ldgv - Label dependent session types

This repository contains an implementation of a frontend (parser and type checker), an interpreter and a C backend for LDGV.

LDGV stands for Label Dependency an Gay & Vasconcelos, the authors of multiple papers about Session Types. It mainly implements Label-Dependent Session Types. It additionally includes an implementation of the Cast Calculus from Label-Dependent Lambda Calculus and Gradual Typing.

The syntax is described in syntax.txt. The directory examples/ contains some examples, source files end in .ldgv.

Compilation is done via Stack, which you need to have installed. You may either compile and run LDGV by stack run ldgv (recommended). Or you install LDGV by stack install, essentially copying the executable ldgv to your $PATH.

Interpreter

Running the interpreter is possible via its command line interface

stack run ldgv -- interpret

resp. ldgv interpret.

C backend

The C backend is nearly feature complete, but missing a garbage collector and tests.

It is available only via the command line. See there for building and running.

Compiling the generated C code

The generated C code is dependent on the files in c-support/runtime/:

• LDST.h defines the data structures and interfaces to the scheduler and channel handler.
• LDST.c contains implementations of helper functions for which there is no need to generate them every time.
• LDST_serial.c contains a serial implementation of a scheduler and channel handler. Contexts in this implementation are not thread-safe.
• LDST_concurrent.c contains a concurrent implementation using a threadpool. Its size can be determined at compile time by defining LDST_THREADPOOL_SIZE, the default is four.
• thpool.h, thpool.c come from Pithikos/C-Thread-Pool and are used in the concurrent scheduler and channel handler implementation.

The generated code requires C11 support and the concurrent runtime requires support for the C11 atomics library and pthreads.

When compiling the generated code an optimization level of at least -O2 should be considered, with this the common compilers (clang, gcc) will transform the abundant number of tail calls into jumps which will keep the stack size from exploding.

The command line executable has support for invoking the C compiler on the generated source code by passing either -O or -L on the command line, see the output compile --help for more information.

Integrating and using the generated C code

Every top level definition in the source file will correspond to a symbol of type LDST_fp0_t (see LDST.h) with ldst_ prepended. Additional transformations are applied to support primes in function names: a q is replaced by qq and a prime is replaced by qQ.

The top-level functions should not be called directly but only through LDST_fork due to the interaction with the scheduler when using channel operations. There exist LDST_run and LDST_main to help with the execution of top level functions which don't exist as closures. See LDST.h for documentation of these.

Using the option -m IDENT / --main=IDENT with the command line C backend emit a int main(void) function at the end of the resulting C code which executes IDENT and prints the result. See the output of $LDGV compile --help for more information.

Multiple LDST files can be compiled separatly and then linked together. It is also possible to write functions in C and use them in LDST. The function has to properly curried, the top level symbol must have a type signature matching LDST_fp0_t and the name must match the name transformations pointed out above. (NB: this allows to subvert the type system, in all good and bad ways.)

Testing

You can run the full test suite by:

stack test

Web Page (Deprecated)

There was a web version availabe, using ghcjs and the reflex-platform. Because of some regression, presumably in some dependency of reflex, it is currently not available. Code and artifacts were factored out and moved to branch browser-support.