Skip to content

README.md

Latest commit

 

History

History
29 lines (15 loc) · 2.22 KB

README.md

File metadata and controls

29 lines (15 loc) · 2.22 KB

Onramp Compiler

cci is the Onramp preprocessed C compiler. It takes preprocessed C as input and outputs Onramp assembly.

This is a description of the implementation stages of the compiler. The input to the compiler is (a subset of) the C programming language after preprocessing, the subset of which depends on the stage. For a specification of the output of the compiler, see Onramp assembly.

(It's called "cci" because it compiles .i files. The analogous program in other compilers might be called cc1, except a cc1 binary typically also handles preprocessing, which in Onramp is the separate program cpp.)

The compiler is implemented in three stages:

  • 0-omc is written in compound assembly and compiles Onramp Minimal C (omC). It is a single-pass compiler designed to be as simple as possible with no optimizations whatsoever.

  • 1-opc is written in omC and compiles Onramp Practical C (opC). This aims to support a large subset of modern C sufficient to comfortably implement the final stage. It is also a single-pass compiler with only trivial micro-optimizations where it is convenient.

  • 2-full is written in opC and aims to implement most of the C2x language plus several GNU extensions. This adds function pointers, long long, floating point and more. It should be sufficient to compile any modern C software.

Input Format

The input to cci is the output of the Onramp preprocessor cpp. The preprocessor performs the first five phases of translation. This means:

  • Preprocessing is already done. We do not need to handle any preprocessor directives (except for debug info, see below) or include any other files.
  • Comments are stripped. No stage of cci handles comments at all.
  • Line endings are normalized. There should be no carriage returns in the input.

These restrictions greatly simplify our lexer. Unlike most other Onramp programs, cci is not designed to accept hand-written input files.

The only exception for preprocessor directives is #line and #pragma for debug info. These are handled specially by the lexer in each stage.