Fortran Analysis, Engineering & Migration

Analysing and re-engineering WRF

WRF is one of the most important programs used in weather forecasting and in the analysis and prediction of climate change. WRF is maintained at The University Corporation for Atmospheric Research, Boulder, Colorado. It is written in Fortran, and SimCon has a long-term project to analyse and to re-engineer it.

WRF is written to a very high engineering standard. It is a remarkable achievement that it runs on anything from a single processor to large multiprocessor systems with a variety of parallel configurations. However, WRF is a very large program, and it would be surprising if a code of this size did not have some coding anomalies.

The Analyses Have Shown

• Errors in INTENT Declarations: in WRFV3.6.1, 138 of the 33,126 INTENT (IN) declarations are violated.
• Errors in accessing OPTIONAL arguments. There are 3,194 optional arguments in WRFV3.6.1 and of these, 1,360 may be accessed without a check that the argument is present
• Potential errors in the precision of real numbers written in the code. In WRFV3.6.1 there are 157,120 floating point numbers in the code, and of these, 26,814 have no specification of precision and are used in a context where a higher precision than the default would be appropriate

All of the scripts used in the analyses can be downloaded from this site. They have been written to be as simple and straightforward to use as possible and can be run quite quickly and very easily on a single processor Linux system.

You can do this at home!

And we would like you to, because with different compilers, different systems and different configurations you may see different results and we would like to know that.

The scripts detect the coding anomalies, and re-write the WRF code to correct them. Please install and run them as described here.

We have carried out this study for three reasons:

• As a service to the WRF community. All of the changes and corrections made to the WRF code are to the original pre-processor files and could be incorporated directly into future WRF distributions.
• As a demonstration of what can be done to re-engineer and to test large Fortran codes.
• To expose issues in the Fortran language. Some of the issues arise because of traps in the language, and it is important to see how these could be addressed.

We would welcome feedback from the WRF community on the priorities for this research. Is it more important to analyse other versions of WRF, or to extend the analyses to other issues? Please contact us if you would like to comment.

This work is carried out in collaboration with Prof. Mark Anderson at Edge Hill University and Dave Gill at The University Corporation for Atmospheric Research, Boulder, Colorado, and by Brian Farrimond and John Collins at SimCon. The project is partly funded by The United States Air Force.

Copyright ©1995 to 2024 Software Validation Ltd. All rights reserved.