path: root/mcs/CodingStyle

* Coding Style for the Mono C# source code.

* Class Libraries and Assembly Layout

	The class libraries are grouped together in the assemblies
	they belong.
	
	Each directory here represents an assembly, and inside each
	directory we divide the code based on the namespace they
	implement.
	
	In addition, each assembly directory contains a Test directory
	that holds the NUnit tests for that assembly.
	
	We use a new build system which is described by various README
	files in mcs/build
	
	The build process typically builds an assembly, but in some
	cases it also builds special versions of the assemblies
	intended to be used for testing.

* Missing implementation bits

	If you implement a class and you are missing implementation bits,
	please use the attribute [MonoTODO].  This attribute can be used
	to programatically generate our status web pages:

	[MonoTODO("My Function is not available on Mono")]
	int MyFunction ()
	{
		throw new NotImplementedException ();
	}

	Ideally, write a human description of the reason why there is
	a MonoTODO, this will be useful in the future for our
	automated tools that can assist in developers porting their code.

* Supporting .NET 1.2, .NET 1.1 and .NET 1.0 builds

	The defines NET_1_1 and NET_2_0 are used to include
	features.   When NET_2_0 is defined, it also implies that the
	NET_1_1 is defined.

	To have code which is only available in an old version, use ONLY_1_0,
	ONLY_1_1

* Tagging buggy code

	If there is a bug in your implementation tag the problem by using
	the word "FIXME" in the code, together with a description of the 
	problem.

	Do not use XXX or obscure descriptions, because otherwise people
	will not be able to understand what you mean.

* Tagging Problematic specs.

	If the documentation and the Microsoft implementation do
	differ (you wrote a test case to prove this), I suggest that you edit
	the file `mcs/class/doc/API-notes' so we can keep track of these problems
	and submit our comments to ECMA or Microsoft and seek clarification.

	Sometimes the documentation might be buggy, and sometimes the implementation
	might be buggy.  Lets try to identify and pinpoint which one
	is the correct one.

	Sometimes the specification will be lame (consider Version.ToString (fieldCount)
	where there is no way of knowing how many fields are available, making the API
	not only stupid, but leading to unreliable code).

	In those cases, use the keyword "LAMESPEC".
	

* Coding considerations and style.

	In order to keep the code consistent, please use the following
	conventions.  From here on `good' and `bad' are used to attribute
	things that would make the coding style match, or not match.  It is not
	a judgement call on your coding abilities, but more of a style and 
	look call.  Please try to follow these guidelines to ensure prettiness.

	Use 8 space tabs for writing your code (hopefully we can keep
	this consistent).  If you are modifying someone else's code, try
	to keep the coding style similar.

	Since we are using 8-space tabs, you might want to consider the Linus
	Torvals trick to reduce code nesting.  Many times in a loop, you will
	find yourself doing a test, and if the test is true, you will nest.
	Many times this can be changed.  Example:


		for (i = 0; i < 10; i++) {
			if (something (i)) {
				do_more ();
			}
		}

	This take precious space, instead write it like this:

		for (i = 0; i < 10; i++) {
			if (!something (i))
				continue;
			do_more ();
		}

* Performance and readability

	It is more important to be correct than to be fast.

	It is more important to be maintainable than to be fast.

	Fast code that is difficult to maintain is likely going to
	be looked down upon.

* Style Guidelines

		* Use a space before an opening parenthesis when calling
		  functions, or indexing, like this:

			method (a);
			b [10];

		* Do not put a space after the opening parenthesis and the 
		  closing one, ie:

			good: method (a);	array [10];

			bad:  method ( a );	array[ 10 ];

		* Inside a code block, put the opening brace on the same line
	  	  as the statement:

			good:
				if (a) {
					code ();
					code ();
				}

			bad:
				if (a) 
				{
					code ();
					code ();
				}

		* Avoid using unecessary open/close braces, vertical space
		  is usually limited:

			good:
				if (a)
					code ();

			bad:
				if (a) {
					code ();
				}

		* When defining a method, use the C style for brace placement, 
		  that means, use a new line for the brace, like this:

			good:
				void Method ()
				{
				}

			bad:
				void Method () {
				}

		* Properties and indexers are an exception, keep the
		  brace on the same line as the property declaration.
		  Rationale: this makes it visually
		  simple to distinguish them.

			good:
				int Property {
					get {
						return value;
					}
				}

			bad:
				int Property 
				{
					get {
						return value;
					}
				}

		  Notice how the accessor "get" also keeps its brace on the same
		  line.

		  For very small properties, you can compress things:

			ok:
				int Property {
					get { return value; }
					set { x = value; }
				}

		* Use white space in expressions liberally, except in the presence
		  of parenthesis:

			good:

				if (a + 5 > method (blah () + 4))

			bad:
				if (a+5>method(blah()+4))

		* For any new files, please use a descriptive introduction, like
		  this:

			//
			// System.Comment.cs: Handles comments in System files.
			//
			// Author:
			//   Juan Perez (juan@address.com)
			//
			// (C) 2002 Address, Inc (http://www.address.com)
			//

		* If you are modyfing someone else's code, and your contribution
		  is significant, please add yourself to the Authors list.

		* Switch statements have the case at the same indentation as the
		  switch:

			switch (x) {
			case 'a':
				...
			case 'b':
				...
			}

		* Argument names should use the camel casing for
		  identifiers, like this:

		 	good:
				void Method (string myArgument)

			bad:
				void Method (string lpstrArgument)
				void Method (string my_string)

		* Empty methods: They should have the body of code using two 	
		  lines, in consistency with the rest:

			good:
				void EmptyMethod ()
				{
				}

			bad:
				void EmptyMethod () {}

				void EmptyMethod () 
				{}
		
		* Line length: The line length for C# source code is 134 columns.


