Fix time interval test

[ChrisBr]

and use RubyTime.convertTimeInterval for parsing to raise correct ArgumentError and TypeError exceptions

[kares]

looks good but it introduced one regression failure as a trade for one fixed, see CI's log.
with JI on places were a Ruby Numeric is expected (such as sleep) Java proxy numbers work as well.
... sample failure for reference https://travis-ci.org/jruby/jruby/jobs/351484156#L1555

[ChrisBr]

@kares thanks for the hint.

Unfortunately, the Java::JavaLang::Long gets passed in as a org.jruby.RubyString. So there is no way to distinguish if the RubyString is from the Long or if it was really a String. So seems there is no (easy) way to fix it or do I miss something?

[kares]

oh really? that ain't good than and the "regression" is legit, I will look into a way of pushing this forward ...

[ChrisBr]

@kares tried again, seems I did sth wrong the last time 😢 It get's of course passed in as a ConcreteJavaProxy. Anyway, see my updated patch which passes now both tests.

[kares]

Thanks Chris, poked around and found out the time-interval method is used for elsewhere.
So, I made sure it works from all the places: d19e850 when a Number proxy is being passed in ...

isn't nice but at least works, for now. the whole piece is a new feature in 9.2, and I recall specifically sleep(java_num) was a reported issue that motivated implementing support for Java numbers.

The commit from here is on master as: fd9c772