Analyze and Redesign vs.
Automated Conversion of MUMPS Code to C++

Analysis and Redesign

Short term costs are high

Long term / maintenance costs are low

Documented system

Stable design

Database design

Short term costs are moderate

MUMPS uses internal hierarchical database

C++ must have an interface to a database manager (SQL)

C++ code can be no better than the MUMPS code it was derived from

C++ does not execute comments

C++ does not execute data

Requires proprietary C++ library

May require manual intervention

Database design

MUMPS has only one data type (and they are all different)

MUMPS will create a new variable if one is misspelled (for a set statement)

C++ has strong data typing

MUMPS vendor specific functions

C++ implementation must provide MUMPS vendor functionality

C++ implementation will have to provide MUMPS functionality

Database manager must be selected

Side effects (some programmers rely on side effects)

MUMPS code has GOTOs

MUMPS code is sometimes self modifying

Trade-Off
	Analyze / Redesign	Automated Conversion
Short term costs	High	Moderate
Long term / maintenance	Low	Moderate
Documentation	Current design	`As was' design (extracted from code)
Design stability	High	Moderate
Data base design	Required	Required
Implementation of sparse arrays	No	Required
Implement context sensitive data typing	No	Required
Implement MUMPS functionality	No	Required
Parse MUMPS code	Required	Required
Run mflow analysis	Required	Required
Run naked analysis	Required	Required
Run metrics analysis	Required	Required
Transform script	No	Required
Transform filter	No	Required
Class library	No	Required
Header Files for Proprietary Classes	No	Required