The Micro Focus COBOL compiler *is* written in Micro Focus COBOL, even to this day. With various code generators, it can emit proprietary executable intermediate code, native code, .NET IL, and Java byte code, and is used to compile itself. That allows the compiler to be built and to run on a variety of processors — one just needs a code generator for the target processor.

For comparison, RM/COBOL is written in a proprietary assembly-language-like interpreted language called POPS (short for “Programmed Operators”) that was developed by Ryan-McFarland’s immediate predecessor, Digitek, in the 1960s. The operations are designed for the kinds of things compilers need to do (parsing and managing tables), are very compact, and implementing the compiler on a new platform involved just rewriting (or more properly, translating) the well-structured POPS interpreter using the target processor’s instruction set, and similarly translating the runtime system. This made the compilers very portable.

POPS was used in RM compilers for various languages (COBOL, BASIC, FORTRAN). The compilers and the matching runtime systems ran on computers with as little as 32KB (that’s kilobytes, not MB or GB) of RAM. Radio Shack/Tandy sold RM/COBOL and RM/BASIC systems for the TRS-80 back in the late 70s, along with accounting applications developed with those compilers. In the 1980s, IBM Professional Fortran (also sold directly as RM/FORTRAN) for MS-DOS was also developed by RM and used the POPS technology in the compiler with native code generation.

RM/COBOL never emitted native code, but instead emits a variant of POPS that’s highly optimized for what COBOL programs need to do: file I/O, decimal arithmetic, etc. RM experimented with generating native code but for typical COBOL I/O bound COBOL programs that spend much of their time performing exact-precision decimal arithmetic and moving data around within memory, it just wasn’t worth the added complexity.