lyx_mirror/development/Code_rules/Rules

Rules for the code in LyX
-------------------------
[updated from the C++STYLE distributed with the GNU C++ Standard]

The aim of this file is to serve as a guide for the developers, to aid us to
get clean and uniform code. This document is still incomplete.

We really like to have new developers joining the LyX Project. However
since we have had problems in the past with developers leaving the
project and their contributed code in a far from perfect state. Most
of this happened before that we really became aware of these issues,
but still, we don't want it to happen again. So we have put together
some guidelines and rules for the developers.

General
-------

These guidelines should save us a lot of work while cleaning up the code and 
help us to have quality code. LyX has been haunted by problems coming from 
unfinished projects by people who have left the team. Those problems will 
hopefully disappear if the code is easy to hand over to somebody else.

In general, if you want to contribute to the main source, we expect at least 
that you:

- the most important rule first: kiss (keep it simple stupid), always
  use a simple implementation in favor of a more complicated one.
  This eases maintenance a lot.
- write good C++ code: Readable, well commented and taking advantage of the 
  OO model. Follow the formatting guidelines. See Formatting.
- adapt the code to the structures already existing in LyX, or in case that 
  you have better ideas, discuss them on the developer's list before writing 
  the code.
- take advantage of the C++ standard library. especially don't use
  custom containers when a standard container is usable, learn to use
  the algorithms and functors in the standard library.
- be aware of exceptions and write exception safe code. See Exceptions.
- document all variables, methods, functions, classes etc. We are
  using the source documentation program doc++, a program that handles
  javadoc syntax, to document sources. See Source Documentation.
- we have certain code constructs that we try to follow. See Code
  Constructs.


Submitting Code
---------------

It is implicitly understood that all patches contributed to The LyX
Project is under the Gnu General Public License, it you have a problem
with that, don't contribute code.

Also please don't just pup up out of the blue with a huge patch (or
small) that changes something substantial in LyX. Always discuss your
ideas with the developers on the developers mailing list.

When you create the patch, please use "diff -up" since we find that a
lot easier to read than the other diff formats. Also please do not
send patches that implements/fix several different things, several
patches is a much better option.

We also expect you to provide a ChangeLog entry with every patch, this
describes shortly what the patch is doing. The ChangeLog entry follows
this syntax:

1999-12-13  Lars Gullik Bj<42>nnes  <larsbj@lyx.org>

	* src/support/lyxstring.C (find): assert bug fixed.


Code Constructs
---------------

We have several guidelines on code constructs, some of these exists to
make the code faster, others to make the code clearer. Yet others
exists to make us able to take advantage of the strong type checking
in C++.
  
- Declaration of variables should wait as long as possible. The rule
  is: "Don't declare it until you need it." In C++ there are a lot of
  user defined types, and these can very often be expensive to
  initialize. This rule connects to the next rule too.

- Make the scope of a variable as small as possible.

- Prefer preincrement to postincrement whenever possible.
  Preincrement has potential of being faster than postincrement. Just
  thing about the obvious implementations of pre/post-increment. This
  rule applies to decrement too.

	++T;
	--U;
	-NOT-
	T++; // wrong
	U--; // wrong

- Try to minimize evaluation of the same code over and over. This is
  aimed especially at loops.

	Container::iterator end = large.end();
	for (Container::iterator it = large.begin(); it != end; ++it) {
		...;
	}
	-NOT-
	for (Container::iterator it = large.begin();
	     it != large.end(); ++it) {
		...;
	}

- For functions and methods that return a non-POD type T, return T
  const instead. This gives better type checking, and will give a
  compiler warning when temporaries are used wrongly.

