CRV2 MetricsCodeRev

No code metric will ever be able to substitute for a code review. There is a long list of quality and security characteristics that can be considered when reviewing code (such as but not limited to correctness, efficiency, portability, maintainability, reliability, securability, etc. sometimes called the '-ilities'). In this paper we will discuss metrics to reach the quickest and best solution for a given code review. No two code review sessions will be the same so some judgement will be needed to decide the best path. Ideally all code needs to be reviewed, but let's face it, not all code can be reviewed. So the best approach is to review the code with the highest risk.

Some Metric Benefits
The objective of code review is to detect development errors which may cause vulnerabilities, and hence give rise to an exploit. Code review can also be used to measure the progress of a development team in their practice of secure application development. It can pinpoint areas where the development practice is weak, areas where secure development practice is strong, and give a security practitioner the ability to address the root cause of the weaknesses within a developed solution. It may give rise to investigation into software development policies and guidelines and the interpretation of them by the users; communication is the key.

Metrics can also be recorded relating to the performance of the code reviewers and the accuracy of the review process, the performance of the code review function, and the efficiency and effectiveness of the code review function.



The figure above describes the use of Metrics throughout the code review process.

Defect Density
The average occurrence of programming faults per Lines of Code (LOC). This gives a high level view of the code quality but not much more. Fault density on its own does not give rise to a pragmatic metric. Defect density would cover minor issues as well as major security flaws in the code; all are treated the same way. Security of code can not be judged accurately using defect density alone.

Lines of Code (LOC)
The count of the executable lines of code. Commented-out code or spaces don't count. This is another metric in an attempt to quantify the size of the code. This gives a rough estimate but is not particularly scientific. Some circles of thinking believe that the estimation of an application size by virtue of LOC is professional malpractice!

Function Point
The estimation of software size by measuring functionality. The combination of a number of statements which perform a specific task, independent of programming language used or development methodology.

Risk Density
Similar to defect density, but discovered issues are rated by risk (high, medium & low). In doing this we can give insight into the quality of the code being developed via a  [X Risk / LoC] or [Y Risk / Function Point] value. (X&Y being high, medium or low risks) as defined by your internal application development policies and standards.

Example: 4 High Risk Defects per 1000 (Lines of Code) 2 Medium Risk Defects per 3 Function Points

Cyclomatic Complexity
Cyclomatic complexity (CC) is just one static analysis metric used to help establish risk and stability estimations on an item of code, such as a class or method or even a complete system. It was defined by Thomas McCabe in the 70's and it is easy to calculate and apply, hence its usefulness.

The McCabe cyclomatic complexity metric is designed to indicate a program's testability, understandability and maintainability, etc. This is accomplished by measuring the control flow structure, in order to predict the difficulty of understanding, testing, maintaining, etc. Once the control flow structure is understood one can gain a realization of the extent to which the program is likely to contain defects. The cyclomatic complexity metric is intended to be independent of language and language format that measures the number of linearly-independent paths through a program module. It is also the minimum number of paths that should be tested.

By knowing the cyclomatic complexity of the product, one can focus on the module with the highest complexity. This will most likely be one of the paths data will take, thus able to guide one to a potentially high risk location for vulnerabilities. (This is where the OWASP Code Review Top 9 comes in handy. ) The higher the complexity the greater potential for more bugs. The more bugs the higher the probability for more security flaws.

Does cyclomatic complexity reveal security risk? One will not know until after a review of the security posture of the module. The cyclomatic complexity metric provides a risk based approach on where to begin to review and analyse the code. Securing an application is a complex task and in many ways, complexity is just one enemy of security. Software complexity can make security and corruption hard to detect. Complexity of software increases over time as the product is updated or maintained.

Security has become a major component of software quality. Functional testing and software validation contribute to the quality of a given body of code, and yet allow the software to remain insecure and potentially vulnerable. An insecure product is not a quality product.

CC = Number of decisions +1

A decision could be considered commands such as:

If....else switch case catch while do and so on.....

As the decision count increases, so does the complexity and the number of paths. Complex code leads to less stability and maintainability.

