Hi,
I have some good news, I added a test named "coverage" that reads Defaults.commands and tests every command it finds in it (1012 tests for the moment). For each command the tests checks if coverage-data/COMMAND.scm exists, and execute it if it exists. Else it simply launch denemo with '-a "(d-COMMAND)(d-Quit)".
This is still a very weak test, since there is no argument passed to the commands, and since that results are not checked. But at least it can check for some segfaults. This test double our test coverage (we were around 10% now, we are at 20%). The coverage test is very long.
To keep things clean, a test should be created every time a command is created (in tests/coverage-data/COMMAND.scm). Tests should be as simple as possible, and provoke failure when the result obtained is not the one expected.
More generally, each implemented feature and each fixed bug should have its own test.
Next step is to also test builtin functions declared in scheme-identifiers.c and find a way to run GUI tests, and add some unit tests for often used functions.