Saxon

Yes, I too observed this when investigating

https://stackoverflow.com/questions/48823316/xslt-condition-and-decimal-values/48823982#48823982

The query works correctly when running Saxon-EE with a license file. In Saxon-PE and EE, we use the ICU library for handling collations. In HE mode, we fall back to what's available from the JDK (unless you specify fallback=no). And I don't think the JDK collation machinery has any way of implementing numeric=yes.

However, with some investigation, it appears to be possible to combine Saxon's AlphanumericCollator class (which is used for Saxon collation URIs) with a base RuleBasedCollator supplied by the JDK for sorting the alphabetic parts of the sort key, so I shall add this capability.

I run Saxon 9.8 EE from the command line with the @-t@ option and it indicates me that it uses a license, nevertheless I get the results mentioned, namely @compare("Chap2", "Chap10", "http://www.w3.org/2013/collation/UCA?lang=en;numeric=yes")@ evaluating to @1@ and @sort(('chap12', 'chap2', 'chap101', 'chap20'), 'http://www.w3.org/2013/collation/UCA?lang=en;numeric=yes')@ to @chap101 chap12 chap2 chap20@.

Does the command line code path somehow not pick up the ICU library or why do I get the wrong results?

I'm somewhat mystified by this because I'm getting the incorrect answer when I run in the terminal:

Saxon-EE 9.8.0.7J from Saxonica
Java version 1.7.0_55
Using license serial number K006480
Analyzing query from {compare('Chap2', 'Chap10', 'http://www.w3.org/2013/collation/UCA?lang=en;numeric=yes')}
Analysis time: 351.96 milliseconds
1
Execution time: 25.566ms
Memory used: 44695432

and the correct answer when I run under IntelliJ:

Saxon-EE 9.8.0.9J from Saxonica
Java version 1.7.0_55
Using license serial number K006480
Analyzing query from {compare('Chap2', 'Chap10', 'http://www.w3.org/2013/collation/UCA?lang=en;numeric=yes')}
Analysis time: 490.972 milliseconds
Processing file:/Users/mike/Desktop/temp/test.xml
Source document ignored - query does not access the context item
-1
Execution time: 18.012ms

I know that's a different Saxon version but I've been experimenting with variations. I've also been playing with different Java versions.

A possibility which I will now investigate is that there's something materially different between the built JAR files and the version of the code that we're running in the IDE.

Simple and infuriating. It works correctly if you put icu4j-59_1.jar on the classpath.