Skip to content
U.S. flag

An official website of the United States government

Official websites use .gov
A .gov website belongs to an official government organization in the United States.

Secure .gov websites use HTTPS
A lock ( ) or https:// means you’ve safely connected to the .gov website. Share sensitive information only on official, secure websites.

How To Write ISIS Tests with CTest and GTest

Helpful External documentation

Consider referencing these as you go along in case you need more in-depth info on how to work with gtest.

  • gtest primer : It is highly recommended that you read over the primer if you are not familiar with gtest.
  • Advanced gtest : This document covers a wide range of advanced options for testing tricky logic and customizing test output.
  • gmock Introduction : A great introduction to how gmock can be used to remove dependencies from tests.
  • gmock Matcher List : This whole document is helpful for gmock, but these matchers have some nice uses in gtest too.
App Conversion + Gtest PR Checklist

Things that all new test additions need to have done for an ISIS app that is still using the old paradigm.

  • New .cpp/.h source files added as <appname>.cpp/<appname>.h in all lowercase.
  • Contents of main.cpp moved into <appname>.cpp with a function named <appname> in lower camel case.
  • Main.cpp replaced with call to app function in <appname>.cpp
  • New app tests added as ISIS3/isis/tests/FunctionalTests<appname>.cpp in upper camel case
  • New GTests prefixed with FunctionalTest<appname>
  • Old Makefile tests removed
  • New tests passing on Jenkins
  • Old test data removed

Refactoring ISIS3 Applications

In order to better integrate with the gtest unit test framework, all of our application logic needs to be callable programmatically.

This is a first step towards several places:

  1. Improving granularity of application testing
  2. Reducing testData size
  3. Improving application test run time

Creating a basic callable function

For the rest of this document, we will use appname as the name of the application that is being worked on. Simple example at https://github.com/DOI-USGS/ISIS3/tree/dev/isis/src/base/apps/crop

  1. In the appname folder, create two new files, appname.cpp and appname.h. These files are where the application logic will live.
  2. In appname.h and appname.cpp create a new function in the Isis namespace with the following signature void appname(UserInterface &ui). If the application has a from cube, add a second function with the input cube as the first argument void appname(Cube incube, UserInterface &ui).
  3. Copy the contents of the existing IsisMain function currently located in main.cpp into the new void appname(Cube incube, UserInterface &ui) function. (if the app doesn't take any file parameters (i.e., network, text, cube list...) copy the contents intovoid appname(UserInterface &ui)). If there is no input cube, but there are other input files, add them as input parameters similar to how the input cube was done, see spiceinit and cnetcheck tests for examples.
  4. Modify void appname(UserInterface &ui) to open the input cube, usually "from", and/or other files and call the second function void appname(Cube incube, UserInterface &ui)
  5. Copy any helper functions or global variables from main.cpp into appname.cpp. This is to not pollute the Isis namespace and avoid redefining symbols
  6. Prototype any helper functions at the top of appname.cpp and define them at the bottom of appname.cpp. Do not define them in appname.h.
  7. Put all the required includes in appname.cpp and appname.h.
  8. Remove the call to get the UserInterface from appname.cpp; it usually looks like UserInterface &ui = Application::GetUserInterface();.
  9. In main.ccp, add the following
#include "Isis.h"

#include "UserInterface.h"
#include "appname.h"

using namespace Isis;

void IsisMain() {
  UserInterface &ui = Application::GetUserInterface();
  appname(ui);
}

If your application uses Application::Log()

Due to how the Application singleton works, calling Application::Log() outside an ISIS application currently causes a segmentation fault. To avoid this, modify the new appname function to return a Pvl that contains all the PvlGroups that need to be logged instead of calling Application::Log(). Then, change your main.cpp to

#include "Isis.h"

#include "Application.h"
#include "Pvl.h"
#include "UserInterface.h"
#include "appname.h"

using namespace Isis;

void IsisMain() {
  UserInterface &ui = Application::GetUserInterface();
  Pvl results = appname(ui);
  for (int resultIndex = 0; resultIndex < results.groups(); resultIndex++) {
    Application::Log(results.group(resultIndex));
  }
}

Wherever Application::log is called with some pvl group (here called pvlgroup), in the app func replace with:

if (log){ 
  log->addGroup(pvlgroup);
}

Creating a more complex callable function

The basic interface that we've created so far is simply a mirror of the application command line interface. In order to further improve the testability of the application, we should break the application down into functional components.

