Workflow Engines: Draft

gistfile1.txt

- Software consists of algorithms and data structures. (Niklaus Wirth published
  a book called _Algorithms = Data Structures = Programs_.)
- Algorithms work on data structures by manipulating them in place or by
  transforming them into other data structures.
- Data structures are usually the starting and end point of a program.
  Exceptions are programs that make up their own data (e.g. by generating
  random data or by gathering measurements of sensors), but even then the
  definition of the data structures predates the algorithm.
- The data structure an algorithm works on shall be called an _input data
  structure_, and a data structure an algorithm is producing shall be called an
  _output data structure_. If the algorithm manipulates the input data
  structure in place, the output data structure is the same as the input data
  structure (with only different values), but input and output data structures
  remain two distinct concepts for that matter.
- Changes to an input or output data structure affect the algorithm. Changes to
  an algorithm can be made without changing either input or output data
  structure. Therefore, changing a data structure has a wider impact on the
  software than changing an algorithm, different runtime properties
  notwhithstanding. And therefore, design decisions affecting the data
  structures are more consequential than design decisions affecting the
  algorithms.
- Workflow engines make the algorithm transparent, but hide the data
  structures. The foremost design decisions made by their users affect the
  algorithms of the software, with data structures only being an afterthought.
  Even if the analysis and design starts with the data structure, the resulting
  program is foremost a visualization of the algorithm, not of the data
  structures.
- Any developer analyzing a program created using a workflow engine first and
  foremost gets acquainted with its algorithms, not with its data structures.
  The program is perceived as an algorithm, not as a combination of data
  structures and algorithms.
- In such an environment, the data structures become an implementation detail
  to be dealt with in the individual steps of a workflow (its sub-algorithms).
  However, changing such an alleged implementation detail has consequences for
  the entire program, which usually cannot be foreseen.
- Even though workflow engines are helpful when creating a first version of a
  program, the way in which such software are designed leads to fragile
  programs that hardly can be modified and improved.