Control table
If there are several hierarchical levels of control table they may behave in a manner equivalent to UML state machines[3] Control tables often have the equivalent of conditional expressions or function references embedded in them, usually implied by their relative column position in the association list.
It is sometimes generated by an optimizing compiler to execute a switch statement – provided that the input range is small and dense, with few gaps (as created by the previous array example) [1].
A control table can be constructed along similar lines to a language dependent switch statement but with the added possibility of testing for combinations of input values (using boolean style AND/OR conditions) and potentially calling multiple subroutines (instead of just a single set of values and 'branch to' program labels).
(The switch statement construct in any case may not be available, or has confusingly differing implementations in high level languages (HLL).
A multi-dimensional control table will normally, as a minimum, contain value/action pairs and may additionally contain operators and type information such as, the location, size and format of input or output data, whether data conversion (or other run-time processing nuances) is required before or after processing (if not already implicit in the function itself).
The table may or may not contain indexes or relative or absolute pointers to generic or customized primitives or subroutines to be executed depending upon other values in the "row".
conditions and actions implied by structure (This side-by-side pairing of value and action has similarities to constructs in Event-driven programming, namely 'event-detection' and 'event-handling' but without (necessarily) the asynchronous nature of the event itself) The variety of values that can be encoded within a control table is largely dependent upon the computer language used.
Assembly language provides the widest scope for data types including (for the actions), the option of directly executable machine code.
Typically a control table will contain values for each possible matching class of input together with a corresponding pointer to an action subroutine.
Comments positioned above each column (or even embedded textual documentation) can render a decision table 'human readable' even after 'condensing down' (encoding) to its essentials (and still broadly in-line with the original program specification – especially if a printed decision table, enumerating each unique action, is created before coding begins).
The interpreter does not need to be unduly complex, or produced by a programmer with the advanced knowledge of a compiler writer, and can be written just as any other application program – except that it is usually designed with efficiency in mind.
The program instructions are, in theory, infinitely extensible and constitute (possibly arbitrary) values within the table that are meaningful only to the interpreter.
For particular platforms/language, they can be specifically designed to minimize instruction path lengths using branch table values or even, in some cases such as in JIT compilers, consist of directly executable machine code "snippets" (or pointers to them).
By separating (or 'encapsulating') the executable coding from the logic, as expressed in the table, it can be more readily targeted to perform its function most efficiently.
The examples below have been chosen partly to illustrate potential performance gains that may not only compensate significantly for the additional tier of abstraction, but also improve upon – what otherwise might have been – less efficient, less maintainable and lengthier code.
The following examples are arbitrary (and based upon just a single input for simplicity), however the intention is merely to demonstrate how control flow can be effected via the use of tables instead of regular program statements.
To handle all 256 different input values, approximately 265 lines of source code would be required (mainly single line table entries) whereas multiple 'compare and branch' would have normally required around 512 source lines (the size of the binary is also approximately halved, each table entry requiring only 4 bytes instead of approximately 8 bytes for a series of 'compare immediate'/branch instructions (For larger input variables, the saving is even greater).
If pointer arrays are not supported, a SWITCH statement or equivalent can be used to alter the control flow to one of a sequence of program labels (e.g.: case0, case1, case2, case3, case4) which then either process the input directly, or else perform a call (with return) to the appropriate subroutine (default, Add, Subtract, Multiply or Divide,..) to deal with it.
CT4 (a complete 'program' to read input1 and process, repeating until 'E' encountered) In the specialist field of telecommunications rating (concerned with the determining the cost of a particular call), table-driven rating techniques illustrate the use of control tables in applications where the rules may change frequently because of market forces.
The table itself can be essentially a collection of "raw data" values that do not even need to be compiled and could be read in from an external source (except in specific, platform dependent, implementations using memory pointers directly for greater efficiency).
Programming languages that are able to use pointers have the dual advantage that less overhead is involved, both in accessing the contents and also advancing the counter to point to the next table entry after execution.
Peter Naur recently wrote me that he considers the use of tables to control program flow as a basic idea of computer science that has been nearly forgotten; but he expects it will be ripe for rediscovery any day now.
The advantage of interpretive techniques are the compactness of representation, the machine independence, and the increased diagnostic capability.
A typical example is the use of a finite-state machine to encode a complex protocol or lexical format into a small tableJump tables can be especially efficient if the range tests can be omitted.
For example, if the control value is an enumerated type (or a character) then it can only contain a small fixed range of values and a range test is redundant provided the jump table is large enough to handle all possible valuesPrograms must be written for people to read, and only incidentally for machines to execute.Show me your flowchart and conceal your tables, and I shall continue to be mystified.