Separating parameter parsing

The first step is to separate the UserInterface parsing from the program logic.

All the UserInterface parsing should be done in the appname(UserInterface &ui) function. Then, all the program logic should be moved into another function also called appname. The signature for this function can be quite complex and intricate. Several tips for defining the appname function signature are in the next sections, but there is no perfect way to do this for every application.

Once the appname function is defined, the appname(UserInterface &ui) function simply needs to call it with the appropriate parameters once it has parsed the UserInterface.

Separating I/O

Most ISIS3 applications were designed to read their inputs from files and then output their results to the command line and/or files. Unfortunately, gtest is very poorly setup to test against files and the command line. To work around this, it is necessary to remove as much file and command line output from the new appname functions as possible. Here are some examples of how outputs can be separated from the application logic:

  1. Anything that would be logged to the terminal should be simply returned via the log pointer. This way, it can be programmatically validated in gtest. This already needs to be done because of issues with Application::Log().
  2. No input filenames should be passed as arguments. All files required by the program should be opened and converted into in-memory objects and then passed to the function. This will help eliminate the need for test data files in many applications. Make sure that for cubes, the appropriate CubeAttributeInput and CubeAttributeOutput values are set!

Process, helper functions, and global variables

Many ISIS3 applications make sure of the Process class and its subclasses. While these classes make file I/O easier, they also tightly couple it to the application logic! Some of them can take objects like cubes instead of filenames, but not all of them. They may need to be refactored to use objects instead of filenames. For now, where-ever possible use objects, but it is acceptable to use filenames if an object cannot be used.

Many of the Process subclasses use process functions to operate on cubes. These helper functions will need to be pushed into the ISIS3 library. There is a chance that there will be symbol conflicts due to multiple applications using functions with the same name. In this case, the function can simply have its name modified. This is also a natural point at which application logic can be segmented and tested separately.

Because of how the Process subclasses use process functions, many older ISIS3 applications use global variables to pass additional parameters to helper functions. This can cause serious issues because applications are no longer self-contained executables. To help with this, ProcessByBrick and ProcessByLine have been modified to use a lambda function with captures as the process function. All the previously global variables can now be defined in the appname function and then captured in the lambda function.

For example, here's an app called checkerboard which generates an artificial cube with a checkerboard pattern. It requires access to the variable size to work, but ProcessByLine.StartProcess only supports functions Buffer objects as input. Therefore, the app uses a global variable in order for the checkerboard function to have size in its scope:

// original foo.cpp
#include "Isis.h"

// .. some other incs 

int size;

void do_process(Buffer &in, Buffer &out) {
  // do stuff with in, out and size
}

void isismain() {
  UserInterface &ui = Application::GetUserInterface();
  size = ui.GetInteger("SIZE");

  ProcessByLine p;

  p.SetInputCube("FROM");
  p.SetOutputCube("TO");

  p.StartProcess(do_process);
}

can be refactored using a lambda which captures the size variable:

// callable foo.cpp
#include "Isis.h"

// .. some other incs 

void IsisMain() {
  UserInterface &ui = Application::GetUserInterface();

  // int is defined in the scope of main
  int size = ui.GetInteger("SIZE");

  // lambda equivalent of the checkerboard function, capture is set to get 
  // all local variables by reference 
  auto do_process = [&](Buffer &in, Buffer &out)->void {
    // do stuff with in, out and size 
  };  

  ProcessByLine p;

  p.SetInputCube("FROM");
  p.SetOutputCube("TO");

  p.StartProcess(do_process);
}

Creating a new test suite

For a Class

  1. Create a new file ClassNameTests.cpp in isis/tests/. For example, the tests for the Cube class should be in CubeTests.cpp.
  2. Add #include "gmock/gmock.h" to the file.
  3. Write your test cases.
  4. Delete old tests

For an Application

  1. Create a new file FunctionalTestAppname.cpp in isis/tests/. For example, the tests the application isisimport should be in FunctionalTestIsisImport.cpp.
  2. Add your tests to this file
  3. Delete the tests and all the test directories

Note: If your tests require the use of ISIS's Null value, you will need to include SpecialPixel.h

Test names

We use gtest as our unit testing framework, but use ctest to actually run the tests. Test name can be defined in the test definition (cpp file). The naming convention for gtests is UnitTest<Object><test case> for unit tests and FunctionalTest<app name><test case> for app tests. Both are upper camel case. The documentation for how this works can be found in cmake's documentation.

