WARNING: this project is no longer being developed; the author no longer believes that a language without manual memory management can be made sufficiently fast.

Optimising compiler for Awelon Bytecode, see Awelon Project.

To build, run ./build. As abcc depends only on the C standard library, you shouldn't need to do anything else. If you are debugging abcc, look in the build script for options.

If you wish to use the show-graph debugging tool you will need to download the GraphStream libraries and place them in show-graph/lib. Then run ./build from show-graph. The libraries you will need are gs-core-1.2.jar and gs-ui-1.2.jar.

Run abcc --help for usage information. abcc only compiles Awelon Bytecode (ABC); if you want to compile Awelon Object code you must compile it to ABC first. Use something like ao abc <your function here> | abcc. Replace ao abc with ao abc.ann if you want better error reporting.

Inputs to abcc may be either from stdin or a file. Executables are written to a.out and text based output is passed to stdout.

The type inferencer prints the input and output types on separate lines of stdout. -> represents a polymorphic block, and => a monomorphic one. Void is shown as a free type variable, and currently substructural attributes are not inferred.

The graphviz output is designed to help visualise the generated program, and may be useful if you want to see how well the optimiser is doing. Pipe it into dot for easy viewing: abcc -G | dot -Tx11. dot may either take a while or segfault when producing large graphs.

Programs are assumed to have type (N*1)*(1*1) -> (N*1)*(1*1). Statically linked amd64 linux executables are generated, which require SSE4.1 (for implementing Q efficiently) and do not depend on the standard C library. At present memory is neither freed or reused, and text constants are unimplemented.

These are mostly for use in development, but some may be useful to others.

• build: Builds abcc.

• compare-simplified: Simple tool for visualising what impact the ao simplifier has.

• constantify: Converts a file's contents into a string constant stored in a C file for embedding into the abcc binary. This allows the abcc binary to be self-contained.

• expect: Auto generates part of the type inferencer for a small but noticable performance improvement.

• memusg: Uses valgrind to print the peak memory usage of a process.

• regen-all: Creates a file containing every word in the dictionary (minus the blacklist) as a block literal which is immediately dropped. Useful to check that every thing typechecks in one hit. Relies on the cache built by regen-words.

• regen-words: Produce a file in words/w-<word name> containing the annotated ABC of each word in the dictionary. This allows us to avoid paying startup costs for ao. There is a small blacklist of words which are not included; these are the transitive closure of words which are unfinished (marked as TODO).

• runabc, runtests: Remnants of the old prototype compiler. Due to be purged once everything useful has been extracted from them.