The machine code compiler is invoked with
python -m fold.machinecode
in the root of the repository. It compiles Fold programs to LLVM IR, and if invoked with the --jit-exec or -e option, it passes the IR to a JIT compiler targeting native machine code and immediately executes the program.
The machine code compiler ignores all architectural hints aimed at compilation to logic, but interprets a few hints of its own.
The fork keyword is implemented with system-level process forks, which in some cases can lead to excessive number of processes and exhaustion of resource limits. In some use cases one can write
` mcode_fork take_jump.
fork target;
to signal to the machine code compiler it should ignore the thread of execution below the fork statement. Effectively the hint turns the fork into a goto as far as the machine code compiler is concerned. The hint spares the system-level process fork if the user knows the behavior of the program is unaffected.
Similarly, one writes
` mcode_fork ignore_jump.
fork target;
to simply ignore the fork statement by the machine code compiler, and proceed with execution of the statements below it.