History: Visual Unit Testing Framework
Source of version: 5
Copy to clipboard
__NOTE: As of right now, this documentation is preliminary, incomplete, and subject to change.__
{maketoc}
!__Abstract__
This is the documentation for OGRE's visual unit testing framework. The framework allows you to perform image-based comparisons of test scenes between builds.
!__Introduction__
OGRE already uses [http://sourceforge.net/projects/cppunit/|CppUnit] for a selection of unit tests that cover the basics: Vectors, String functions, etc; however, this handful of existing tests is far from comprehensive. Moreover, given the high degree of interdependency and the dependence on graphics API's and the like, traditional unit testing really wouldn't be possible without a very complex testing setup involving mock rendersystems and a huge amount of testing code.
Even if traditional unit testing were doable without significant commitment, it still wouldn't be especially helpful in the context of a rendering engine (you really can't effectively decide "Is the rendered image correct?" with just assertions and so forth).
Given that the output of a rendering engine is an image, why not test it using just that? This framework aims to make testing possible by creating test scenes that can be screen-captured and compared between builds. This allows for features to be tested very simply (just implement the feature in a simple test scene, and the framework does the rest; there's no need for assertions or elaborate test cases).
!__Build Details__
Simply build with the OGRE_BUILD_TESTS option enabled in CMake. Also, note that test plugins are fully compatible with the sample browser.
!__Running Tests__
!!__Generating Reference Images__
Due to differences in drivers and so forth, it works best to generate a reference image set for each machine you will be testing with, this greatly reduces the chance of false positives due to driver issues.
The easiest way to generate reference image sets, is as follows:
''__1.__ Build Ogre with the CMake option __OGRE_BUILD_TESTS__ enabled.''
''__2.__ Run the TestContext executable from the command line with the following options:''
{CODE(wrap="1", colors="c++")}TestContext.exe -r -d{CODE}
The __-r__ generates a reference set, and the __-d__ forces the config dialog to appear. Select your preferences from the dialog.
A few notes:
* Lower resolutions are highly recommended, time to do the image comparison and disk space used scale up quickly with over 100 tests.
* VSync isn't recommended, it just slows down execution of the tests (since they run for a set number of frames, and VSync locks the max framerate).
* Fullscreen isn't recommended, I haven't tested with it much, and it may produce different results than windowed mode.
''__3.__ Repeat step 2 for each render system you're using.''
!!__Testing__
Once you have reference image sets ready, you can run tests.
The testing is integrated with Ogre's CMake system, so the easiest way of running tests is through your build environment:
__Using Unix Makefiles:__
* {CODE(wrap="1", colors="c++")}make test{CODE}
__Using Visual Studio:__
* Build the RUN_TESTS project
__Using NMake Makefiles: (run from the "Visual Studio Command Prompt")__
* {CODE(wrap="1", colors="c++")}nmake test{CODE}
For other build systems, consult CMake/CTest docs, in general there should be a "target" or "project" or the equivalent for running tests.
Alternatively, you can run it directly from ctest, from the command line in your build directory (you may need to specify a build configuration):
{CODE(wrap="1", colors="c++")}ctest -C Release{CODE}
!__CDash and Automated Testing__
Using CMake and CTest, it's possible to upload build and test results to a CDash web dashboard.
!!__Experimental Builds__
The simplest way of running tests and uploading results to the dashboard is through an 'Experimental' target in your build system (with makefiles this'll be "make Experimental", VS will have an "Experimental" target).
As above, you can run it directly from ctest:
{CODE(wrap="1", colors="c++")}ctest -C Release -D Experimental{CODE}
!!__CTest Scripts__
CTest allows you to use .cmake scripts to script the build/test process, which is useful for automated building, you can test with a script like so:
{CODE(wrap="1", colors="c++")}ctest -S script.cmake -V{CODE}
The -V sets verbose output (otherwise it's almost silent, -VV makes it even more verbose).
See below for example Nightly and COntinuous build scripts.
__Note that the scripts I've provided use NMake in Windows, so you'll need to run them from the "Visual Studio Command Prompt".__
!!__Nightly Builds__
Nightly builds grab a snapshot of the source repository once a day, built, test, and upload results.
You'll need to use your OS's scheduling system to run the script (see this CTest wiki entry for details on setting this up in various OS's).
Here's the script, it's sort of s template, there's a lot of specific details that will need configuration, so you'll want to read through it and set it up to match your needs:
{CODE(wrap="1", colors="c++")}
# Where to find OGRE's dependencies (very important in Windows, /usr should work in Unix)
set(DEPENDENCIES_DIR "...")