The more complex the code, the higher risk of defects. One could establish thresholds for Cyclomatic complexity:

0-10: Stable code. Acceptable complexity 11-15: Medium Risk. More complex 16-20: High Risk code. Too many decisions for a unit of code. Modules with a cyclomatic complexity higher than 30 are extremely complex and should be split in smaller methods. Any module with a complexity greater than 50 is considered untestable.

Depth of Inheritance
Depth of inheritance: measures the inheritance level of a method. The deeper the level the greater the complexity involved in predicting its behavior and more importantly the impact changes will have.

Bad Fix Probability
Bad fix probability: This is the probability of an error accidentally inserted into a program while trying to fix a previous error. Cyclomatic Complexity: 1 – 10 == Bad Fix Probability: 5% Cyclomatic Complexity: 20 –30 == Bad Fix Probability: 20% Cyclomatic Complexity:  > 50 == Bad Fix Probability: 40% Cyclomatic Complexity: Approaching 100 == Bad Fix Probability: 60%

As the complexity of software increase so does the probability to introduce new errors.

Other aspects of software complexity
Thanks to John Wilander for pointing out that complexity in software has five dimensions
 * 1) Scale. More lines of code means higher complexity. This is often hard to avoid although code refactoring can reduce the code mass somewhat.
 * 2) Diversity. The number of technologies, frameworks, languages, versions, and even varying coding conventions -- they all drive complexity. This can be handled with policies but often times developers will oppose a strict list of what to use and what not.
 * 3) Connectivity. On the code level it's about coupling and cohesion and depth of inheritance. On a system level it's about the architecture number of connections and functions. Even higher it's about interconnecting systems and services both internally and across organization. In the server room it is about switches, routers, firewalls and other appliances which boils down to cabling complexity.
 * 4) Dynamics. Could the system even be described in a state diagram? On any reasonable level? As the number of states the system can in be grows, so does its complexity. The verdict is still out on if we have the tools or skills to manage software dynamics yet.
 * 5) Refinement. Over time software is refined or maintained. Understanding why things are done the way they are in a highly refined product is complex. This kind of complexity can be lowered through documentation, but not avoided.
 * 6) There are four categories of maintenance according to ISO/IEC 14764:
 * 7) Adaptive: Modification of a software product performed after delivery to keep a software product usable in a changed or changing environment.
 * 8) Corrective: Reactive modification of a software product performed after delivery to correct discovered problems.
 * 9) Perfective: Modification of a software product after delivery to improve performance or maintainability.
 * 10) Preventive: Modification of a software product after delivery to detect and correct latent faults in the software product before they become effective faults.

These are just some of the metrics to consider in an attempt to improve your code. From here things become more complex.

Inspection Rate
This metric can be used to get a rough idea of the required duration to perform a code review. The inspection rate is the rate of coverage a code reviewer can cover per unit of time. From experience, a rate of 250 lines per hour would be a baseline. This rate should not be used as part of a measure of review quality, but simply to determine duration of the task.

Defect Detection Rate
This metric measures the defects found per unit of time. Again, can be used to measure performance of the code review team, but not to be used as a quality measure. Defect detection rate would normally increase as the inspection rate (above) decreases.

Code Coverage
Measured as a % of LoC of function points, the code coverage is the proportion of the code reviewed. In the case of manual review we would aim for close to 100%, unlike automated testing wherein 80-90% is considered good. In order to ensure that the code coverage standards are met, some organizations might implement a safety check during the build process, so the build will fail if there are any piece of code that has not been tested or if it the coverage is under the desired percentage. The higher the percentage of code coverage, the better to ensure quality and prevent logic errors. When it comes to Unit Testing, code coverage is an important element because one might determine how to write unit test that covers all the possible execution paths and avoid any logic error that might cause a wrong behaviour in a piece of software at runtime.

Defect Correction Rate
The amount of time used to correct detected defects. This metric can be used to optimise a project plan within the SDLC. Average values can be measured over time, producing a measure of effort which must be taken into account in the planning phase.

Reinspection Defect Rate
The rate at which upon re-inspection of the code more defects exist, some defects still exist, or other defects manifest through an attempt to address previously discovered defects.