A Foreign Language Interface from ML to Shell

This paper proposes a new approach to embedding shell scripts in an ML-style typed functional language. Instead of representing each construct in the shell script language as a typed component, the proposed approach treats an untyped shell script program as a typed function. The programmer writes a shell script and specifies its type according to the intention of his/her script. The ML compiler automatically generates glue codes that serialize and de-serialize input and output data. With this embedding mechanism, an ML programmer can enjoy the full benefits of shell programming directly within an ML program. The proposed approach has been implemented in our SML# compiler.