# determine home directory
if (WIN32)
# ctest seems to choke on backslash path separators...
string(REPLACE "\\" "/" HOME_DIRECTORY "$ENV{HOMEDRIVE}$ENV{HOMEPATH}")
elseif (UNIX)
set(HOME_DIRECTORY "$ENV{HOME}")
endif ()
# choose where you want source/builds to go
set(CTEST_SOURCE_DIRECTORY "${HOME_DIRECTORY}/dashboards/ogre/nightly/source")
set(CTEST_BINARY_DIRECTORY "${HOME_DIRECTORY}/dashboards/ogre/nightly/build")
# set any additional build options you need here
set(BUILD_OPTIONS
"OGRE_BUILD_TESTS=ON" # tests need to be enabled to do unit testing
"OGRE_DEPENDENCIES_DIR=${DEPENDENCIES_DIR}" # set where to find dependencies
)
# Everything after here should (hopefully) work without any editing
#--------------------------------------------------------------
# site name will just be hostname
site_name(CTEST_SITE)
if(UNIX)
# In Unix (including OSX, Cygwin, et al) use plain ol' makefiles
set(CTEST_CMAKE_GENERATOR "Unix Makefiles")
set(CTEST_BUILD_NAME "${CMAKE_SYSTEM_NAME}-make-gcc")
elseif(WIN32)
# This must be run from the Visual Studio command prompt for CMake to find everything
set(CTEST_CMAKE_GENERATOR "NMake Makefiles")
set(CTEST_BUILD_NAME "${CMAKE_SYSTEM_NAME}-nmake-vs")
endif()
# Configuration (release will be quickest to run the tests)
set(CTEST_BUILD_CONFIGURATION "Release")
# Setup build options
set(CTEST_BUILD_OPTIONS "")
foreach (opt ${BUILD_OPTIONS})
set(CTEST_BUILD_OPTIONS "${CTEST_BUILD_OPTIONS}-D${opt} ")
endforeach ()
# memcheck/coverage aren't set up (yet?)
set(WITH_MEMCHECK FALSE)
set(WITH_COVERAGE FALSE)
# look for mercurial executable
find_program(CTEST_HG_COMMAND NAMES hg)
# do a fresh clone if necessary
if(NOT EXISTS "${CTEST_SOURCE_DIRECTORY}")
set(CTEST_CHECKOUT_COMMAND "${CTEST_HG_COMMAND} clone https://bitbucket.org/RileyA/ogre-gsoc-testingframework ${CTEST_SOURCE_DIRECTORY}")
endif()
# tell CTest how to update
set(CTEST_UPDATE_COMMAND "${CTEST_HG_COMMAND}")
# set this all up
set(CTEST_CONFIGURE_COMMAND "${CMAKE_COMMAND} -DCMAKE_BUILD_TYPE:STRING=${CTEST_BUILD_CONFIGURATION}")
set(CTEST_CONFIGURE_COMMAND "${CTEST_CONFIGURE_COMMAND} ${CTEST_BUILD_OPTIONS}")
set(CTEST_CONFIGURE_COMMAND "${CTEST_CONFIGURE_COMMAND} \"-G${CTEST_CMAKE_GENERATOR}\"")
set(CTEST_CONFIGURE_COMMAND "${CTEST_CONFIGURE_COMMAND} \"${CTEST_SOURCE_DIRECTORY}\"")
# do a full build, from an empty bin directory
SET (CTEST_START_WITH_EMPTY_BINARY_DIRECTORY_ONCE 1)
# build and quit
ctest_start("Nightly")
ctest_update()
ctest_configure()
ctest_build()
ctest_test()
ctest_submit()
# manually call post-test cleanup script
exec_program("cmake -P ${CTEST_BINARY_DIRECTORY}/Tests/VisualTests/PostTest.cmake")
{CODE}
!!__Continuous Builds__
You can also run a continuous build setup, that checks the repository every few minutes, and does a build/test/upload whenever new changes are found.
This script will probably take more setup work than the Nightly one, so be sure you look over it and configure as needed:
{CODE(wrap="1", colors="c++")}
# Where to find OGRE's dependencies (very important in Windows, /usr should work in Unix)
set(DEPENDENCIES_DIR "...")
# How long to run in seconds
set(TIME_TO_RUN 43200) # default of 12hrs
# How long between checking for changes
set(TIME_BETWEEN_UPDATES 300) # default of 5mins
# determine home directory
if (WIN32)
