fn:compare: Support for arbitrary atomic types

[ChristianGruen]

Inspired by #866:

We should extend fn:compare to support arbitrary atomic types. The comparison rules…

• would be unchanged for strings,
• would rely on fn:numeric-compare for numbers, and
• would rely on the existing op: functions for the remaining types.

For example, the rule for dates would be:

• 0 is returned if op:date-equal(A, B) is true,
• -1 is returned if op:date-less-than(A, B) is true,
• 1 is returned otherwise.

Some types will be rejected (xs:duration, xs:QName, xs:NOTATION, Gregorian types).

In addition, I would vote for making fn:numeric-compare and fn:atomic-equal private. I don’t see a benefit to expose them; I rather expect people to be confused.

[michaelhkay]

We chose to expose fn:atomic-equal because it was useful as a callback to functions like fn:for-each-pair() and subsequence-starts-with() and because the alternatives, like count(distinct-values($a, $b)) = 1 are horrible.

[ChristianGruen]

We chose to expose fn:atomic-equal because it was useful as a callback to functions like fn:for-each-pair() and subsequence-starts-with() and because the alternatives, like count(distinct-values($a, $b)) = 1 are horrible.

I’d probably have written op('=') or fn:deep-equal#2.

But maybe it's good to motivate people to learn more about the internals of the language this way, and to give experts direct access to it.

[ChristianGruen]

Accepted and merged (#909).