This guide will first explain how to setup the folder structure in your module to allow testing. Afterwards the method of creating and running tests is explained.
Note that it is required to have the R package jaspTools installed.
Your module should include the following structure:
- ModuleName/
An empty folder, until you add tests, at that point it will contain a number of test files as described in Creating a test file.
An empty folder, will be automatically created and filled later, see Adding plot tests.
library(jaspTools)
library(testthat)
jaspTools::runTestsTravis(module = getwd())
The default workflow is
on: [push, pull_request]
name: unit-tests
jobs:
unit-tests:
runs-on: ${{ matrix.config.os }}
name: ${{ matrix.config.os }} (${{ matrix.config.r }})
strategy:
fail-fast: false
matrix:
config:
- {os: windows-latest, r: "3.6.1"}
- {os: macOS-latest, r: "3.6.1"}
env:
R_REMOTES_NO_ERRORS_FROM_WARNINGS: true
R_REMOTES_UPGRADE: never
VDIFFR_RUN_TESTS: true
steps:
- uses: actions/checkout@v2
- uses: r-lib/actions/setup-r@master
with:
r-version: ${{ matrix.config.r }}
- uses: jasp-stats/jasp-actions/setup-test-env@master
- name: Run unit tests
run: source("tests/testthat.R")
shell: Rscript {0}
JAGS can be enabled by setting requiresJAGS: true:
- uses: jasp-stats/jasp-actions/setup-test-env@master
with:
requiresJAGS: true
The analysis tests in modules are based on the R package testthat (for more information about this package see this online book chapter by Hadley Wickam).
The basic idea is that your testthat folder will contain one or more files that each hold a number of unit tests. Each unit test (identified by test_that(...)) is be comprised of one or more expectations (identified by expect_...()). These expectations allow you to verify that certain aspects of the analysis output did not change when compared to earlier versions of your module.
A module should have one test file per analysis. This file will check multiple aspects of the analysis, such as if the error handling works, if the data of the output tables did not change (results of our calculations should almost never change!) and if the plots still look the same.
jaspTools offers some functionality to test tables (expect_equal_tables()) and plots (expect_equal_plots()), but if you wish to test other aspects of the analysis output, you should look into the functionality of testthat.
Before you create any unit tests, we first need to create the file(s) that will store these tests.
The file needs to be located in tests/testthat/ and named according to the scheme test-'analysisname'.R, where 'analysisname' refers to the name of the main R function of an analysis (e.g., test-anova.R).
Every analysis in the module should have a separate test file.
This section explains how to manually create a unit test for a table. For automatic creation of this type of test look here.
We want to make sure that the table data does not change as a result of unintended side effects of code changes. To ensure this, we create a snapshot of the table data and hardcode this in our test file. We then use jaspTools::expect_equal_tables() to check if the snapshot is identical to the table data obtained at a later point in time (and if it's not the test will fail).
expect_equal_tables has the following arguments:
test: The new table data, this should point to the [["data"]] entry of a table in a results list obtained from jaspTools::runAnalysis().
ref: The table data snapshot obtained by applying jaspTools::makeTestTable() to the [["data"]] entry of a table.
label [optional]: If you have multiple expectations in a single unit test, this allows you to identify a specific expectation.
The snapshot of the table data can be created by supplying a table to jaspTools::makeTestTable():
options <- jaspTools::analysisOptions("BinomialTest")
options$variables <- "contBinom"
options$testValue <- 0.5
results <- jaspTools::runAnalysis("BinomialTest", "test.csv", options)
table <- results[["results"]][["binomialTable"]][["data"]]
jaspTools::makeTestTable(table)
Note that the creation of the options shown in the above example can be done in various different ways as described here.
The above returns the output:
list("TRUE", 58, 0, 0.133210619207213, 0.58, 100, "contBinom", "FALSE", 42, 1, 0.133210619207213, 0.42, 100, "contBinom")
And with this we can write the table test:
test_that("Binomial table results match", {
options <- jaspTools::analysisOptions("BinomialTest")
options$variables <- "contBinom"
options$testValue <- 0.5
results <- jaspTools::runAnalysis("BinomialTest", "test.csv", options)
table <- results[["results"]][["binomialTable"]][["data"]]
jaspTools::expect_equal_tables(table,
list("TRUE", 58, 0, 0.133210619207213, 0.58, 100, "contBinom", "FALSE",
42, 1, 0.133210619207213, 0.42, 100, "contBinom"))
})
This section explains how to manually create a unit test for a plot. For automatic creation of this type of test look here.
As with tables, we want to make sure that the plot does not change as a result of unintended side effects of code changes. To ensure this, we create a reference .svg of the plot and store this in tests/testthat/_snaps/ (note that the storage will be done automatically). We then use jaspTools::expect_equal_plots() to check if the reference .svg is identical to the plot obtained at a later point in time (and if it's not the test will fail). This function is a wrapper around vdiffr::expect_doppelganger() (for more information about vdiffr see their GitHub page).
expect_equal_plots has the following arguments:
test: The new plot object, this should point to the [["state"]][["figures"]][[INDEX/PLOTNAME]][["obj"]] entry of a plot in a results list obtained from jaspTools::runAnalysis().
name: The name the plot (will be) stored under (commonly the name of the plot type, e.g., "prior-posterior").