# ctest seems to choke on backslash path separators...
string(REPLACE "\\" "/" HOME_DIRECTORY "$ENV{HOMEDRIVE}$ENV{HOMEPATH}")
elseif (UNIX)
set(HOME_DIRECTORY "$ENV{HOME}")
endif ()
# choose where you want source/builds to go
set(CTEST_SOURCE_DIRECTORY "${HOME_DIRECTORY}/dashboards/ogre/continuous/source")
set(CTEST_BINARY_DIRECTORY "${HOME_DIRECTORY}/dashboards/ogre/continuous/build")
# set any additional build options you need here
set(BUILD_OPTIONS
"OGRE_BUILD_TESTS=ON" # tests need to be enabled to do unit testing
"OGRE_DEPENDENCIES_DIR=${DEPENDENCIES_DIR}" # set where to find dependencies
)
# Everything after here should (hopefully) work without any editing
#--------------------------------------------------------------
# site name will be hostname
site_name(CTEST_SITE)
if(UNIX)
# In Unix (including OSX, Cygwin, et al) use plain ol' makefiles
set(CTEST_CMAKE_GENERATOR "Unix Makefiles")
set(CTEST_BUILD_NAME "${CMAKE_SYSTEM_NAME}-make-gcc")
elseif(WIN32)
# In Windows, NMake seems to be the only way of automating this
# This must be run from the "Visual Studio Command Prompt" for CMake to find everything
set(CTEST_CMAKE_GENERATOR "NMake Makefiles")
set(CTEST_BUILD_NAME "${CMAKE_SYSTEM_NAME}-nmake-vs")
endif()
# Configuration (release will be quickest to run the tests)
set(CTEST_BUILD_CONFIGURATION "Release")
# Setup build options
set(CTEST_BUILD_OPTIONS "")
foreach (opt ${BUILD_OPTIONS})
set(CTEST_BUILD_OPTIONS "${CTEST_BUILD_OPTIONS}-D${opt} ")
endforeach ()
# memcheck/coverage aren't set up (yet?)
set(WITH_MEMCHECK FALSE)
set(WITH_COVERAGE FALSE)
# look for mercurial executable
find_program(CTEST_HG_COMMAND NAMES hg)
# do a fresh clone if necessary
if(NOT EXISTS "${CTEST_SOURCE_DIRECTORY}")
set(CTEST_CHECKOUT_COMMAND "${CTEST_HG_COMMAND} clone https://bitbucket.org/RileyA/ogre-gsoc-testingframework ${CTEST_SOURCE_DIRECTORY}")
endif()
# tell CTest how to update
set(CTEST_UPDATE_COMMAND "${CTEST_HG_COMMAND}")
# set this all up
set(CTEST_CONFIGURE_COMMAND "${CMAKE_COMMAND} -DCMAKE_BUILD_TYPE:STRING=${CTEST_BUILD_CONFIGURATION}")
set(CTEST_CONFIGURE_COMMAND "${CTEST_CONFIGURE_COMMAND} ${CTEST_BUILD_OPTIONS}")
set(CTEST_CONFIGURE_COMMAND "${CTEST_CONFIGURE_COMMAND} \"-G${CTEST_CMAKE_GENERATOR}\"")
set(CTEST_CONFIGURE_COMMAND "${CTEST_CONFIGURE_COMMAND} \"${CTEST_SOURCE_DIRECTORY}\"")
# do a full build once
set(CTEST_START_WITH_EMPTY_BINARY_DIRECTORY_ONCE 1)
set(STARTED_CTEST FALSE)
# start continuous integration
while (${CTEST_ELAPSED_TIME} LESS ${TIME_TO_RUN})
set (START_TIME ${CTEST_ELAPSED_TIME})
# make sure this only gets called once for each build/test
if(NOT STARTED_CTEST)
ctest_start("Continuous")
set(STARTED_CTEST TRUE)
endif()
ctest_update(RETURN_VALUE NUM_UPDATED)
# if files were updated: build, test and submit
if (${NUM_UPDATED} GREATER 0)
ctest_configure()
ctest_build()
ctest_test()
ctest_submit()
# manually call post-test cleanup script
exec_program("cmake -P ${CTEST_BINARY_DIRECTORY}/Tests/VisualTests/PostTest.cmake")
set(STARTED_CTEST FALSE)
else ()
exec_program("cmake -E echo \"No Updates found, waiting...\"")
endif ()
# wait before checking again
ctest_sleep( ${START_TIME} ${TIME_BETWEEN_UPDATES} ${CTEST_ELAPSED_TIME})
endwhile()
{CODE}
!__TestContext Command Line Options__
Note that this is using Ogre's built-in [http://www.ogre3d.org/docs/api/html/group__General.html#gad78e25bde5597796c07e75d4f857a3cd|findCommandLineOpts] which does not allow for combining of options (i.e. foo -abc bar would have to be written as foo -a -b -c bar)
__-r__
Generate a reference set.
__-m "[[comment]"__
Add an optional comment to be associated with the generated image set.
__-ts "[[test set name]"__
Select the test set to use (default is 'VTests').
__-c "[[image set name to compare against]"__
Select which image set you want to compare this run with (default is 'Reference').
__-n "[[name]"__
Specify a name for this set (omitting this, or choosing 'AUTO' will result in an automatically generated name).
__-h -help or --help__
Usage details.
__--no-html__
Suppress html output.