		  If your function declaration arguments go beyond
		  this point, please align your arguments to match the
		  opening brace, like this:

	 		void Function (int arg, string argb,
	 			       int argc)
	 		{
	 		}
	 
	  	  When invoking functions, the rule is different, the
		  arguments are not aligned with the previous
		  argument, instead they begin at the tabbed position,
		  like this:
	  
			void M ()
			{
				MethodCall ("Very long string that will force",
					"Next argument on the 8-tab pos",
					"Just like this one")
		
			}

		* Variable declaration indentation.

		  Sometimes it is convenient to indent the variables to make the code
	 	  look pretier, but do not add gratuitous space, try to use the minimally
	  	  necessary space, for example:

		  Good:

		  	void Method ()
	 	  	{
				string b;
		  		int    a;
				byte   c;
			}

		  Bad:

		  	void Method ()
	 	  	{
				string 		b;
		  		int    		a;
				byte   		c;
			}

		* Braces and the `else' clause

		  If there are braces closing or opening next to the else clause,
	  	  they go on the same line as the word `else', for example:

		  Good:

			if (..) {

			} else {
		
			}
	
		  Bad:

			if (..) {

			} 
			else {
		
			}

		  Bad:

			if (..) {

			} else 
			{		
			}

		  Bad:

			if (..) {

			} 
			else 
			{
		
			}

* RCS and CVS tags

	Some users like to use the special RCS/CVS tags in their
	source code: $id$, $log$ and so on.  

	The use of these is not permitted on the Mono source code
	repository.   This metadata belongs on a ChangeLog or in the
	SVN metadata. 

* File formats

	Historically our repository has used a mix of line-endings,
	this is a mistake that we are trying hard to fix.

	For existing files, please make sure that you do not convert 
	the file, as that causes us to loose precious history (the
	full file is commited).

	For new files that you create, please make sure that you use
	Subversion's support for mapping the line endings
	automatically, after adding your file:

		$ svn add file.cs

	Execute this command:

		$ svn propset svn:eol-style native file.cs

	Which will make the file automatically receive the proper
	treatment from that point on.

	Please verify before commiting that your changes wont loose
	history, you can do this by running:

		$ svn diff

	And examining the output.

* ChangeLogs

	ChangeLogs are the files that we use to track the project
	history.  ChangeLogs are found one per directory, or in small
	projects, one per project.

	The format looks like this:

	2004-11-19  Raja R Harinath  <rharinath@novell.com>

		* Makefile (%-profiles): Go through an intermediate
		set of rules.  Move body to ...
		(profiles-do--%): ... this.
		(profiles-do--run-test): Customized rule that usefully
		runs with 'make -j' and 'make -k'.
		(profiles-do--all, profile-do--%--all): Orchestrate
		the bootstrap process.

		* file.cs (MainForm): Updated version.

	The date, author, email address in the first line.

	From that point on a list of changes in a file-by-file basis,
	describing what changes were done.

	This information must be cut and pasted into your commit
	message, so the information ends up in two places: in the
	subversion repository metadata and also on the source code
	distirbution (which does not have the Subversion metadata).
		
* Warnings

	Avoid commiting code with warnings to the repository, the use
	of #pragmas to disable warnings is strongly discouraged, but
	can be used on unique cases.  Please justify the use of the
	warning ignore clause on a comment.

	Do not commit changes to the Makefiles that removes warnings,
	if anything warnings should be eliminated one at a time, and
	if not possible, they must be flagged.


* Examples:

class X : Y {

	bool Method (int argument_1, int argument_2)
	{
		if (argument_1 == argument_2)
			throw new Exception (Locale.GetText ("They are equal!");

		if (argument_1 < argument_2) {
			if (argument_1 * 3 > 4)
				return true;
			else
				return false;
		}

		//
		// This sample helps keep your sanity while using 8-spaces for tabs
		// 
		VeryLongIdentifierWhichTakesManyArguments (
			Argument1, Argument2, Argument3,
			NestedCallHere (
				MoreNested));
	}

	bool MyProperty {
		get {
			return x;
		}

		set {
			x = value;
		}
	}

	void AnotherMethod () 
	{
		if ((a + 5) != 4) {
		}

		while (blah) {
			if (a)
				continue;
			b++;
		}
	}
}

* Conditional compilation

	Ideally we would not need conditional compilation, and the use
	of #ifdef is strongly discouraged.  But due to our support for
	old C# 1.0 compilers we have to use it in a few places.

	Try to avoid negative tests that have an else clause, for
	example:

	    #if !NET_2_0
	    	CODE_FOR_1_0
            #else
		CODE_FOR_2_0
	    #endif

        Instead use:

	    #if NET_2_0
	    	CODE_FOR_2_0
	    #else
	        CODE_FOR_1_0
            #endif

	When a major feature differs across compilation targets, try
	to factor out the code into a separate class, a helper class
	or a separate file, and include that in your profile while
	surrounding that helper file/class with the ifdefs to reduce
	the amount of ifdefs in the code.

	For instance, this is used for some parts of Grasshopper where
	the code is ifdefed out, when large parts of a file would have
	been ifdefed out, we moved the code into a MyOtherFile.jvm.cs

	For 2.0 classes, this is even simpler as code can be trivially
	factored out into 

		 MyHelperClass.cli.cs
		 MyHelperClass.jvm.cs

	By using partial classes.