[bugfix] spec compliant random-number-generator

[line-o]

Description:

• add tests for fn:random-number-generator
• xs:anyAtomic allowed as a seed value
• calls to fn:random-number-generator()?next() are deterministic

Reference:

fixes #3071

Type of tests:

xqsuite

[duncdrum]

[ERROR]   java:org.exist.xquery.XPathException exerr:ERROR cannot convert value of type xs:dateTime to Java object of type long [at line 32, column 31, source: /home/travis/build/eXist-db/exist/exist-core/src/test/xquery/xquery3/fnRandomNumberGenerator.xql]
1481In function:
1482	fn-rng:seed-dateTime() [17:5:/home/travis/build/eXist-db/exist/exist-core/src/test/xquery/xquery3/fnRandomNumberGenerator.xql]
1483	test:apply(function(*), item()*) [560:9:/home/travis/build/eXist-db/exist/exist-core/target/classes/org/exist/xquery/lib/xqsuite/xqsuite.xql]
1484	test:apply(function(*), element(), item()*) [467:9:/home/travis/build/eXist-db/exist/exist-core/target/classes/org/exist/xquery/lib/xqsuite/xqsuite.xql]
1485	test:call-test(function(*), element(), element()*) [279:32:/home/travis/build/eXist-db/exist/exist-core/target/classes/org/exist/xquery/lib/xqsuite/xqsuite.xql]. cannot convert value of type xs:dateTime to Java object of type long [at line 32, column 31]
1486[ERROR]   err:FORG0001 Invalid value for cast/constructor. can not convert 'sample seed' to xs:long [at line 20, column 27]

some type casting problem it seems

[line-o]

Yes, this is will fail until the function is implemented according to spec.

[duncdrum]

To merge this now i d suggest expecting the error and linking to #3071 in an inline comment, this way we would track changes in behaviour moving forward.

Unless you have a fix to the underlying java implementation ready to go, so we could add it to this PR ?

[line-o]

I would rather leave this open until the the fix is in place, @duncdrum .

[adamretter]

@line-o happy to take a look at this with you when we meet during XML Prague 2020.

[line-o]

@adamretter I just checked the spread of the random numbers (next() called 10000 times). The results looks OK.
The delta of each tenth is below one hundred in each segment (-100, 100).

[adamretter]

@line-o you should find that they are better than java.util.Random

[line-o]

I would would like to see this merged as it is.

[adamretter]

@line-o sounds good, but as you and I both worked on it - you will have to find someone else to review and merge (if they are happy).

The rule is - "you never merge your own PRs"

[line-o]

I know :)

[joewiz]

I re-ran the one failing CI test, and it now all passes.

[joewiz]

I’m curious to know, are we passing more of the qt3 tests for this function after the bug fix?

[adamretter]

@joewiz with the XQTS runner after modifying build.sbt to reflect the correct eXist-db version, running:

$ java -jar exist-xqts-runner-assembly-1.0.0.jar --xqts-version HEAD --test-set fn-random-number-generator

Yields the results:

1. 45.45% for eXist-db 5.2.0
2. 56.82% for eXist-db 5.3.0-SNAPSHOT

Not great I know! But... in this PR I only focused on fixing the two bugs that @line-o reported to me.

[line-o]

Well these two amounted for more than 10%, I think it was worth the effort.
@adamretter could you publish the failing ones here?

[adamretter]

@line-o No, you can run the XQTS.

I think you will particularly groan at the fn-random-number-generator-30 failure though, as it has implications on your xBow work too!

[adamretter]

See also - #3249