From Free Pascal wiki
Revision as of 16:29, 17 May 2018 by Pdenommee (talk | contribs) (Any field in a subclass of TTestCase may get cleared at the beginning of a new test)

English (en) français (fr)


fpcunit is a unit testing framework a la DUnit/JUnit/SUnit. This allows you to quickly write a set of test for a (logical) unit of code (not necessarily the same as a Pascal unit, though often it is).

Development methodologies like Test Driven Design use this to make sure you code your expectations/specifications in your unit tests first, then write your main code, then run your tests and improve the code until all tests pass.

Not only does fpcunit allow you to visually inspect test runs, you can also collect the results systematically (using the XML output), and use that to compare versions for e.g. regression errors (i.e. you run your regresssion tests using the unit test output).

Screenshot of the GUI test runner:


The image shows that out of 10 tests run, 6 tests failed. The EAssertionFailure exceptions indicate the test assertions (see below) were not met - i.e. the test failed. The associated messges indicate the result the test expected and the actual result achieved.

Use in FPC/Lazarus

FPCUnit tests are used in the FPC database test framework: Databases#Running_FPC_database_tests

There are also tests for the FPC compiler/core packages, but these presumably predate fpcunit and use a simpler approach.


It's easiest to use Lazarus to set up a new test project for you. Below are some descriptions of which procedures/methods to use for what purpose.


This procedure is present in all FPCUnit tests. It sets up the test environment before each test is run - in other words not only before and after the complete test suite is run, but for every test. You can use this to e.g. fill a database with test data.


This procedure is present in all FPCUnit tests and is the reverse of Setup. It cleans up the test environment after each test is run. You can use this to e.g. clear test data from a database.

Test decorator: OneTimeSetup and OneTimeTearDown

The Setup and Teardown procedures mentioned above are run once per test. You can also run Setup and Teardown procedures once per instance/execution of your test run.

To do this, use OneTimeSetup and OneTimeTearDown and inherit from the TTestSetup "test decorator" and register it, e.g.:

  TDBBasicsTestSetup = class(TTestSetup)
      procedure OneTimeSetup; override;
      procedure OneTimeTearDown; override;
// make sure you register your test along with the decorator so it knows to run the setup/teardowns
  RegisterTestDecorator(TDBBasicsTestSetup, TTestDBBasics);


You write your own tests as published procedures in the test class (private, protected or public will not work). You can use AssertEquals etc to specify what should be tested, and give a suitable message when the test fails.

If you want to fail a test, you can e.g. use

if 5=0 then //ridiculous example but you understand what I mean. You can use other Assert* procedures to test equality much easier etc.
  AssertTrue('This part of the code should never have been reached in this test.',false);

Note: there must be a better way of doing this...

If the test fails, an EAssertionFailedError will be raised with the message you specify in Assert*. This way, you can add a series of subtests and tell which subtest failed. Note: the test runner will stop at the first assertion failure, so subsequent subtests will not be performed. If you do want to always test everything, split out these subtests in separate test procedures.

Instead of the Assert* procedures, you can also use the DUnit compatible Check* procedures (e.g. CheckEquals) which give more descriptive error messages in the test results: they include expected and actual values.

The order in which the tests are run are the order in which they appear in the test class definition.

Example test

  Ttestexport1 = class(Ttestcase)
    procedure TestOutput;
procedure Ttestexport1.TestOutput;

  if FileExists(OutputFilename) then DeleteFile(OutputFileName);
  // Fill test data
  TestDataset.FieldByName('ID').AsInteger := 1;
  // Data with quotes
  TestDataset.FieldByName('NAME').AsString := 'J"T"';
  TestDataset.FieldByName('BIRTHDAY').AsDateTime := ScanDateTime('yyyymmdd', '19761231', 1);

  AssertEquals('Number of records in test dataset', 1, TestDataset.RecordCount);

Test hierarchy

In simple cases, you (or Lazarus) would register all your test cases with calls like:

  RegisterTest(Ttestexport1); //you pass the class name to register it for running

However, you can also create multiple layers to group your test cases if your project gets big:

  RegisterTest('WorldDominationApp.ExportCheesePlan',Ttestexport1); //The levels are separated by periods 
  RegisterTest('WorldDominationApp.Obsolete',TtestPinkysBrain1); //another category

Automatic modification to the fields

If you add any field to a subclass of TTestCase, you should be aware that those fields will be reseted to their default values before the beginning of a test. I do not consider this to be a bug because tests results should be independent: the result of a test should not depend on the result of any other test. I have reproduced this behaviour in both Windows and Linux Mint XFCE.