	T const add(...);
	-NOT-
	T add(...);

- Avoid using the default cases in switch statements unless you have
  too. Use the correct type for the switch expression and let the
  compiler ensure that all cases are exhausted. 

	enum Foo {
		foo,
		bar
	};
	Foo f = ...;
	switch (f) {
	case foo: ...; break;
	case bar: ...; break;
	default: ...; break; // not needed and would shadow a wrong use of Foo
	}

Exceptions
----------

Even if LyX currently is not using exceptions we need to be aware of
them. One important thing to realize is that you often do not have to
use throw,try or catch to be exception safe. Let's look at the
different types of exceptions safety: (These are taken from Herb
Sutters book[ExC++]

"
1. Basic guarantee: Even in the presence of exceptions thrown by T or
	other exceptions, Stack objects don't leak resources.
	Note that this also implies that the container will be
	destructible and usable even if an exception is thrown wile
	performing some container operation. However, if an exception
	is thrown, the container will be in a consistent, but not
	necessarily predictable, state. Containers that support the
	basic guarantee can work safely in some settings.

2. Strong guarantee: If an operation terminates because of an
	exception, program state will remain unchanged.
	This always implies commit-or-rollback semantics, including
	that no references or iterators into the container be
	invalidated if an operation fails. For example, if a Stack
	client calls Top and then attempts a Push that fails because
	of an exception, then the state of the Stack object must be
	unchanged and the reference returned from the prior call to
	Top must still be valid. For more information on there
	guarantees, see Dave Abrahams's documentation of the SGI
	exception-safe standard library adaption at:
	http://www.metabyte.com/~fbp/stl/eg_contract.html 

	Probably the most interesting point here is that when you
	implement the basic guarantee, the strong guarantee often
	comes for free. For example, in our Stack implementation,
	almost everything we did was needed to satisfy just the basic
	guarantee -- and what's presented above very nearly satisfies
	the strong guarantee, with little of no extra work. Not half
	bad, considering all the trouble we went to.

	In addition to these two guarantees, there is one more
	guarantee that certain functions must provide in order to make
	overall exception safety possible: 

3. Nothrow guarantee: The function will not emit an exception under any
   	circumstances.
	Overall exception safety isn't possible unless certain
	functions are guaranteed not to throw. In particular, we've
	seen that this is true for destructors; later in this
	miniseries, we'll see that it's also needed in certain helper
	functions, such as Swap(). 
"

For all cases where we might be able to write exception safe functions
without using try, throw or catch we should do so. In particular we
should look over all destructors to ensure that they are as exception
safe at possible.

Later when more compiler support exceptions sufficiently well we will
begin using them too. One reason for this is that the C++ standard
library actually requires exceptions, e.g. "new" will throw
bad_allocation if the requested memory is not available.


Formatting
----------

* Only one declaration on each line.
	int a;
	int b;
	-NOT-
	int a, b; // wrong
  This is especially important when initialization is done at the same
  time:
	string a("Lars");
	string b("Gullik");
	-NOT-
	string a("Lars"), b("Gullik"); // wrong

* Pointers and references
	char * p = "flop";
	char & c = *p;
	-NOT-
	char *p = "flop"; // wrong
	char &c = *p;     // wrong

  Some time ago we had a huge discussion on this subject and after
  convincing argumentation from Asger this is what we decided. Also note
  that we will have:
	char const * p;
	-NOT-
	const char * p; // wrong

* Operator names and parentheses
	operator==(type)
	-NOT-
	operator == (type)  // wrong

  The == is part of the function name, separating it makes the
  declaration look like an expression.

* Function names and parentheses
	void mangle()
	-NOT-
	void mangle ()  // wrong

* Enumerators
	enum {
		one = 1,
		two = 2,
		three = 3
	};
	-NOT-
	enum { one = 1, two = 2, three 3 }; // wrong

* Naming rules for classes

  - Use descriptive but simple and short names. For stuff specific to LyX
    use LyX as prefix. Some modules, like mathed or spellchecker, could have
    other prefixes.
    [I am not so sure about the LyX prefix]

  - Class names are usually capitalized, and function names lowercased.
    Enums are named like Classes, enum values in CAPS.

  - Long variables are named like thisLongVariableName.

  New types are capitalized, so this goes for typedefs, classes, structs
  and enums.

