Daniel Quinlan wrote:
No, moving most of check() to where the check function can be placed in
different plugins is a very good move. This allows us to provide
options for how check() works. For example, trying out Bayesian
decision-making, decision tree for faster operation, etc.
Moving some or even "most of" check() to plugins is at least defensible,
but the statement that I was responding to had no such qualifiers.
I think something needs to be done to improve the maintainability and
cross-version-compatibility of the EvalTests.pm module, but I'm not sure
what it is yet.
I think one weakness in the design of eval tests in general is that
their definition is split into two different places, a function
definition in some .pm file and a function call in some .cf file. The
.cf file isn't very amenable to multiple-line chunks of perl code, so
perhaps it would be good if the .pm files directly defined the tests.
Theo is doing a valuable experiment and I think he has
a very good shot of ending up somewhere that makes sense. Let's give
him some time to work it out before rejecting it.
I am trying to raise my concerns in a timely manner, not waiting until
the 11th hour.
