Detecting numerical problems in geodetic software
a consequence of the automation of calculation processes the demands on
geodetic software have risen tremendously. Automation, as a rule, puts
an end to direct checking by the user. This means that the software
needs to be extremely reliable and intelligent.
numerical weaknesses may result in hidden mistakes or cause program
crashes. Hidden mistakes produce results that are not entirely wrong
and are therefore not found immediately. Program crashes may cause data
losses. This paper therefore focuses on the finding of numerical
errors. It has been the aim of this paper to work out concrete tests
which enable users to check their software with a view to numerical
weaknesses. Tools have been worked out and precise instructions for
program testing have been included.
The methods applied are
derived from that field of software engineering which is concerned with
program testing. The program is looked at as a black box into which
inputs are fed and from which outputs are expected. Test specifications
for each test case contain precise instructions for the performance of
each test, the test data proper and finally the correct results.
Evaluation of the test results has been based on standard DIN 66 272
"Software product evaluation". An evaluation chart has been worked out
to categorize and assess the individual test cases.
To try out
the effectiveness of the test cases worked out the author has examined
and assessed two software products. The results have shown that the
tests are successful in finding numerical weaknesses and in identifying
differences in quality between the individual programs.