Note: App names are considered a single word for camel case purposes. So when naming a test file or function, only capitalize the first letter of the app name, even if the app name is made up of multiple words.

To run the gtests for a specific class, use ctest -R ClassName

To run all gtests, use ctest -R "\." -E "(_app_|_unit_)". Not that this command simply excludes old Makefile based tests that contain the substrings _app_ and _unit_ in their respective names.

Basic tests

TEST(Foo, Bar) will produce the test Foo.Bar.

Test fixtures

TEST_F(Foo, Bar) will also produce the test Foo.Bar.

The names of fixtures are at the developer's decretion. Fixtures that require extra data should have their data files placed in ISIS3/isis/tests/data inside a folder matching the fixture name.

Parameterized tests

TEST_P(Foo, Bar) {
...
}
INSTANTIATE_TEST_CASE_P(
      Baz,
      Foo,
      ::testing::Values(T x, T y, T z));
will produce 3 different tests that are named 1. Baz/Foo.Bar/x 1. Baz/Foo.Bar/y 1. Baz/Foo.Bar/z

General Testing Tips

Testing exceptions

Testing for exception throws in gtest can be rather convoluted. Through testing and looking at what others have done, we have settled on using the following setup to test an exception throw:

try {
  // Code that should throw an IException
  FAIL() << "Expected an exception to be thrown";
}
catch(IException &e) {
  EXPECT_TRUE(e.toString().toLatin1().contains("Substring of the expected exception message"));
}
catch(...) {
  FAIL() << "Expected an IExcpetion with message message: \"The full expected exception message.\"";
}

Be sure to choose a substring of the exception message that will uniquely identify it, but only use parts of the string identified in the actual code. Anything like **USER ERROR** or in IsisAml.cpp at 1613 that are added based on user preference settings should not be included in the substring. The IException and Preferences classes test that these are properly added onto the error messages.

Normally, if the error message does not contain the substring, gtest will only output the expression evaluated to false. To make the failure message more helpful, add << e.toString().toStdString() which outputs the full error message if the test fails.

Testing floating point numbers

Comparison between floating point numbers is a common problem in both writing and testing software. Carefully choosing test data can help avoid comparisons that are likely to cause problems. Whenever possible, use test data such that output values can be exactly stored in a double-precision floating point number (c++ double). For example, if a function takes its input and divides it by 116, a good test input would be 232 because 232/116 = 2 which can be stored exactly in a double. On the other hand, 1 would be a poor test input because 1/116 cannot be precisely stored in a double. If a comparison between numbers that cannot be stored precisely is required, gtest provides Special Assertions to make the comparisons more tolerant to different rounding.

Test fixtures

Because each test case is a completely fresh environment, there is often set-up code that is shared between multiple test cases. You can use a test fixture to avoid copying and pasting this set-up code multiple times. See the gtest documentation for how to create and use fixtures in gtest.

In order to conform to our test case naming conventions, all test fixtures need to be named ClassName_FixtureName. For example, a fixture that sets up a BundleSettings object in a not-default state for the BundleSettings unit test could be called BundleSettings_NotDefault.

Test parameterization

If the same process needs to be run with several different inputs, it can be easily automated via value-parameterized tests. In order to maintain test naming conventions, when you call the INSTANTIATE_TEST_CASE_P macro, make sure the first parameter is. ClassName. For example

INSTANTIATE_TEST_CASE_P(
      BundleSettings,
      ConvergenceCriteriaTest,
      ::testing::Values(BundleSettings::Sigma0, BundleSettings::ParameterCorrections));

This will ensure that the tests start with BundleSettings.

Tip

See the ISIS App Testing Notebook for examples you can copy from.