An experimental COBOL compiler, written in Rust.

Overview

🚧 This project is experimental and not ready for production use. Read the disclaimer for more details. 🚧

Cobalt is an experimental COBOL compiler targeting x86 and aarch64 Linux systems, providing blazing fast compile times and runtime performance. Some key features include:

• Built on Rust and Cranelift for fast modern code generation.
• Small, efficient output binaries.
• Security by default, utilising hardware features such as PAC+BTI/Intel CET where available.
• Supports alternative linkers such as gold and mold for link-time performance.

For a peek inside the internals of Cobalt, see this project's documentation. A basic usage outline and performance comparison can be found below.

Building

This project depends on the presence of a Rust toolchain, with an MSRV of 1.76. The build process may also install the rust-src toolchain component, if not already present.

To build the compiler in release mode, simply run the following:

git clone [email protected]:c272/cobalt.git
cd cobalt && ./build.sh release

Build artifacts for the project can be found in the ./build directory. For testing debug builds during development of the compiler, the "run" and "clean" utilities may also be useful, for example:

# Build and execute an example with all opts.
./run.sh build examples/algorithms/primality.cobol -O speed_and_size
./out/primality

# Clean up any compiler build artifacts.
./clean.sh

Usage

Builds a single COBOL file.

Usage: cobalt build [OPTIONS] <FILE>

Arguments:
  <FILE>  The file to build

Options:
  -o, --output-name <OUTPUT_NAME>  The name of the output executable. By default, the primary input file name with extensions removed
  -d, --output-dir <DIR>           The output directory to save to. By default, `./out`
  -v, --verbose...                 Set the verbosity of compiler output. Can be specified multiple times
  -O, --opt-level <OPT_LEVEL>      The optimisation level to compile the provided code at. By default, optimises for compile speed (no optimisations) [possible values: none, speed, speed_and_size]
  -p, --prefer-platform-linker     Actively selects the platform linker over other available linkers
      --disable-security-features  Disables the generation of instructions utilising hardware security features within output binaries (e.g. PAC/BTI on aarch64)
  -h, --help                       Print help

Benchmarks

A basic set of benchmarks are available in this repository for testing performance against a baseline solution (GnuCobol). These can be run through the benchmarking crate, which requires the following additional dependencies:

• When comparing against GnuCobol, cobc in PATH.
• When comparing against GnuCobol+clang, clang in PATH.

Results from benchmarking runs can be found, by default, within the ./bench_out directory as JSON files. For documentation on the structure of these output files, as well as how benchmark TOML files are formatted, view the bench crate documentation.

Some example uses of the benchmarking suite can be seen below:

# Display a manual page for the benchmarking tool.
./benchmark.sh --help

# Execute all benchmarks with no comparisons, all opts.
./benchmark.sh -O speed_and_size

# Execute a specific benchmark TOML file.
./benchmark.sh --ignore-default --benchmarks ./benchmarks/foo.toml

# Compare all against `cobc` performance, all opts.
./benchmark.sh --run-cobc -O speed_and_size -C 3

# Compare against `cobc`+`clang`, build performance only.
./benchmark.sh --run-cobc --run-clang --build-only

Disclaimer

This compiler is experimental and feature incomplete, only supporting a minimal set of the COBOL specification. In addition, the only format supported for parity is GnuCobol's -free format.

In accordance, this compiler is not intended for production use and should not be used as such. No support is provided for production use cases.