Support union type as expected type for `data.xmldata` module.

[prakanth97]

Description:
$title

Describe your task(s)

Related Issues (optional):

Suggested Labels (optional):

Suggested Assignees (optional):

[prakanth97]

In the xml, the content type is always string, but in the data.xmldata module we allow it to convert to a expected basic type. According to the union type algorithm, we convert them into a JSON and then use data.jsondata, parseAsType API to convert that into a expected type.

In the JSON, string source must be converted to string always. If the expected type is non-string for string literal then it is an error. But we have to allow this for xmldata module hence we have to introduce another option allowConvertFromString for parseAsType API

[prakanth97]

While working on this found a blocker #6659. I will continue this once fix the issue.

[prakanth97]

While working on this found a blocker #6659. I will continue this once fix the issue.

This issue is fixed with ballerina-platform/module-ballerina-data.xmldata#26

[prakanth97]

Ballerina types represents the XSD instead of a particular XML. Hence we have to support the complex union type cases which are possible in the XSD. Currently preparing the doc for this.

[prakanth97]

According to our offline discussion, we concluded that we should support union types. The reason is that the data.xmldata module takes an XML source and converts it to a Ballerina record type. Our concern should be whether the provided source can be converted to the expected type.

Case 1: Union case without namespace

Sample 1:

<service>
    <port name="8000">
        <address location="exampleweb1.com" />
    </port>
    <port name="8008">
        <address location="exampleweb2.com" />
        <id>2345</id>
    </port>
</service>

Ballerina types =>

Option 1:

type Service record {|
    (PortA|PortB)[] port;
|};

type PortA record {|
    @xmldata:Attribute
    string name;
    Address location;
|};

type Address record {|
    string location;
|};

type PortB record {|
    @xmldata:Attribute
    string name;
    Address location;
    int id; 
|};

Option 2:

type Service record {|
    Port[] port;
|};

type Address record {|
    string location;
|};

type Port record {|
    @xmldata:Attribute
    string name;
    Address location;
    int id?;
|};

Given the above sample, both option 1 and option 2 are valid expected types. The module should allow both cases, but for the XSD of the above XML instance, option 2 is the one we have to generate. At the moment, this needs to be considered when Ballerina supports XSD to Ballerina record generation.

Case 2: Union case with namespace

<A xmlns:ns1="example1.com" xmlns:ns2="example2.com">
    <ns1:port>1</ns1:port>
    <ns2:port>1</ns2:port>
</A>

Option 1:

type A record {|
    (B|C)[] port;
|};

@xmldata:Namespace {
    prefix: "ns1",
    uri: "example1.com"
}
type B record {|
    string \#content; 
|};

@xmldata:Namespace {
    prefix: "ns2",
    uri: "example2.com"
}
type C record {|
    string \#content;
|};

Option 2:

type A record {|
    @xmldata:Name {
        value: "port"
    }
    string ns1Port;
    @xmldata:Name {
        value: "port"
    }
    string ns2Port; 
|};

@xmldata:Namespace {
    prefix: "ns1",
    uri: "example1.com"
}
type B record {|
    string \#content; 
|};

@xmldata:Namespace {
    prefix: "ns2",
    uri: "example2.com"
}
type C record {|
    string \#content;
|};

If we take option 1, it allows you to add sequences of ns1:port and ns2:port. Option 2 allows only one of each of those elements. However, both are valid expected types for the provided XML sample. As mentioned above, this is a concern when generating Ballerina records from the XSD and is not part of this module.

Hence I will resume the union type support.

Reference: https://docs.google.com/document/d/1fE6SjAznfFJGMoDzjVTT84qxA5QmiR37C_ulHtKOyv0/edit?usp=sharing