Lists Home |
Date Index |
Jonathan Robie wrote:
>They are not reference implementations, they are sample implementations.
>>>Are they tied normatively to the spec or informatively?
>They are tied to the spec only as an existence proof, showing that the spec
>can be implemented.
The weakness with this approach appears to be completeness. Someone can
come forward with a partial implementation that covers most of the
obvious stuff while ignoring the more complex issues, and unless those
involved take a very hard look they're unlikely to realize that the
implementation just papers over the cracks in the specification. At
least one case comes to mind where this appears to have happened (WXS).
The JCP process used by Sun at least attempts to deal with this through
the compatibility test kits required for each standard. This seems like
a solid approach - if your implementation passes the test kit it really
*is* an implementation of the standard, if it doesn't it's not. The
standard authors should be required to certify that the test kit truly
does represent a comprehensive test of the standard as a necessary
precondition to the specification being made final.