!__The Output__
Whenever a set is created, the test images themselves, along with a small config file containing data about the set (resolution, date/time, name, etc), are created in a new directory (see ((Visual Unit Testing Framework|#Where to Find the Output|below)) for details on directory structure).
The primary output is an HTML document containing an overview of the test, and side-by-side images of the reference image set and the newly generated images. A small linked javascript file allows for some basic diffing to be done within a web browser (requires HTML5/Canvas).
[http://rileyadams.net/gsoc/July5/out.html|Here] is a sample of the html output.
!!__Where to Find the Output__
Output is generated in the same directory as the logs and cfg's for the sample browser. This is generally in your My Documents or home directory (or your OS's equivalent). There should be an Ogre directory, with a subdirectory for the Ogre version. From there the structure looks like:
*VisualTests
** [[Test set name]
*** [[Rendersystem]
**** out.html
**** [[Reference]
***** info.cfg
***** Reference screenshots (.png's)...
**** [[Test set name]_[[date]
***** Info.cfg
***** Screenshots (.png's)...
!__Image Comparison__
The images are compared to reference images using a selection of common metrics. A failed test will report values for the following metrics in the HTML output:
The most basic is just the __absolute difference__; how many pixels differ between two images.
Next is the __Mean Squared Error (MSE)__, which, as the name suggests produces the average squared error (difference between the images). Lower is better.
Next is the __Peak Signal-to-Noise Ratio (PSNR)__, which measures the ratio between the maximum signal (in this case, full color values in each channel), and corrupting noise (the differences in the images). Higher values of this metric are better.
Last, is the __Structural Similarity (SSIM) index__, which is a more recent development (see [https://ece.uwaterloo.ca/~z70wang/publications/ssim.html|this] 2004 paper for in-depth details), and aims to provide a metric better related to human preception (images with identical MSE may actually be of very different quality levels). It gives a value in the range of -1 to 1, (with 1 being identical).
!__Creating New Tests__
The testing framework is built on top of the existing sample framework, so it is very similar to creating a sample. Tests are created in plugins that the TestContext is able to load dynamically.
!!__General__
Create a class derived from VisualTest, override whichever functions you need (the same FrameListener-style functions used in Samples apply here), and add it to a test plugin.
__Some Things to Note:__
* You will need to specify when you want test screenshot(s) to be taken, with addScreenshotFrame (timing is done by frame to prevent floating point issues).
* Tests must be deterministic, so use the delta (time since last frame) time passed to the frameStarted/frameEnded functions for any timing needs.
* Keep tests simple, the idea is to isolate and test a single feature as completely as possible.
!!__Defining Test Sets__
Test sets (a grouping of tests that will be generated and compared together) are defined as a collection of test plugins in the 'tests.cfg' file. Plugins can belong to more than one test set.
An example configuration of test.cfg is below
{CODE(wrap="1", colors="c#")}
# where the test plugins are located
TestFolder=[Ogre lib dir]
# A set of all visual tests
[VTests]
TestPlugin=PlayPenTests
TestPlugin=VTests
# Only the playpen tests
[Playpen]
TestPlugin=PlayPenTests
{CODE}
!__Tests__
The following are the initial tests being used with the system:
{FANCYTABLE(head="Playpen Tests")}
__Name:__ | __Description:__ | __Image:__
Test Project Sphere | Tests sphere projection. | No image yet.
Camera Set Direction | Tests setting of camera direction. | No image yet.
Manual Blending | Tests manual texture blending. | No image yet.
Manual LOD | Tests manual level of detail setting. | No image yet.
Manual LOD from file | Tests manual level of detail setting. | No image yet.
Morph Animation w/ normals | Tests morph animation. | No image yet.
Morph Animation w/o normals | Tests morph animation. | No image yet.
Pose Animation w/ normals | Tests pose animation. | No image yet.
Pose Animation w/o normals | Tests pose animation. | No image yet.
Particle Effects | Tests simple particle effects. | No image yet.
Stencil Shadows | Tests simple stencil shadows. | No image yet.
Transparency | Tests simple alpha blending. | No image yet.
Texture Effects | Tests simple scrolling/scaling/rotation texture effects. | No image yet.
SceneNodeTracking| Tests simple node tracking functionality. | No image yet.
CubeMapping | Tests basic fixed-function cube mapping. | No image yet.
{FANCYTABLE}
!__Known Issues__
* None! (so far...)
!__Future Improvements__
* More Tests!
* A more robust image comparison algorithm (e.g. something along the lines of [http://pdiff.sourceforge.net/|PerceptualDiff])
!__Additional Links__
* [http://www.ogre3d.org/forums/viewtopic.php?f=13&t=63582|Summer of Code forum thread]
* [http://aras-p.info/blog/2007/07/31/testing-graphics-code/|Unity3d's test framework]