A Tour Through the SubC C Compiler

I've always wanted to grok how a real compiler works. I've done compiler courses which were theoretical, and I've written my own toy compiler.

This is my own personal tour through the SubC C compiler, written by Nils M Holm. I've done this to help me grok a compiler that is small but which is complete enough to be able to recompile itself.

I've added a lot of comments to the original compiler code, subc-20161212.tgz, and I've written the descriptions of the stages below.

I haven't really covered much of the compiler theory behind the operation of SubC, so you really should buy and read Nils' book about SubC: Practical Compiler Construction. What I have done is to give an alternate viewpoint of the operation of SubC, and to cover the additions to the compiler written after Nils' book. Please don't read my descriptions instead of the book; Nils has put a lot of effort into SubC and he deserves some recompense.

The topics that I will cover are:

• The Basic Operation of a Compiler
• Lexical Analysis
• The C Pre-processor
• Parsing of Statements
• The Symbol Table
• Parsing of Declarations
• Abstract Syntax Trees
• Parsing of Expressions
• General Optimisation
• General Code Generation
• The x86-64 Code Generator
• Cyclic Register Allocation

Putting It Into Action

If you've cloned this GitHub repository to a Linux or BSD box, you should be able to do:

$ ./configure

to get the source ready to compile on your system. Then cd src into the source code tree and do:

$ make

to build the scc0 compiler binary. To prove that the compiler works, you can do this:

$ make scc

which is the compiler triple test: compile the compiler with another compiler (e.g. Gnu C), then compile the compiler with itself, then use this second compiler executable to compile itself again.

After that, you can do:

$ sudo make install

to install scc into /usr/local/bin.

Now go back and read through the descriptions and following along in the source code.

Copyrights

Unless otherwise noted,

• all source code and scripts are © Nils M Holm and have been placed in the public domain.
• all non-source code documents (e.g. English documents, image files) are © Warren Toomey under the Creative Commons BY-NC-SA 4.0 license.