ILSpy/ICSharpCode.NRefactory.CSharp/Parser/mcs/compiler.doc

Compiler operation

The compiler has a number of phases:

* Parsing.

	Initially the compiler parses all the source files and keeps a
	parsed representation in memory.  Very syntax error checking
	is performed at this point.

	The compiler stores the information in classes whose names
	represent the language construct, for example, the "if"
	construct is stored in an `If' class.  A class is stored in a
	`Class'.

* The TypeManager

	The TypeManager loads all the assemblies that were referenced
	by the programmer.  The CLR type system is used as our
	repository for types defined as well.

	So the same interface that is used to query the types,
	properties and flags about system types is the same interface
	that we use for our types.

	As we work our way through the code generation and semantic
	analysis, new types are entered into the Type system through
	the use of System.Reflection.Emit.  The TypeManager will
	lookup types on both the user defined types and on the system
	defined ones.

	So special care has to be used.  The order in which we
	proceeed from here is important. 

* Base class resolution and type definition.

	Once the parsing has happened, the compiler resolves the
	inheritance tree for interfaces.   This is done recursively
	and we catch recursive interface definitions here.

	After this is done, we continue on with classes.  Classes have
	can have an optional "parent" inherit from or the implicit
	System.Object class (for normal builds, builds with /nostdlib
	will allow you to compile class System.Object with no parent).

	At this point we do some error checking and verify that the
	inherits/implements section of a class is correct (since we
	have previously built the interface inheritance).

	By the time we are done, all classes, structs and interfaces
	have been created using System.Reflection.Emit and registered
	with the Type Manager.

	This allows us to define fields and resolve argument names for
	methods, properties, indexers and events. 

* Field generation

	Fields are generated next, we go through all the type
	containers (classes and structs) and enter the fields into
	their types.

* Method, Properties, Indexers and events definitions

	Now all the methods, constructors, properties, indexers and
	events are entered.  They are only `defined' using
	System.Reflection.Emit.  No code generation will happen until
	everything has been entered into System.Reflection.Emit.

	This is important because to actually generate code we need to
	know everything about the environment in which the code is
	being generated.

* Code Generation

	At this point all the definitions have been entered into the
	type manager through System.Reflection.Emit.  We can now use
	System.Reflection to query all the information about the
	types. 

	Your normal semantic analysis and code generation phase lives
	here.

* Statements

	Most of the statements are handled in the codegen.cs file.

* Expressions

* Error reporting

	Always use `Report::Error' or `Report::Warning' methods of Report
	class. The actual Report instance is available via local context.
	An expression error reporting has to be done during Resolve phase,
	except when it's Emit specific (very rare).

	Error reporting should try to use the same codes that the
	Microsoft compiler uses (if only so we can track which errors
	we handle and which ones we dont).

	If there is an error which is specific to MSC, use negative
	numbers, and register the number in mcs/errors/errors.txt

	Try to write a test case for any error that you run into the
	code of the compiler if there is none already.

	Put your test case in a file called csNNNN.cs in the
	mcs/errors directory, and have the first two lines be:

	// csNNNN.cs: This is the error message
	// Line: XXX
	// Compiler options: an optional compiler options

	Where `XXX' is the line where the error ocurrs.  We will later
	use this as a regression test suite for catching errors in the
	compiler.