99% – is it enough or not?
Today is a great day. I will try to explain why. As I mentioned in my intro-post our team is creating several different TCK’s. The area that I work on is so-called LANG – I develop tests for Java Language. Long ago, more than 2 years from now, we started to work on . We had to solve many problems which often occur during spec change (I promise to write more about that). Our team is finishing JCK 6a, lang tests is part of this JCK. Today I run the coverage scripts and we can finally say that we have 99% assertion coverage for JLS 3. To be more precise we have 99.4%. It means that we wrote tests for 99% of sentances in JLS 3 that we had marked as potentially testable. Isn’t it cool? I bet it is!
The work is certainly not over yet and will not be so - there are many reasons why more tests are needed :
- depth coverage improvement – more tests for several assertions are needed;
- there are sentances that are testable, but for several reasons we hadn’t marked them as potentially testable;
- there will be JLS 4 soon, we should start working on it as soon as possible.
Different people might have opposite answers for a question in a title. Most would say "Yes, of course". Indeed 99% is almost 100%. And what is 100% – it is a perfection. 99% looks great, and it is great. But we must understand what this number stands for, and what can be improved. My opinion is "yes, it is great, colossal, tremendous; but no, it’s not enough, I want more, even more than 100%", that’s why I plan to create a script for depth coverage calculation.
Thanks to all SUN developers who did JCK-Lang work, thanks to people who helped (especially to compiler team) and certainly great thanks to all developers who use Java 
Java world became even more compatible and safer!
assertion, Java, Java Language, JLS, language, spec, specification, Sun, TCK, tests