A CIL Tutorial - Department of Computer Science - ETH Zürich

Chapter 0
Overview and Organization
Overall, this tutorial is organized as an OCaml program that uses CIL as a library. A Perl script
sits in front of the resulting OCaml program to make the result resemble various o-the-shelf C
compilers as much as possible. The code accompanying this tutorial provides nearly all of the
necessary boilerplate, and will let you dive right in and start analyzing and manipulating the AST.
0.1 Source le to AST
Day-to-day, it is generally not needed to know how this boilerplate code produces an AST that we
can work with in our OCaml program. It is good to understand how it works, though, in case your
project requires you to modify the compilation process in some way, for example to force linking
against a custom runtime library.
By default, CIL operates on one source le at a time (whole-program analysis is discussed by
Chapter 13). The result of parsing a preprocesssed source le is a data-structure of type Cil.file.
This structure contains a list of C denitions and declarations at global scope.
The details of parsing are handled in two places. Firstly in the Perl script mentioned above. It
is called cil/lib/Cilly.pm, and is part of the CIL installation. For a project that uses the CIL
library, like this one, we use our own Perl script, lib/Ciltut.pm, that derives from, overrides, and
extends Cilly.pm to specialize it to our purposes 1 . This Perl script also takes care of running the
preprocessor on the source le, interpreting some of the compiler ags, and performing the nal link.
Secondly, the OCaml program mentioned above reads in the preprocessed source le and passes it
to Frontc.parse with cabs, which parses the le, and turns it into a Cil.file.
The details of the Cil.file instance are covered in Chapter 1. For now it should suce to
mention that the boilerplate code in src/main.ml takes care of setting up CIL's internal datastructures
and options, parsing the input source le, and later on, emitting the .cil.c le that the
perl script passes to gcc, or whatever your favorite C compiler happens to be.
Everything that happens in between parsing the input and emitting the output happens in source
les like src/tut0.ml. Here, we write a function that takes a Cil.file, analyzes it, performs some
1 This is where we could provide additional command line ags to the back-end compiler or to force linking with
a custom runtime library, see the comments in that source le for details.
8