fpt - Fortran Program Metrics and Code Quality Indices
How do you assess the workload in maintaining a code - the cost of ownership?
How do you measure the quality of a Fortran code?
How do you assess the risk in using and maintaining a code?
Measuring the Workload of Maintaining Fortran Codes
There are three issues:
WinFPT and fpt report the size of the code as part of the code metrics report, for example:
Program Metrics =============== Main Code System Total Files Primary files 8 0 8 Include files 7 0 7 Code and comments Declaration lines 2817 0 2817 Executable lines 3629 0 3629 Total code lines 6446 0 6446
The cyclomatic complexity is a measure of the number of branches in the code. It is reported for the entire code, for example:
Main Code System Total Program flow and cyclomatic complexity Normal path code nodes 1785 0 1785 Exception code nodes 30 0 30 Total nodes 1815 0 1815 Normal path code edges 2562 0 2562 Exception code edges 218 0 218 Total code edges 2780 0 2780 Executable sub-programs 10 0 10 Normal path code complexity 777 0 777 Total code complexity 965 0 965 Complexity per sub-program 96.500 96.500
Code which handles exceptions, and code in system routines (if included in the analysis) are identified separately in the reports.
The program analysed for this report is a pre-compiler which contains several switches (Computed GOTO statements) and many handlers for specialised constructs. It is therefore very complex. A mean cyclomatic complexity of 10 to 20 per sub-program is more usual.
The cyclomatic complexities of the individual sub-programs are also reported, for example:
Cyclomatic Complexities ======================= Nodes Edges Complexity ERRPAR 29 43 14 ERRSCL 21 33 12 GNX 9 13 4 GNXEND 76 124 48 GNXINI 70 91 21 GNXPAR 1096 1688 592 GNXREA 35 53 18 GNXREL 335 521 186 GNXSDA 124 189 65 GNXTXT 20 25 5
This is useful in identifying hot-spots in maintenance. In the code analysed the routine GNXPAR is the parser. This has several 128-way switch lists on input character used to recognise and process variable names and numbers. The high complexity is a cause for concern!
Metrics of Fortran Code Quality - Quality from Counting Errors
fpt and WinFPT generate a simple measure of the quality of code based on the number of errors and warnings generated per thousand lines of code. Some errors and warnings are always generated when code is analysed. Some are generated only in response to specific tests such as the argument check and name check. It is best to write a script of standard tests which will be used at a site to test the code quality. For example:
! millibug.fsp 13-Sep-02 John Christopherson ! ***************************************************************************** ! % check alignment ! ! Ignore distinction between e.g. REAL X and REAL*4 X % argument check to ignore default sizes % check arguments ! % check data in COMMON blocks % check data in structures ! % check expressions ! % equivalence check to ignore default sizes % check equivalence ! % check names ! % check program flow ! % check usage ! ! End of millibug.fsp ! *******************************************************
fpt and WinFPT report the index of quality in the list file at the end of processing. For example, the report for GNX, an old VAX/VMS program, is:
Diagnostic Summary Total diagnostics 590 Highest severity 2 Notes 168 Warnings 273 Errors 149 Code Quality Index Error weighting 10 Total lines 10017 Index per 1000 lines 176.001
The index per 1000 lines, the "millibug rating" is computed here as:
1000 * (10 * Number_of_Errors + Number_of_Warnings) / Number_of_lines
The weighting factor for errors may be changed by the user. It is probably best written in the configuration file, config.fsp in the main installation directory.
The listing also shows the contributions made by different types of disgnostic to the overall quality index. The table for GNX, for example, shows:
Contributions to Code Quality Index No. Description Count Contrib 1269 FORTRAN keyword used as local identifier name. 2 2 1273 FORTRAN auxiliary keyword used as identifier name. 4 0 1281 FORTRAN intrinsic used as an array name 1 0 1459 Fortran keyword EXIT used, presumably to call system service 1 1 1585 Arguments inconsistent with sub-program declaration. 85 850 1887 Arguments inconsistent with other calls. 3 30 1967 Unable to determine value of a PARAMETER. 5 0 2241 References are made to sub-programs which have not been read. 1 0 2315 The file has been referenced elsewhere by a different name 24 0 2517 ANSI FORTRAN 77 intrinsic used as a local identifier 2 2 2847 Multiple initialisation of COMMON block 1 1 2849 DATA for COMMON outside BLOCK DATA 1 1 3047 Fortran 90 intrinsic name used as an identifier 2 2 3073 Danger of overflow due to different integer sizes 229 229 3235 Symbol is read but never written to 61 610 3237 Symbol is written to but never read 2 2 3239 Symbol is declared but never used 128 0 3241 Name is used inconsistently in different contexts 1 1 3243 Object is mis-aligned in memory 14 14 3247 Mixed EQUIVALENCE - Objects of different type or size 18 18 3291 String shorter than parameter in parameter assignment 5 0
Our experience is that any program with a millibug rating greater than 10 is unlikely to run without problems. The rating of 176 for GNX is alarming. The table of contributions shows that most of the problems come from mis-matched arguments and unassigned variables. This was a VAX program, where arguments with size mis-matches were still likely to be handled correctly and all unassigned variables were set to zero. It would not migrate well to a PC!
Metrics of Fortran Code Quality - Quality of Comments
fpt and WinFPT generate a measure of the number of comments in the code. This forms part of the metrics report. For example:
Main Code System Total Code and comments Declaration lines 2817 0 2817 Executable lines 3629 0 3629 Total code lines 6446 0 6446 Comment text lines 2019 0 2019 Comment separator lines 221 0 221 Blank lines 1331 0 1331 Total comment lines 3571 0 3571 Total lines 10017 0 10017 Trailing comments 540 0 540 Words in comments 13001 0 13001
Comment lines are classified as:
A count is also made of the number of trailing comments, delimited by "!" or (non-standard) "(* ... *)" symbols.
Probably the best measure of the level of commentary is the count of the number of separate words in the comment text lines and the trailing comments.
Metrics of Fortran Code Quality - Quality of Variable Names
Code written with short and meaningless variable names can be almost unmaintainable. It is difficult to write a tool to measure the quality of variable names, but two issues have been identified:
WinFPT and fpt count the number of names visible in multiple routines which are also used locally, and measure the mean lengths of the names visible in multiple routines. These are, for example, module, function, subroutine and COMMON block names, and the names of objects which are public in modules or in COMMON blocks.
The measures are written in the metrics report. For example:
Usage of Names for Objects Other than Local Variables Names used for non-locals 864 % used inconsistently 0.694 Mean length (Characters) 5.231
In the code analysed, only 0.694% of the names visible in multiple routines (exactly one name) was also used for a local variable. This is very good. The non-local names were on average only 5.231 characters long. This is not really long enough for a reliable naming convention or for the names to be fully meaningful. It is still very much better than some of the commercial codes which we have seen!
Risk Assessment of Fortran Codes
You inherit a code from another company, another department or from a legacy environment.
What is the risk in using and maintaining it?
There are two issues:
Copyright ©1995 to 2017 Software Validation Ltd. All rights reserved.