* Formatting

  - Please adapt the formatting of your code to the setting in LyX in that
    particular file. Lars and Asger are slowly, but surely moving the source 
    towards Linux kernel style formatting, aka K&R style. We suggest that you 
    also do this, but this is NOT something that has been decided generally.


* Use existing structures

  - Use string wherever possible. LyX will someday move to Unicode, and
    that will be easy if everybody uses string now.

  - Check out the filename and path tools in filetools.h

  - Check out the string tools in lstring.h, and the SubString class
    and the regex class.

  - Use the DebugStream class to report errors and messages using
    the lyxerr instantiation.

  [add description of other existing structures]


* Declarations
  
  - Use this order for the access sections of your class: public,
    protected, private. The public section is interesting for every
    user of the class. The private section is only of interest for the
    implementors of the class (you). [Obviously not true since this is
    for developers, and we do not want one developer only to be able to
    read and understand the implementation of class internals. Lgb]
  
  - Avoid to declare global objects in the declaration file of the class. 
    If the same variable is used for all object, use a static member.

  - Avoid global or static variables. An exception to this rule is 
    very private stuff like the math stack.

  - Use the const keyword like this: char const * instead of const char *
    because this is more logical.


* Documentation

  - The documentation is generated from the header files.
  - You document for the other developers, not for yourself.
  - You should document what the function does, not the implementation.
  - in the .C files you document the implementation.
  - Single line description (///), multiple lines description (/** ... */)
  - You make the documentation by doing "make srcdoc" in the root,
    and then you'll find HTML in the srcdoc/ directory. Read with
    Netscape for best results.


* NAMING RULES FOR USER-COMMANDS
   
  Here's the set of rules to apply when a new command name is introduced:
 
  1) Use the object.event order. That is, use `word-forward' instead of 
     `forward-word'.
  2) Don't introduce an alias for an already named object. Same for events.
  3) Forward movement or focus is called `forward' (not `right').
  4) Backward movement or focus is called `backward' (not `left').
  5) Upward movement of focus is called `up'.
  6) Downward movement is called `down'.
  7) The begin of an object is called `begin' (not `start').
  8) The end of an object is called `end'.


* Using external GUI constructors (XForms fdesign)

  - Fdesign generated files should not be changed at all. The only changes
    needed are gettext, compability with 0.88 or when you have made your own
    xforms objects and have just a dummy in the .fd file in place of your
    own. In case you have to change the generated files for any of the
    reasons above, you should provide a patch against the clean generated
    file. Your callbacks must be in a separate file.

 *************************************************************

 How to create class interfaces.
 (a.k.a How Non-Member Functions Improve Encapsulation)
 ======================================================

	I recently read an article by Scott Meyers in C/C++ Users
Journal (Vol.18,No.2), where he makes a strong case on how non-member
functions makes classes more encapsulated, not less. Just to skipping
to the core of this provides us with the following algorithm for
deciding what kind of function to add to a class interface:

	- We need to add a function f to the class C's API.

	if (f needs to be virtual)
		make f a member function of C;
	else if (f is operator>> or operator<<) {
		make f a non-member function;
		if (f needs access to non-public members of C)
			make f a friend of C;
	} else if (f needs type conversions on its left-most argument) {
		make f a non-member function;
		if (f needs access to non-public members of C)
			make f a friend of C;
	} else if (f can be implemented via C's public interface)
		make f a non-member function;
	else
		make f a member function of C;
 
Unfortunately, to make the best use of this kind of Class API's we
need namespaces. As soon as Jean-Marc stops using gcc 2.8 and other
compilers seem more or less up to date on namespaces we will begin to
use them. _BUT_ we should begin to use the above algoritm ASAP. We
should also go through old code and apply this algorithm to the
existing member functions. That will help maintainability in the
future.

(I'll fill in more from Scott Meyers article when time allows.)

References
----------

[ExC++] Sutter, Herb. Exceptional C++: 47 engineering puzzles,
	programming problems, and solutions. ISBN 0-201-61562-2