Fortran Analysis Overview

Return to the Fortran tools home page

The analysis of code in a programming language is conventionally divided into three parts:

• lexical analysis to recognize the language's tokens;
• syntactical analysis to recognize significant groupings of tokens; and
• semantic analysis to associate meanings with the components of the resulting structure.

Fortran is unusual in that it was initially developed before techniques were developed to separate and simplify the three stages. It now has two source forms, fixed (the original Fortran source form), and free (a source form incorporating most modern concepts about language design). However, even the free source form has aspects that are recognized as awkward for analysis:

• No reserved keywords (which I consider a plus);
• Implicit typing;
• Statement functions with a syntax similar to assignments to an array element;
• Syntactical distinction between user defined and many language defined keywords;
• Global namespace for most language defined procedures;
• Minor other awkwardnesses, e.g., ENTRY, EQUIVALENCE.

In addition, the fixed form has several other problems:

• No significant spaces outside of character constants;
• Position dependent interpretation of syntactic entities; and
• The assumption that each text line has exactly 72 characters.

As a result the lexical analysis requires a subset of the information normally obtained from the parser, and the lexical analysis of a statement can sometimes require two passe. Note, however that normal code is sufficiently regular that it is very rare that a statement will require more than one pass by a well written Lexer. After the lexical analysis the syntactic analysis of Fortran is comparable to other large languages, and the semantic analysis is perhaps simpler, as almost all properties of interest (except potential benefits from inlining) can be determined by local analysis at the procedure level.

The detailed description of the lexical analysis of Fortran was given by Arthur H. J. Sale, "The Classification of FORTRAN Statements" The Computer Journal 14(1): 10-12 (1971), which was revised for Fortran 77 by J. K. Slape and Peter J. L. Wallis, "A Modification of Sale's Algorithm to Accommodate FORTRAN 77," The Computer Journal (34)1: 373-376 (1991). This algorithm, with a minor correction, is the basis of William Waite's analysis available as part of Eli's online documentation, and the online documentation for G77 by Craig Burley, particularly the section on "The Gnu Fortran Language".

Return to the Fortran tools home page