Torsten
a3effe29f6
true, false and nll were not handled correctly as returns returns assume int, and reduce. The solution (hack?) is to add fake numbers into true,false and nil that get returned from from the arm syscall (and it works!, which means i can fix other tests now) |
||
|---|---|---|
| .. | ||
| source | ||
| helper.rb | ||
| README.md | ||
| test_arm.rb | ||
| test_assign.rb | ||
| test_calling.rb | ||
| test_class.rb | ||
| test_conditional.rb | ||
| test_int_cmp.rb | ||
| test_int_math.rb | ||
| test_interpreted.rb | ||
| test_new.rb | ||
| test_while.rb | ||
Mains testing
Test methods by their output and exit codes (return, since it is the main).
There are only two tests here (plus one, see below), one for interpreter, one for arm. Both run the same tests. The actual ruby code that is run is in the source dir. Test methods are generated, one for each source file.
Files
File names follow [order,name,stdout,exitcode] joined by _ pattern. Stdout may be left blank, but exit code must be supplied.
The order number is some number giving the difficulty of the test, higher is more. The first digit represents how many external methods the code relies on, the second is some general indicator, ie recursive is more difficult than not, syscalls more than normal calls, if or while more than nothing etc.
Arm
Obviously the arm tests need an arm platform. This may be defined by ARM_HOST, eg for simulated ARM_HOST=localhost
Also port and user may be specified with ARM_PORT and ARM_USER , they default to 2222 and pi if left blank. SSH keys must be set up so no passwords are required (and the users private key may not be password protected)
Developing
Since the Framework always runs all tests, it is a little cumbersome for developing a single new test. Since all get run and it is slow.
To develop the next test, one can edit test_new.rb . Once it runs on the interpreter, move the changes to a source file and revert test_new changes.