1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
|
* MCS: The Ximian C# compiler
MCS began as an experiment to learn the features of C# by
writing a large C# program. MCS is currently able to parse C#
programs and create an internal tree representation of the
program. MCS can parse itself.
MCS now does type checking at the class, interface and struct
levels and can resolve the class hierarchy and as of last week
can generate interface code.
Work is progressing quickly on various fronts in the C#
compiler. Recently I started using the System.Reflection API
to load system type definitions and avoid self-population of
types in the compiler and dropped my internal Type
representation in favor of using the CLI's System.Type.
** Phases of the compiler
The compiler has a number of phases:
<ul>
* Lexical analyzer: hand-coded lexical analyzer that
provides tokens to the parser.
* The Parser: the parser is implemented using Jay (A
Berkeley Yacc port to Java, that I ported to C#).
The parser does minimal work and syntax checking,
and only constructs a parsed tree.
Each language element gets its own class. The code
convention is to use an uppercase name for the
language element. So a C# class and its associated
information is kept in a "Class" class, a "struct"
in a "Struct" class and so on. Statements derive
from the "Statement" class, and Expressions from the
Expr class.
* Parent class resolution: before the actual code
generation, we need to resolve the parents and
interfaces for interface, classe and struct
definitions.
* Semantic analysis: since C# can not resolve in a
top-down pass what identifiers actually mean, we
have to postpone this decision until the above steps
are finished.
* Code generation: The compiler recently started generating IL
executables that contain interfaces. Work is
progressing in other areas.
The code generation is done through the System.Reflection.Emit API.
</ul>
<a name="tasks">
** Current pending tasks
Simple tasks:
<ul>
* Array declarations are currently being ignored,
* PInvoke declarations are not supported.
* Pre-processing is not supported.
* Attribute declarations and passing currently ignored.
* Compiler does not pass around line/col information from tokenizer for error reporting.
* Jay does not work correctly with `error'
productions, making parser errors hard to point. It
would be best to port the Bison-To-Java compiler to
become Bison-to-C# compiler.
Nick Drochak has started a project on SourceForge for this.
You can find the project at: <a href="http://sourceforge.net/projects/jb2csharp/">
http://sourceforge.net/projects/jb2csharp/</a>
</ul>
Interesting and Fun hacks to the compiler:
<ul>
* Finishing the JB port from Java to C#. If you are
interested in working on this, please contact the project admin on SourceForge:
<a href="http://sourceforge.net/projects/jb2csharp/">
http://sourceforge.net/projects/jb2csharp/</a>
More on JB at: <a href="http://www.cs.colorado.edu/~dennis/software/jb.html">
http://www.cs.colorado.edu/~dennis/software/jb.html</a>
JB will allow us to move from the Berkeley Yacc
based Jay to a Bison-based compiler (better error
reporting and recovery).
* Semantic Analysis: Return path coverage and
initialization before use coverage are two great
features of C# that help reduce the number of bugs
in applications. It is one interesting hack.
* Enum resolutions: it is another fun hack, as enums can be defined
in terms of themselves (<tt>enum X { a = b + 1, b = 5 }</tt>).
</ul>
** Questions and Answers
Q: Why not write a C# front-end for GCC?
A: I wanted to learn about C#, and this was an exercise in this
task. The resulting compiler is highly object-oriented, which has
lead to a very nice, easy to follow and simple implementation of
the compiler.
I found that the design of this compiler is very similar to
Guavac's implementation.
Targeting the CIL/MSIL byte codes would require to re-architecting
GCC, as GCC is mostly designed to be used for register machines.
The GCC Java engine that generates Java byte codes cheats: it does
not use the GCC backend; it has a special backend just for Java, so
you can not really generate Java bytecodes from the other languages
supported by GCC.
Q: If your C# compiler is written in C#, how do you plan on getting
this working on a non-Microsoft environment.
We will do this through an implementation of the CLI Virtual
Execution System for Unix (our JIT engine).
Q: Do you use Bison?
A: No, currently I am using Jay which is a port of Berkeley Yacc to
Java that I later ported to C#. This means that error recovery is
not as nice as I would like to, and for some reason error
productions are not being caught.
In the future I want to port one of the Bison/Java ports to C# for
the parser.
Q: Should someone work on a GCC front-end to C#?
A: I would love if someone does, and we would love to help anyone that
takes on that task, but we do not have the time or expertise to
build a C# compiler with the GCC engine. I find it a lot more fun
personally to work on C# on a C# compiler, which has an intrinsic
beauty.
We can provide help and assistance to anyone who would like to work
on this task.
Q: Should someone make a GCC backend that will generate CIL images?
A: I would love to see a backend to GCC that generates CIL images. It
would provide a ton of free compilers that would generate CIL
code. This is something that people would want to look into
anyways for Windows interoperation in the future.
Again, we would love to provide help and assistance to anyone
interested in working in such a project.
Q: What about making a front-end to GCC that takes CIL images and
generates native code?
A: I would love to see this, specially since GCC supports this same
feature for Java Byte Codes. You could use the metadata library
from Mono to read the byte codes (ie, this would be your
"front-end") and generate the trees that get passed to the
optimizer.
Ideally our implementation of the CLI will be available as a shared
library that could be linked with your application as its runtime
support.
Again, we would love to provide help and assistance to anyone
interested in working in such a project.
Q: But would this work around the GPL in the GCC compiler and allow
people to work on non-free front-ends?
A: People can already do this by targeting the JVM byte codes (there
are about 130 compilers for various languages that target the JVM).
Q: Why are you writing a JIT engine instead of a front-end to GCC?
A: The JIT engine and runtime engine will be able to execute CIL
executables generated on Windows.
You might also want to look at the <a href="faq.html#gcc">GCC</a>
section on the main FAQ
|