The creation of the reference .svg is an automatic process which cannot be started until the entire unit test has been written. So first we create a unit test for a specific descriptives plot:
test_that("Descriptives plot matches", {
options <- jaspTools::analysisOptions("BinomialTest")
options$variables <- "contBinom"
options$testValue <- 0.5
options$descriptivesPlots <- TRUE
results <- jaspTools::runAnalysis("BinomialTest", "test.csv", options)
plotName <- results[["results"]][["containerPlots"]][["collection"]][["containerPlots_contBinom"]][["collection"]][["containerPlots_contBinom_0"]][["data"]]
testPlot <- results[["state"]][["figures"]][[plotName]][["obj"]]
expect_equal_plots(testPlot, "descriptives")
})
Note that the creation of the options shown in the above example can be done in various different ways as described here.
Notice also that we first obtain the plotname and then use this to index in the state. Plot objects are not stored in results[["results"]] but instead in results[["state"]]. You can also use a numeric index instead of the plotname.
After creating the unit test and storing it in a file we'll want to generate the reference image. To accomplish this you should run all unit tests (through jaspTools::testAll() or jaspTools::testAnalysis()). It does not matter when you create the reference image of a unit test: this could be after you create only a single plot unit test, or after multiple.
Running the unit tests the very first time causes a reference .svg to be created and placed in tests/testthat/_snaps/ (in the above example: tests/testthat/_snaps/BinomialTest/descriptives.svg).
In our example we created a plot test for one analysis so we can be specific in what test file we want to look:
jaspTools::testAnalysis("BinomialTest")
Of course, the above set of steps might be a bit tedious to perform for every single table and plot. You can let jaspTools::runAnalysis() take care of making unit tests by setting makeTests = TRUE. So if we take the example from the manual section:
options <- jaspTools::analysisOptions("BinomialTest")
options$variables <- "contBinom"
options$testValue <- 0.5
options$descriptivesPlots <- TRUE
results <- jaspTools::runAnalysis("BinomialTest", "test.csv", options, makeTests = TRUE)
Which prints the following tests to the R console:
options <- analysisOptions("BinomialTest")
options$variables <- "contBinom"
options$descriptivesPlots <- TRUE
options$testValue <- 0.5
set.seed(1)
results <- runAnalysis("BinomialTest", "test.csv", options)
test_that("Binomial Test table results match", {
table <- results[["results"]][["binomialTable"]][["data"]]
jaspTools::expect_equal_tables(table,
list("TRUE", 58, 0, 0.133210619207213, 0.58, 100, "contBinom", "FALSE",
42, 1, 0.133210619207213, 0.42, 100, "contBinom"))
})
test_that("0 plot matches", {
plotName <- results[["results"]][["containerPlots"]][["collection"]][["containerPlots_contBinom"]][["collection"]][["containerPlots_contBinom_0"]][["data"]]
testPlot <- results[["state"]][["figures"]][[plotName]][["obj"]]
jaspTools::expect_equal_plots(testPlot, "0")
})
test_that("1 plot matches", {
plotName <- results[["results"]][["containerPlots"]][["collection"]][["containerPlots_contBinom"]][["collection"]][["containerPlots_contBinom_1"]][["data"]]
testPlot <- results[["state"]][["figures"]][[plotName]][["obj"]]
jaspTools::expect_equal_plots(testPlot, "1")
})
You will be able to copy-paste the output directly into a test file. Just make sure that each test/expectation title makes some sense (e.g., "Binomial Test table results match" makes sense but "1 plot matches" could use some work).
In the example above we should adjust the two plot tests, because "0" and "1" are not very informative:
test_that("Descriptives plot contBinom-level 0 matches", {
plotName <- results[["results"]][["containerPlots"]][["collection"]][["containerPlots_contBinom"]][["collection"]][["containerPlots_contBinom_0"]][["data"]]
testPlot <- results[["state"]][["figures"]][[plotName]][["obj"]]
jaspTools::expect_equal_plots(testPlot, "descriptives-0")
})
test_that("Descriptives plot contBinom-level 1 matches", {
plotName <- results[["results"]][["containerPlots"]][["collection"]][["containerPlots_contBinom"]][["collection"]][["containerPlots_contBinom_1"]][["data"]]
testPlot <- results[["state"]][["figures"]][[plotName]][["obj"]]
jaspTools::expect_equal_plots(testPlot, "descriptives-1")
})
Note that it is not recommended to add multiple tests for the same element if the tests do not test distinct aspects of the element.
