Learn R Programming

svUnit (version 1.0.6)

svUnit-package: svUnit: 'SciViews' - Unit, Integration and System Testing

Description

A complete unit test system and functions to implement its GUI part.

Arguments

Details

The SciViews 'svUnit' package defines a framework for testing R code, not unlike jUnit for Java. It is inspired on the checkxxx() functions from the 'RUnit' package and the same test unit files should be compatible with both 'svUnit' and 'RUnit'. However, the internal implementation is completely different and svUnit can also be used interactively, while 'RUnit' is only designed to run test units written in files on disks.

The test unit framework provided in 'svUnit' is based on tests, also called assertions, implemented in checkxxx() functions. For instance, the checkTrue(expr) function check if its 'expr' argument returns TRUE. Results of these assertions are collected in a centralized logger located in the .Log object in .GlobalEnv. This is a 'svSuiteData' object with data about the context of the tests (see for instance, lastTest(), lastSuite() or metadata(.Log)).

Assertions can give three results: (1) TRUE if success, (2) FALSE in case of failure (in our example, 'expr' in checkTrue(expr) did not return TRUE), and (3) NA if the code in 'expr' cannot be parsed or executed correctly. All these errors or failures are catch and recorded in the logger, as individual 'svTestData' objects.

Both the logger ('svSuiteData' object) and test records inside it ('svTestData' objects) have convenient methods to visualize information they contain: print(), summary() and stats() methods. Access to the individual test records in the logger is done with list-like instructions: .Log$mytest returns the 'svTestData' object named 'mytest', itself the result of running test in the 'mytest' test function (i.e., runTest(mytest), see hereunder). Assertions run at the command line, outside of specific contexts provided by test functions, test units and test suites (see hereunder) are recorded under the 'eval' 'svTestData' object in the logger (i.e., .Log$eval).

Since a 'svSuiteData' object (the logger) is also an environment, you can get the list of all test records it contains using ls(.Log), and you can eliminate a given test record using rm(mytest, envir = .Log).

Test cases are collections of assertions with the satellite code needed to build example or situations to be tested. They are collected together in argument-less functions with class being 'svTest'. See ?svTest for further explanations and a couple of example test cases/test functions.

In its simplest instance, a test function is defined as a separate R object loaded in memory (unlike RUnit where all test must be defined in files). You run it simply by using runTest(mytest). A slightly more structured way to work is to attach the test function to the object it testes. You use test(myobj) <- testmyobj to do so, and retrieve it with test(myobj). Now, the test function always follows the tested object. Testing the object is still simple by using runTest(myobj), which is totally equivalent to runTest(test(myobj)). One can determine if an object has a test function associated, or is a test function itself by using is.test(myobj).

Several test functions can be collected together in so-called test units. A test unit only exists on disk. It is a file named 'runit*.R' containing sourceable R code with test functions having names starting with 'test' (unlike 'RUnit', the default convention of file names starting with 'runit' and test function names starting with 'test' is not customizable in 'svUnit'). One can also define special .setUp() and .tearDown() functions in the test unit. The first function will be run before each test function, and the latter one will be run after it. Test units are created manually, or from a collection of objects with associated test functions loaded in an environment (usually .GlobalEnv) thanks to the makeUnit() method. These units should be mutually compatible with those used in the 'RUnit' package (at least this is verified with version 0.4-17 of 'RUnit').

Test units defined for packages should be located in the package /runitTests subdirectory (/inst/runitTests for source of the package) or one of its subdirectories. That way, they are located automatically by the function svSuiteList() that also automatically detects all objects with associated test functions loaded in .GlobalEnv. Test suites are 'svSuite' objects with a list of test units or test objects to collect in the suite. Thus, svSuiteList() automatically builds such a suite with all tests it finds in R, with many possibilities to filter packages' test units, objects' test functions, or to add non standard directories with test units, for instance. See ?svSuite for more details on creating and using these suites.

A GUI (Graphical User Interface) is provided to automatically build and run tests suites and to get a graphical (tree) interactive report of the results in the Komodo Edit or IDE code editor, together with the SciViews-K extension. If you want to use this (optional) GUI, you have to install required software components on your machine.

Finally, the 'svUnit' framework is compatible with R CMD check (see the manual "Writing R extensions"). You simply define man pages (.Rd files) with an example section running selected test units from your package. The function errorLog() is designed to generate and error if one or more tests failed or raised an error during R CMD check, and it should be used at the end of the example that runs your unit test(s). That way, R CMD check is interrupted and a detailed report of the tests that failed or raised an error is printed. See an example in ?unitTests.svUnit.

References

There is a huge literature and unit testing. An easy starting point is: https://en.wikipedia.org/wiki/Unit_test.

See Also

Useful links:

Examples

Run this code
# NOT RUN {
# Clear the logger
clearLog()

# Design and attach a simple test function to an object
foo <- function(x, y = 2)
  return(x * y)
testfoo <- function() {
  #DEACTIVATED()	# Use this to deactive the test (notice placed in the log)
  checkEqualsNumeric(5, foo(2),  "Check return of foo()")
  checkException(foo("b"),       "Wrong first argument")
  checkException(foo(2, "a"),    "Wrong second argument")
}
# Attach this to the foo function
test(foo) <- testfoo

# Run this test
runTest(foo)

# Inspect the result
ls(.Log)
.Log$`test(foo)`
# This test fails. You see that the test function requires that foo(2) = 5
# and the actual implementation returns 4. This is a trivial, useless example,
# but you are supposed to correct the function. For instance:
foo <- function(x, y = 2)
  return(x * y + 1)
test(foo) <- testfoo

(runTest(foo))	# Now, that's fine!
# }

Run the code above in your browser using DataLab