More precise typing of XML elements

[jclark]

We can build on #176 to provide more precise typing of XML elements, by making an xmlelement have type parameters xmlelement<E,A,C>, where

• E is a subtype of string specifying the element name
• A is a subtype of map<string> specifying attributes, and
• C is a subtype of xml<xmlelement|string> specifying type of x.getChildren().strip() (this is because you want comments, PIs, and whitespace to always automatically allowed)

We can then use our type system to write a useful schema for an XML document. Other things that would make this work better include:

• add xml.getChildrenStrip()
• xmlempty that matches an empty xml sequence (for empty elements)
• if ns is declared as a namespace prefix
  • allow ns:x as a const string and type standing for the corresponding expanded name
  • allow ns:foo as the name of a field in a record type (for namespace qualified attributes)
  • allow ns:* as a subtype of string type

This doesn't deal with the ordering or cardinality of child elements. I think ordering would be very hard, but perhaps cardinality is doable. By cardinality I mean saying that this element has between M and N elements named foo (typically M is 0 or 1 and M is 1 or unlimited).

[sanjiva]

If we are doing cardinality we should consider it for records to be able to do better schemas for json too.

See: https://docs.google.com/document/d/1U7tCYaTC9Xx2pKeNrH-Xdw14zPHGm9T2uWTE52SPw10/edit for json schema to Ballerina mapping work.

[jclark]

What is cardinality for records in JSON schema?

[sanjiva]

There are two cardinality aspects:

• arrays: https://docs.google.com/document/d/1U7tCYaTC9Xx2pKeNrH-Xdw14zPHGm9T2uWTE52SPw10/edit#heading=h.wahvnz621312
• records: https://docs.google.com/document/d/1U7tCYaTC9Xx2pKeNrH-Xdw14zPHGm9T2uWTE52SPw10/edit#heading=h.6q78rf64ij9x

[jclark]

So in our terms a min/max constraint on the length of a mapping value? That's pretty funky. Not obvious to me what the use cases for that are.

[sanjiva]

Yes that's what it means. Definitely funky ..

[jclark]

There's a reasonably obvious syntax for us for the array case T[5...17], which I think it much more common.

[jclark]

Note that this issue is not about extending the value space of the XML basic type to handle the PSVI (see issue #249), it's about being able to describe useful subtypes of the our current XML value space.

[jclark]

There is an obvious way to handle ordering and cardinality, which is described in issue #251.