In our example, we could get rid of one of the two plots, in which case we end up with:
options <- analysisOptions("BinomialTest")
options$variables <- "contBinom"
options$descriptivesPlots <- TRUE
options$testValue <- 0.5
set.seed(1)
results <- runAnalysis("BinomialTest", "test.csv", options)
test_that("Binomial Test table results match", {
table <- results[["results"]][["binomialTable"]][["data"]]
jaspTools::expect_equal_tables(table,
list("TRUE", 58, 0, 0.133210619207213, 0.58, 100, "contBinom", "FALSE",
42, 1, 0.133210619207213, 0.42, 100, "contBinom"))
})
test_that("Descriptives plot matches", {
plotName <- results[["results"]][["containerPlots"]][["collection"]][["containerPlots_contBinom"]][["collection"]][["containerPlots_contBinom_0"]][["data"]]
testPlot <- results[["state"]][["figures"]][[plotName]][["obj"]]
jaspTools::expect_equal_plots(testPlot, "descriptives")
})
You can do this for how ever many jaspTools::runAnalysis()'s to get a good coverage. If you use a .jasp file with multiple analysis calls as input to jaspTools::analysisOptions() then you can quickly create many tests:
options <- jaspTools::analysisOptions("path/to/file.jasp") # there are multiple analyses so options is a list of lists
jaspTools::runAnalysis("BinomialTest", "test.csv", options[[1]], makeTests = TRUE)
jaspTools::runAnalysis("BinomialTest", "test.csv", options[[2]], makeTests = TRUE)
As explained earlier (see Adding plot tests), to validate the new plots you should run:
jaspTools::testAll() # or jaspTools::testAnalysis()
If you have .hasErrors() calls within your analysis, then it's a good idea to verify that they all work correctly.
You can do this by creating a unit test that checks if your analysis exits when you supply it incompatible data (e.g., data that contains infinities).
For these types of checks we use expectations provided by testthat:
test_that("Analysis handles errors", {
options <- jaspTools::analysisOptions("RegressionLinear")
options$dependent <- "debInf"
options$covariates <- "contGamma"
options$modelTerms <- list(list(components="contGamma", isNuisance=FALSE))
results <- jaspTools::runAnalysis("RegressionLinear", "test.csv", options)
expect_identical(results[["status"]], "validationError", label = "Inf dependent check")
options$dependent <- "contNormal"
options$covariates <- "debInf"
options$modelTerms <- list(list(components="debInf", isNuisance=FALSE))
results <- jaspTools::runAnalysis("RegressionLinear", "test.csv", options)
expect_identical(results[["status"]], "validationError", label = "Inf covariate check")
})
Maybe your analysis does not exit when it has invalid data, but instead it logs this in a footnote
test_that("Analysis handles errors", {
options <- jaspTools::analysisOptions("TTestIndependentSamples")
options$variables <- "debInf"
options$groupingVariable <- "contBinom"
results <- jaspTools::runAnalysis("TTestIndependentSamples", "test.csv", options)
notes <- unlist(results[["results"]][["ttest"]][["footnotes"]])
expect_true(any(grepl("infinity", notes, ignore.case=TRUE)), label = "Inf check")
options$variables <- "debSame"
options$groupingVariable <- "contBinom"
results <- jaspTools::runAnalysis("TTestIndependentSamples", "test.csv", options)
notes <- unlist(results[["results"]][["ttest"]][["footnotes"]])
expect_true(any(grepl("variance", notes, ignore.case=TRUE)), label = "No variance check")
})
You can explore the results object returned by jaspTools::runAnalysis() and test any aspect of it.
It's even possible to create your own expectations.
jaspTools comes bundled with all the .csv datasets available in the data library of JASP.
Additionally, it has a copy of debug.csv named test.csv. It's recommended to use test.csv as it is guaranteed the data in this file will never change.
If you would like to use something else still, that is also possible -- simply add it inside tests/testthat.
Say you have tests/testthat/someData.csv, you will be able to use this like:
jaspTools::runAnalysis("TTestIndependentSamples", "someData.csv", options)
Just ensure that your working directory is set to tests/testthat when making the test.
To run all tests in a module, type
jaspTools::testAll()
Note that if you include several modules in jaspTools::setPkgOption("module.dirs", ...), then the tests of all of these modules will be run sequentially.
Any test that fails is shown in the R console.
Warnings may be ignored, but should be minimized.
It is also possible to test a specific analysis, as running all unit tests in a module may take some time
jaspTools::testAnalysis("Anova")
If a test shows up as failed, you should verify why this is and fix it before making a pull request.
If you made a legitimate change that the test does not cover, then the unit test should be updated.
You can locate the offending test inside tests/testthat; your R console will show the filename and line number -- e.g., test-jagsmodule.R:56: failure:.
If testthat shows that a failure occurred then that means the unit test ran okay, but the results do not match. If it is anything but a plot (e.g., a table) the R console will notify you of the exact problem. You should determine why this change occurred and if it is desired. If it is desired, then you can update the unit test.
To view changes in plot tests you should run jaspTools::manageTestPlots() (e.g., manageTestPlots("jagsModule")). If the change is legitimate you can validate the failing plot in the Shiny application; the reference .svg in the tests/testthat/_snaps/ folder will be updated automatically.
If testthat shows that an error occurred then that means the unit test could not run properly; it terminated too soon. To figure out the cause of termination it is often useful to run the unit test code directly in RStudio. This means going to the test file, locating the unit test and then running everything between jaspTools::analysisOptions(...) and jaspTools::runAnalysis(...) so see what the output of runAnalysis() is.