The compiler verification framework I mentioned in last week’s post is currently under development and is due for completion by the end of next week. Of course, compiler testing never really reaches a completed state; new features can be added or old ones may require verification for a new use-case. So what I mean by completed is that by next week we will have included all the test cases we originally intended to include, which were decided upon after some thought and reasonably made assumptions. In the future, it is very likely that new test cases will be added to the framework. Our intended test cases will cover every expected (and many unexpected) use cases of the IL ops.
The testing framework verifies the correctness of the compiler using behavioural testing, which essentially tests whether the IL to ASM conversions were correct. It is a behavioural testing framework because it tests the functionality of the compiler by testing the behaviour of the output and thus whether the compiler compiled correctly. It is also cross-architecture/cross-platform because it can be used for testing both the x86 and the MIPS architecture libraries (and any others we may add in the future).
The test kernel is called FlingOops™ and consists of a wrapper framework (with variations for different target architectures) and a long list of method calls, where each method contains a particular test relating to an operation as well as a message to the console stating whether the test case passed. Particular attention is paid to testing signed operations and the ones that require the handling of 64-bit values. To verify that signed and 64-bit values are calculated correctly, depending on the operation tested, we need to make sure that the carry- and borrow-bits are handled correctly as well as overflow happens as expected. There aren’t special carry and borrow flags on MIPS, so these have to be implemented by the compiler and therefore it is crucial to test the correctness of these implementations. The testing is going well and it is great to see that so far, all the test cases passed with the exception of a couple of small errors which were in the x86 target library. Thankfully these issues were fixed quickly as we were able draw from the knowledge and experience gained form working with MIPS viagra 100 mg 4 comprim. On the plus side, the detection of bugs in the x86 build demonstrated that our testing works!
For the last week
For the last week of my internship, our plan is to finalise the testing framework. I will post another article next week to confirm the completion of testing and to review my experience working for FlingOS™.
Thanks for reading! See you next week,