Re: Mauve compatibility intent

[Tom Tromey <tromey at cygnus dot com>]

>>>>> "Dave" == Dave Roberts <dave@droberts.com> writes:

Dave> I was going to try my hand at writing some Mauve test cases. I 
Dave> downloaded a CVS snapshot of Mauve and got it running on Linux. I ran 
Dave> Mauve against the recent Sun JDK 1.2 Linux release candidate and got 
Dave> quite a few errors (like over 1000, with a dominant portion of those 
Dave> against the Character class). This surprised me. Is Mauve regularly 
Dave> tested against the Sun JDKs? Either that or I'm doing something wrong.

Dave> Now, the big overall question is, "What is the compatibility intent of 
Dave> Mauve?" Is it to test compatibility with the Sun JDKs (e.g., full "bug 
Dave> compatibility" with Sun's implementation), or is it to test 
Dave> compatibility with the abstract JDK API specs (e.g., Sun's 
Dave> implementation may show errors because it doesn't meet Sun's own 
Dave> abstract API spec)?

Good questions.

Character is an ugly one, because Sun's spec is ambiguous and in some
cases clearly contradictory (to me anyway -- Sun disagrees).  Also,
things change from release to release (and, for Character, the results
also depend on which version of the Unicode tables you use).

We've tried to address this in part by letting tests be specific to a
particular JDK.

In general I try to write tests according to the "spec", to the extent
that there is one.  E.g., I'll use the Java Class Libraries book (or
online docs) as the normative text (except where it is clearly wrong,
sigh).

However, we can use our configurability code to let us test *both*
spec conformance and bug compatibility, by using different tags for
each.

I haven't been running Mauve against Sun's JDK.

Please do write tests!  We're always eager for more.

Tom