Recent work on javac diagnostics have underlined the need for better diagnostics; the diagnostics to be improved, esp. those regarding the type-system area of the compiler, have a problem in that they could in principle add too many lines before the erroneous source line is diaplayed; consider the following cases:

Examples:

1) Type-system diagnostic:

Test.java:3: method test in class Foo<T #0> cannot be applied to given types
   required: T #0
   found: T #1
      where T #0,T #1 are typevariables:
          T #0 extends String
              (declared in class Foo)
          T #1 extends Integer
              (declared in method <T>foo(T))
<ERROR LINE HERE>
1 error

2) Method resolution diagnostics

Test.java:6: no suitable method m found for (String,String)
    method: <T>m(T,T) in Test
      (inferred type argument(s) String do not conform to bounds of type variable(s) T)
    method: m(Integer) in Test
<ERROR LINE HERE>
1 error

Despite being an effective way for separating different diagnostics, having source line printed only at the end of the diagnostic output makes the whole diagnostic (i) less readable - because people have to parser the diagnostic contents in order to find the actual erroneous line and (ii) less scalable - because we would like to be able to give the users the flexibility to turn on/off diagnostic features (e.g. where clauses/detailed diagnostics).