the KviQString::* wrapper functions around QString (take a look at
  kvi_qstring.h) even if there exist QString functions that do the same
  thing. This is because the QString interface has changed a lot between
  Qt 3.x and Qt 4.x and KviQString introduces a compatibility layer.

- Do not use the "char * c = KviQString::toUtf8(string).data();" construct.
  It leads to crashes on many compilers since the returned KviQCString
  goes out of scope just at the end of the instruction.

-------------------------------------------------------------------------------
Code documentation
-------------------------------------------------------------------------------

KVIrc is LARGE. We need to start documenting the source code if we want
to understand our own code in a year from now and if we want help from
others.

Use doxygen.

There is a Doxyfile in the admin subdirectory. You can either
run doxygen from there or simply type "make devdocs" from the
top directory of the source tree.
Then take a look at doc/api/html/annotated.html

Let's also try to document the code we write: the doxygen syntax
is trivial and you can find a 5 minute tutorial by googling.