Introduction

The XML Parser uses XPath to extract field from a complex XML Event. This is generally used in conjunction with the XML Source Reader. The XML Source Reader will provide individual events to the XML Parser and the XML Parser is responsible for extracting fields from the event and mapping them to output schema. 

A simple example - assume you have a XML Event that looks as follows:

<employee>
  <name>
   <first>Joltie</first>
   <last>Root</last>
  </name>
  <address>
   <street> 180, Mars Ave </street>
   <city> Marsville </city>
   <state> Marscity </state>
   <country> M.A.R.S </country>
   <coordinates>
     <lat>89</lat>
     <long>117</long>
   </coordinate>
  </address>
  <dob>
    <day>1</day>
    <month>Jan</month>
    <year>2177</year>
  </dob>
</employee>

User wants to extract following fields from the XML event.

• first
• last
• lat
• long
• dob year

User uses the following XPath to extract those fields

/employee/name/first
/employee/name/last
/employee/address/coordinates/lat
/employee/address/coordinates/long
/employee/dob/year

Use-case

• User should be able to specify the input field that should be considered as source of XML event or record.
• User is able to specify XML encoding (default is UTF-8)
• The Plugin should ignore comments in XML
• User is able to specify a collection of XPath to output field name mapping
  • User is able to extract values from Attribute (as supported by XPath)
  • User is NOT able to XPaths that are arrays. It should be runtime error. 
• User is able to specify the output field types and the plugin performs appropriate conversions
• User is able to specify what should happen when there is error in processing
  • User can specify that the error record should be ignored
  • User can specify that upon error it should stop processing
  • User can specify that all the error records should be written to separate dataset

Design

Approaches:

• To fetch the output field name, its type and xPath from user.

As an alternative to the output schema in the right panel to fetch the output field name and its types from user, following are the approaches that can be used, in order to read the above information using widgets

First Approach:

Use of 3 field widget - “two text boxes and one dropdown” 

In this case, dropdown will be populated with supported data types, and user will select one out of that. It will ensure the proper data type is entered.

Note : The GroupbyAggregate plugin has a widget with 2 text-boxes and a drop-down.

1st text box - xpathMappings

drop-down - field type

2nd text box - fieldName

But the keyword "as" present in the widget may be confusing to the user.

Image Modified

Or

Second Approach(Implemented):

Use two widgets as mentioned below:

1.  keyvalue(xpathMappings) : to get the output field name and its xPath.

2.   keyvalue-dropdown(fieldTypeMapping): to get the output field name and its type. The type dropdown will have the following options : boolean, int, long, float, double, bytes, string

(schema widget is not visible in the "configuration-groups" on UI)

Note : Both the config properties - xpathMappings and schema will be mandatory

 Or

Third Approach:

Use 3 text box widgets, that we already have.

In this case, we will need to add validation part during the pipeline, to make sure that user has entered the right data types.

Also, the description will include the supported data types, which will help user to enter the correct or expected types for particular field.

Note : This will require extra validations to check if the correct value for field type is entered.

• If the xpath evaluates to array for a xml record, it will throw runtime exception. Eg: For the below xml record if the user gives the xpath as  /Cities /City it will result in runtime exception, which will be handled in the way specified by the user.

  Code Block
  language xml
  <Cities> <City>Paris</City> <City>Lyon</City> <City>Marseille</City> </Cities>

• If the xpath evaluates to a node which contains child node, the plugin will return the node along with all its children as a single xml node string.This is done so that the plugin returns valid xml event for every input structured record.

  Eg: For the below xml record if the user gives the xpath as  /Book/author, the plugin will return the xml event - 

             "<author><name>Dan Brown</name><location>U.S.A</location><author>".

  Code Block
  language xml
  <Book> <title>Angels & Demons</title> <author> <name>Dan Brown</name> <location>U.S.A</location> </author> </Book>

• Eg: For the below xml record if the user gives the xpath as  /Cities /City it will result in runtime exception, which will be handled in the way specified by the user.

• If user chooses to write error to dataset the error record will be emitted using the emitter.emitError() available in transform().
• The fieldName specified in "xpathMappings" and "fieldTypeMapping" should exactly match each other.(This will be validated during configurePipeline())
• The schema for all fields will be of type nullable, since the xpath can evaluate to null if the expression is not satisfied by a node.

Assumptions:

For every structured record received to the transform plugin, the output will also be a single structured record.

Example

Properties:

input                         : Specifies the field in input that should be considered as source of XML event or record.

encoding                 : Specifies XML encoding type(default is UTF-8).

xpathMappings       : Specifies a mapping from XPath of XML record to field name.

fieldTypeMapping   : Specifies the field name as specified in xpathMappings and its corresponding data type. The data type can be of following types- boolean, int, long, float, double, bytes, string 

processOnError       : Specifies what happens in case of error.

                                      1. Ignore the error record
                                      2. Stop processing upon encoutering error
                                      3. Write error record to different dataset

Example:

        This example parses the xml record received in the the "body" field of the structured record, according to the xpathMappings specified, for each field name.

        The type output schema will be created, using the type specified for each field in "fieldTypeMapping" .

{

"name": "XMLParser",
"plugin": {

"name": "XMLParser",
"type": "transform",
"label": "XMLParser",
"properties": {

"encoding": "UTF-8",
"processOnError": "UTF-8",
"xpathMappings": "category://book/@category,title://book/title,year:/bookstore/book[price>35.00]/year,price:/bookstore/book[price>35.00]/price,subcategory://book/subcategory",
"fieldTypeMapping": "category:string,title:string,year:int,price:double,subcategory:string",
"input": "body"

}

}

}

Questions and/or Clarifications

Clarifications:

1. For defining the output field types, field names and xpath value, following approach can be used:
   1. Common widget with 2 text boxes and a drop down   or 
   2. key value widget to take the output field name and xpath expression, and a second output schema widget
      
      

2. User is able to specify what should happen when there is error in processing. Errors could be:

1. IllegalCharacter

2. Type conversion error

3. NULL or EMPTY value for non nullable column value

   
   

3. Requirement: User is NOT able to XPaths that are arrays. It should be runtime error. 

    Understanding: xPath returning multiple nodes with same name. Suppose we have below input:

   Code Block
   language xml
   <Cities> <City>Paris</City> <City>Lyon</City> <City>Marseille</City> </Cities>

    And user wants to extract city ['Paris', 'Lyon', 'Marseille'] and provides xPath till /Cities /City. Then as per our use case, we should throw an error.

    a. Is this understanding correct or are we missing anything on xPath arrays?

    

4. In case, user chooses to write the error records to a separate dataset, then record will be emitted using  emitter.emitError() in transform method.

   1. Is the understanding correct?

Questions:

1. If the xpath evaluates to a node with child elements, how should the plugin handle this?

   1. Return the text in child node elements as comma separated values
   2. Return value as an XML record, similar to the record emitted by XMLReader plugin
   3. Throw exception
   4. or anything other than this