If you really need a reliable value for a variable, using class var instead of fields solves the problem.

While performing further testing of this unit, I found that, in the code below, if I replace fx:double; by class var fx:double; the test no longer fails! Declaring one variable a class variable has cleared the problem. This makes me believe that this is a bug.

Usally, bugs are not written to the documentation, but this one has cost us many hours of lost time during unit testing. Furthermore, before reporting any bug, the expected behaviour must be clearly established. This wiki page must describe the expected behaviour. Is the bug is that the fields are assigned to their default values of 0 or is the bug is that when class var is added, they are no longer assigned to 0? According to this reference https://sergworks.wordpress.com/2012/08/31/introduction-to-unit-testing-with-lazarus/ , the bug would be that they are cleared to their default values.

unit TestCase1;

{$mode objfpc}{$H+}


  Classes, SysUtils, fpcunit, testutils, testregistry, Dialogs;


  TTestCase1= class(TTestCase)
    procedure SetUp; override;
    procedure TearDown; override;
    procedure One;
    procedure Two;


procedure TTestCase1.One;
  IF fi <> 3 THEN
     Fail('Must be 3.');
procedure TTestCase1.Two;

  IF fi <> 3 THEN    //Will fail, fi =0, it has been automatically cleared even if test 1 has ran before
     Fail('Must be 3.');

procedure TTestCase1.SetUp;


procedure TTestCase1.TearDown;





The console test runner can output in XML (either original FPCUnit format, or a more advanced DUnit2-like format if you use the xmltestreport unit), plain text and latex (e.g. usable for PDF export) formats. The GUI test runner outputs to XML if needed (using the same xmltestreport XML format).

Customising output

You can use your own "listener" that listens to the test results and outputs the test data in whatever way you want. Create a T*Listener that implements the ITestListener interface. It's only 5 required methods to implement.

In your test runner application (e.g. a copy of fpctestconsole), add a listener object; register that test listener with the testing framework (e.g. your console test runner) via the TestResult.AddListener() call, and it will be fed test results as they happen.


An example of a custom listener is the database output writer available at https://bitbucket.org/reiniero/testdbwriter. This writer will save all test results to a database, which is optimized for receiving large amounts of test results (handy for using on CI server like Jenkins or for importing/consolidating test results). The mentioned repository contains an example that runs the db test framework test results to (another) database.

To do: adapt this; use the new xml unit

An example of an available extra listener is TXMLResultsWriter in the xmlreporter unit in <fpc>\packages\fcl-fpcunit\src\xmlreporter.pas.

todo: actually, dbtestframework seems to use the old xml output method... An example of an adapted test runner that uses an extra listener can be found in <fpc>\packages\fcl-db\tests\dbtestframework.pas, which contains this code to output to custom listeners (an XML writer and a digest writer that stuffs the output in a .tar archive, handy to process remotely):

 ...the rest of the units needed for test runners...
 // the units with TXMLResultsWriter and TDigestResultsWriter
Procedure LegacyOutput;

  FXMLResultsWriter: TXMLResultsWriter;
  FDigestResultsWriter: TDigestResultsWriter;
  testResult: TTestResult;

  testResult := TTestResult.Create;
  FXMLResultsWriter := TXMLResultsWriter.Create;
  FDigestResultsWriter := TDigestResultsWriter.Create(nil);
    // Set some properties specific for this results writer:
    //WriteHeader is specific for this listener; it writes the header to an XML file
    //notice that it is not called for FDigestResultsWriter
    // This performs the actual test run, and the output will be processed by the listeners:
    // Likewise WriteResult is specific for this listener; it writes 


  • DUnit2 - a huge improvement over the original DUnit. Originally written for Delphi only, and which is used by the huge test suite of the tiOPF framework.
  • FPTest - a fork of DUnit2, and which is tuned specifically for use with the Free Pascal Compiler.


Lazarus has the consoletestrunner and GUI test runner units, which can be installed by installing the FPCUnitTestRunner package. This will help you create and run your unit tests using a GUI (or console, if you want to).

The consoletestrunner is compatible with FPC so you don't need Lazarus to compile it. The Lazarus version is slightly different to the one in FPC (e.g. use of UTF8 output etc).

The GUI runner is easier to use.

In the GUI runner, if you want to run all tests, currently you first need to click on a test element before the Run all tests button is activated.

GDB bug/feature

Note (September 2012): a bug/undocumented feature in the debugger used by Lazarus/FPC (gdb) means that passing --all as run parameters has no effect. Passing this parameter can be useful when debugging console fpcunit test runners has no effect. Workaround: use -a. See bug [1]

See also