1 * MCS: The Ximian C# compiler
3 The Mono C# compiler is considered feature C# 1.0 complete at
4 this point and mature. MCS is able to compile itself and many
5 more C# programs (there is a test suite included that you can
6 use). It is routinely used to compile Mono, roughly 1.7
7 million lines of C# code.
9 The compiler is also fairly fast. On a IBM ThinkPad t40 it
10 compiles 18,000 lines of C# code per second.
12 Work on C# 2.0 has started: some pieces of it are available on
13 the standard compiler with the -2 switch (iterators, method
14 conversions) and some others are available on the `gmcs'
15 branch on CVS (generics)
19 The Mono C# compiler is part of the `mcs' module in the Mono CVS
20 you can get it from our <a href="anoncvs.html">Anonymous CVS</a> server,
21 or you can get nightly <a href="download.html">download page</a>.
25 MCS is written in C# and uses heavily the .NET APIs. MCS runs
26 on Linux with the Mono runtime and on Windows with both the
27 .NET runtime and the Mono runtime.
29 ** Reporting Bugs in MCS
31 When you report a bug, try to provide a small test case that would
32 show the error so we can include this as part of the Mono C# regression
35 If the bug is an error or a warning that we do not flag, write
36 a sample program called `csXXXX.cs' where XXXX is the code number
37 that is used by the Microsoft C# compiler that illustrates the
38 problem. That way we can also do regression tests on the invalid
41 ** Phases of the compiler
43 The compiler has a number of phases:
46 * Lexical analyzer: hand-coded lexical analyzer that
47 provides tokens to the parser.
49 * The Parser: the parser is implemented using Jay (A
50 Berkeley Yacc port to Java, that I ported to C#).
51 The parser does minimal work and syntax checking,
52 and only constructs a parsed tree.
54 Each language element gets its own class. The code
55 convention is to use an uppercase name for the
56 language element. So a C# class and its associated
57 information is kept in a "Class" class, a "struct"
58 in a "Struct" class and so on. Statements derive
59 from the "Statement" class, and Expressions from the
62 * Parent class resolution: before the actual code
63 generation, we need to resolve the parents and
64 interfaces for interface, classe and struct
67 * Semantic analysis: since C# can not resolve in a
68 top-down pass what identifiers actually mean, we
69 have to postpone this decision until the above steps
72 * Code generation: The code generation is done through
73 the System.Reflection.Emit API.
78 The compiler performs a number of simple optimizations on its input:
79 constant folding (this is required by the C# language spec) and
80 can perform dead code elimination.
82 Other more interesting optimizations like hoisting are not possible
83 at this point since the compiler output at this point does not
84 generate an intermediate representation that is suitable to
85 perform basic block computation.
87 Adding an intermediate layer to enable the basic block
88 computation to the compiler should be a simple task, but we
89 are considering having a generic CIL optimizer. Since all the
90 information that is required to perform basic block-based
91 optimizations is available at the CIL level, we might just skip
92 this step altogether and have just a generic IL optimizer that
93 would perform hoisting on arbitrary CIL programs, not only
94 those produced by MCS.
96 If this tool is further expanded to perform constant folding
97 (not needed for our C# compiler, as it is already in there)
98 and dead code elimination, other compiler authors might be
99 able to use this generic CIL optimizer in their projects
100 reducing their time to develop a production compiler.
104 See the <a href="bugs.html">bugs page</a> for more information.
106 A test suite is maintained to track the progress of
107 the compiler and various programs are routinely compiled and
112 Slides for the Mono C# Compiler presentation at .NET ONE are
114 href="http://primates.ximian.com/~miguel/slides-europe-nov-2002/Mono_C_Sharp_Overview_1007.sxi">here</a>
115 in StarOffice format.
119 MCS was able to parse itself on April 2001, MCS compiled itself
120 for the first time on December 28 2001. MCS became self hosting
121 on January 3rd, 2002.
123 The Mono Runtime and the Mono execution engine were able to make
124 our compiler self hosting on March 12, 2002.
126 ** Questions and Answers
128 Q: Does the Mono C# compiler support C# 2.0?
130 A: At this point the Mono C# compiler supports some of the features of
131 C# 2.0, but the support has not been completed. To enable 2.0 features
132 you must use the -2 flag to the compiler.
134 Q: What features are available as of Feb 2004?
136 A: Iterators have been implemented as well as method group implicit
137 conversion to delegates on the main compiler branch.
139 We have a branch of the compiler in the module `mcs/gmcs' which is
140 where we are developing the Generics support for the compiler. Plenty
141 of tests work (see mcs/tests/gen-*.cs for a list of tests), but work
144 Q: Will the C# 2.0 features be part of the Mono 1.0 release?
146 A: Only a few, the generic compiler will not be part of the 1.0
147 stable release, but a beta preview will be distributed.
149 Q: Why not write a C# front-end for GCC?
151 A: I wanted to learn about C#, and this was an exercise in this
152 task. The resulting compiler is highly object-oriented, which has
153 lead to a very nice, easy to follow and simple implementation of
156 I found that the design of this compiler is very similar to
157 Guavac's implementation.
159 Targeting the CIL/MSIL byte codes would require to re-architecting
160 GCC, as GCC is mostly designed to be used for register machines.
162 The GCC Java engine that generates Java byte codes cheats: it does
163 not use the GCC backend; it has a special backend just for Java, so
164 you can not really generate Java bytecodes from the other languages
167 Q: If your C# compiler is written in C#, how do you plan on getting
168 this working on a non-Microsoft environment.
170 We will do this through an implementation of the CLI Virtual
171 Execution System for Unix (our JIT engine).
173 Our JIT engine is working for the purposes of using the compiler.
174 The supporting class libraries are being worked on to fully support
179 A: No, currently I am using Jay which is a port of Berkeley Yacc to
180 Java that I later ported to C#. This means that error recovery is
181 not as nice as I would like to, and for some reason error
182 productions are not being caught.
184 In the future I want to port one of the Bison/Java ports to C# for
187 Q: Should someone work on a GCC front-end to C#?
189 A: I would love if someone does, and we would love to help anyone that
190 takes on that task, but we do not have the time or expertise to
191 build a C# compiler with the GCC engine. I find it a lot more fun
192 personally to work on C# on a C# compiler, which has an intrinsic
195 We can provide help and assistance to anyone who would like to work
198 Q: Should someone make a GCC backend that will generate CIL images?
200 A: I would love to see a backend to GCC that generates CIL images. It
201 would provide a ton of free compilers that would generate CIL
202 code. This is something that people would want to look into
203 anyways for Windows interoperation in the future.
205 Again, we would love to provide help and assistance to anyone
206 interested in working in such a project.
208 Q: What about making a front-end to GCC that takes CIL images and
209 generates native code?
211 A: I would love to see this, specially since GCC supports this same
212 feature for Java Byte Codes. You could use the metadata library
213 from Mono to read the byte codes (ie, this would be your
214 "front-end") and generate the trees that get passed to the
217 Ideally our implementation of the CLI will be available as a shared
218 library that could be linked with your application as its runtime
221 Again, we would love to provide help and assistance to anyone
222 interested in working in such a project.
224 Q: But would this work around the GPL in the GCC compiler and allow
225 people to work on non-free front-ends?
227 A: People can already do this by targeting the JVM byte codes (there
228 are about 130 compilers for various languages that target the JVM).
230 Q: Why are you writing a JIT engine instead of a front-end to GCC?
232 A: The JIT engine and runtime engine will be able to execute CIL
233 executables generated on Windows.
235 You might also want to look at the <a href="faq.html#gcc">GCC</a>
236 section on the main FAQ