Abstract
Most of the number-crunching software running on supercomputers lack a concept for an error-free input processing. The programs often terminate after several hours due to user input errors. We present an analysis of this input problem. First, we separate the input parsing system from the number-crunching code. Second, we define a general grammar suitable for an automatic generation of a parsing system using the LEX/YACC tools. This concept leads to an input system that is easy to use and the generated input data for the number-crunching software is as error free as possible. We discuss the implementation of such an input system for an ab initio quantum chemical package.