[
https://issues.apache.org/jira/browse/RNG-142?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17358196#comment-17358196
]
Gilles Sadowski commented on RNG-142:
-------------------------------------
{quote}Running clirr:check with the same change also reports binary
incompatibility.
{quote}
The return type is part of a function's signature (in the bytecode too).
{quote}Add a note that the call to withUniformRandomProvider returns a
SharedStateContinuousSampler but it can be cast to the original Gaussian
sampler type.
{quote}
I think that we should not advertise the use of casting.
I'd consider that {{SharedStateContinuousSampler}} is an implementation detail
which the caller should not have to be aware of; hence the expectation that the
fluent API returns an object of the same type.
{quote}
{code:java}
SharedStateContinuousSampler g = GaussianSampler.of(algorithm, 0.43, 2.1)
.withUniformRandomProvider(RandomSource.create(RandomSource.JSF_64));
{code}
{quote}
It's fine. Thanks.
Could we perhaps go one step further in encapsulating "implementation details",
and use an "enum" when there are several alternatives for some functionality
(like sampling from a Gaussian distribution)? E.g. (untested):
{code:java}
public class GaussianSampler {
private final ContinuousSampler delegate;
// ...
public enum Variant {
ZIGGURAT,
MARSAGLIA,
BOX_MULLER
}
private GaussianSampler(final NormalizedGaussianSampler n, final double
mean, final double stdev) {
if (mean == 0d && stdev == 1d) { // Will displease "SonarCloud" ;-)
delegate = n;
} else {
delegate = new ContinuousSampler() {
public double sample() {
return mean + n.sample() * stdev;
}
}
}
}
/** Request a specific implementation */
public static GaussianSampler of(Variant v, double mean, double stdev) {
switch (v) {
case ZIGGURAT:
return new GaussianSampler(new ZigguratNormalizedSampler(), mean,
stdev);
// ...
}
}
/** Default implementation */
public static GaussianSampler of(double mean, double stdev) {
return of(Variant.ZIGGURAT, mean, stdev);
}
public double sample() {
return delegate.sample();
}
}
{code}
{{NormalizedGaussianSampler}} thus becomes a "low-level" type (for the library
to ensure its internal consistency) and the public API is just
{{GaussianSampler}}.
> Return type of method "withUniformRandomProvider"
> -------------------------------------------------
>
> Key: RNG-142
> URL: https://issues.apache.org/jira/browse/RNG-142
> Project: Commons RNG
> Issue Type: Improvement
> Components: sampling
> Reporter: Gilles Sadowski
> Priority: Major
>
> Expected usage:
> {code:java}
> NormalizedGaussianSampler n01 =
> ZigguratNormalizedGaussianSampler.of(RandomSource.create(RandomSource.KISS));
> GaussianSampler g =
> GaussianSampler.of(n01.withUniformRandomProvider(RandomSource.create(RandomSource.JSF_64)),
> 0.43, 2.1);
> {code}
> Code doesn't compile: Method {{withUniformRandomProvider}} returns a
> {{SharedStateContinuousSampler}} whereas a {{NormalizedGaussianSampler}} is
> required.
> Am I missing something?
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
