This document shows how to use the DBItest package when implementing a new DBI backend or when applying it to an existing backend. The DBItest package provides a large collection of automated tests.

Testing a new backend

The test cases in the DBItest package are structured very similarly to the sections in the “backend” vignette:

vignette("backend", package = "DBI")

Like the “backend” vignette, this vignette assumes that you are implementing the RKazam package that has a Kazam() function that creates a new DBIDriver instance for connecting to a “Kazam” database.

You can add the tests in the DBItest package incrementally, as you proceed with implementing the various parts of the DBI. The DBItest package builds upon the testthat package. To enable it, run the following in your package directory (after installing or updating devtools):

This creates, among others, a file test-DBItest.R in the tests/testthat directory. Replace its entire contents by the following:

Now test your package with devtools::test(). If you followed at least the “Getting started” section of the DBI “backend” vignette, all tests should succeed.

By adding the corresponding test function to your tests/test-DBItest.R file before implementing a section, you get immediate feedback which functionality of this section still needs to be implemented by running devtools::test() again. Therefore, proceed by appending the following to tests/test-DBItest.R, to include a test case for the forthcoming section:

Again, all tests should succeed when you are done with the “Driver” section. Add the call to the next tester function, implement the following section until all tests succeed, and so forth.

In this scenario, you are usually interested only in the first error the test suite finds. The StopReporter of testthat is most helpful here, activate it by passing reporter = "stop" to devtools::test(). Alternatively, call the relevant DBItest::test_() function directly.

The tests are documented with the corresponding functions: For instance, ?test_driver shows a coarse description of all tests for the “Driver” test case. Test failures will include the name of the test that is failing; in this case, investigating the documentation or the source code of the DBItest package will usually lead to the cause of the error.

Not all tests can be satisfied: For example, there is one test that tests that logical variables survive a write-read roundtrip to the database, whereas another test tests that logical variables are converted to integer in such a case. Tests can be skipped by adding regular expressions for the tests to skip as character vector to the call, as in the following1:

Some other reasons to skip tests are: - your database does not support a feature - you want to postpone or avoid the implementation of a feature - the test takes too long to run.

Testing an existing backend

For an existing backends, simply enabling all tests may be the quickest way to get started. Run the following in your package directory (after installing or updating devtools):

This creates, among others, a file test-DBItest.R in the tests/testthat directory. Replace its entire contents by the following:

The notes about “Kazam” and skipping tests from the previous section apply here as well. The test_all() function simply calls all test cases.

External testing

DBItest is currently geared towards usage as part of a package’s test suite. With some effort it is possible to test a database backend against a custom database. This can help verify that your database installation gives expected results when accessed with DBI with specific connection arguments.

The example below shows how to run tests with the RSQLite backend.

Preparation

First, we need to define a test context. It contains:

  • a connector that describes how to establish the database connection, see ?DBI::`DBIConnector-class` for details,
  • tweaks, see ?tweaks,
  • tests skipped by default, as a character vector.

Database backends that use DBItest for testing usually have a file test/testthat/helper-DBItest.R or test/testthat/test-DBItest.R where a call to make_context() can be found. The help for make_context() already contains an example that works for RSQLite. Adapt it to your needs.

The make_context() function must be called before any tests can run.

Testing

Use test_all() to run all tests, and test_some() to run a specific test that failed previously. The test_* functions need to be run with a testthat reporter to avoid stopping at the first error or warning. For interactive use, the “progress” reporter gives good results. In the example below, the “location” and “stop” reporters are combined. Review ?testthat::Reporter for a list of reporters.

testthat::with_reporter(
  c("location", "stop"),
  DBItest::test_some("get_query_atomic")
)
## Start test: DBItest: Result: get_query_atomic
##   spec-result-get-query.R#17:1 [success]
##   spec-result-get-query.R#17:1 [success]
##   spec-result-get-query.R#17:1 [success]
##   spec-result-get-query.R#17:1 [success]
##   spec-result-get-query.R#17:1 [success]
##   spec-result-get-query.R#17:1 [success]
##   spec-result-get-query.R#17:1 [success]
## End test: DBItest: Result: get_query_atomic

Debugging failing tests

DBItest relies heavily on metaprogramming. Unfortunately, this means that a failing test may give no indication of the reason for the failure. One way to scout for the reason of the problem is to review the sources of DBItest and relate the test name (that is printed with each failure) with the human-readable specification embedded with the test code.

Alternatively, the new experimental dblog package helps by printing the DBI code that is executed as the tests are run. For this, use a driver constructed by dblog::dblog() to display DBI code interspersed with testthat output.

## drv1 <- RSQLite::SQLite()
## Start test: DBItest: Result: get_query_atomic
## conn1 <- dbConnect(drv1, dbname = "/tmp/RtmpQBpbDe/DBItest4517733ac8a7.sqlite")
## dbGetQuery(conn1, "SELECT 1.5 AS a")
## ##     a
## ## 1 1.5
##   spec-result-get-query.R#17:1 [success]
##   spec-result-get-query.R#17:1 [success]
##   spec-result-get-query.R#17:1 [success]
##   spec-result-get-query.R#17:1 [success]
##   spec-result-get-query.R#17:1 [success]
##   spec-result-get-query.R#17:1 [success]
## dbDisconnect(conn1)
##   spec-result-get-query.R#17:1 [success]
## End test: DBItest: Result: get_query_atomic

  1. The terminating NULL allows appending new lines to the end by copy-pasting an existing line, without having to take care of the terminating comma.