xmlEventParse: XML Event/Callback element-wise Parser

xmlEventParseR Documentation

XML Event/Callback element-wise Parser


This is the event-driven or SAX (Simple API for XML) style parser which process XML without building the tree but rather identifies tokens in the stream of characters and passes them to handlers which can make sense of them in context. This reads and processes the contents of an XML file or string by invoking user-level functions associated with different components of the XML tree. These components include the beginning and end of XML elements, e.g <myTag x="1"> and </myTag> respectively, comments, CDATA (escaped character data), entities, processing instructions, etc. This allows the caller to create the appropriate data structure from the XML document contents rather than the default tree (see xmlTreeParse) and so avoids having the entire document in memory. This is important for large documents and where we would end up with essentially 2 copies of the data in memory at once, i.e the tree and the R data structure containing the information taken from the tree. When dealing with classes of XML documents whose instances could be large, this approach is desirable but a little more cumbersome to program than the standard DOM (Document Object Model) approach provided by XMLTreeParse.

Note that xmlTreeParse does allow a hybrid style of processing that allows us to apply handlers to nodes in the tree as they are being converted to R objects. This is a style of event-driven or asynchronous calling

In addition to the generic token event handlers such as "begin an XML element" (the startElement handler), one can also provide handler functions for specific tags/elements such as <myTag> with handler elements with the same name as the XML element of interest, i.e. "myTag" = function(x, attrs).

When the event parser is reading text nodes, it may call the text handler function with different sub-strings of the text within the node. Essentially, the parser collects up n characters into a buffer and passes this as a single string the text handler and then continues collecting more text until the buffer is full or there is no more text. It passes each sub-string to the text handler. If trim is TRUE, it removes leading and trailing white space from the substring before calling the text handler. If the resulting text is empty and ignoreBlanks is TRUE, then we don't bother calling the text handler function.

So the key thing to remember about dealing with text is that the entire text of a node may come in multiple separate calls to the text handler. A common idiom is to have the text handler concatenate the values it is passed in separate calls and to have the end element handler process the entire text and reset the text variable to be empty.


xmlEventParse(file, handlers = xmlEventHandler(), 
               ignoreBlanks = FALSE, addContext=TRUE,
                useTagName = TRUE, asText = FALSE, trim=TRUE, 
                 useExpat=FALSE, isURL = FALSE,
                  state = NULL, replaceEntities = TRUE, validate = FALSE,
                   saxVersion = 1, branches = NULL,
                    useDotNames = length(grep("^\\.", names(handlers))) > 0,
                     error = xmlErrorCumulator(), addFinalizer = NA,
                      encoding = character())



the source of the XML content. This can be a string giving the name of a file or remote URL, the XML itself, a connection object, or a function. If this is a string, and asText is TRUE, the value is the XML content. This allows one to read the content separately from parsing without having to write it to a file. If asText is FALSE and a string is passed for file, this is taken as the name of a file or remote URI. If one is using the libxml parser (i.e. not expat), this can be a URI accessed via HTTP or FTP or a compressed local file. If it is the name of a local file, it can include ~, environment variables, etc. which will be expanded by R. (Note this is not the case in S-Plus, as far as I know.)

If a connection is given, the parser incrementally reads one line at a time by calling the function readLines with the connection as the first argument (and 1 as the number of lines to read). The parser calls this function each time it needs more input.

If invoking the readLines function to get each line is excessively slow or is inappropriate, one can provide a function as the value of fileName. Again, when the XML parser needs more content to process, it invokes this function to get a string. This function is called with a single argument, the maximum size of the string that can be returned. The function is responsible for accessing the correct connection(s), etc. which is typically done via lexical scoping/environments. This mechanism allows the user to control how the XML content is retrieved in very general ways. For example, one might read from a set of files, starting one when the contents of the previous file have been consumed. This allows for the use of hybrid connection objects.

Support for connections and functions in this form is only provided if one is using libxml2 and not libxml version 1.


a closure object that contains functions which will be invoked as the XML components in the document are encountered by the parser. The standard function or handler names are startElement(), endElement() comment(), getEntity, entityDeclaration(), processingInstruction(), text(), cdata(), startDocument(), and endDocument(), or alternatively and preferrably, these names prefixed with a '.', i.e. .startElement, .comment, ...

The call signature for the entityDeclaration function was changed in version 1.7-0. Note that in earlier versions, the C routine did not invoke any R function and so no code will actually break. Also, we have renamed externalEntity to getEntity. These were based on the expat parser.

The new signature is c(name = "character", type = "integer", content = "", system = "character", public = "character" ) name gives the name of the entity being defined. The type identifies the type of the entity using the value of a C-level enumerated constant used in libxml2, but also gives the human-readable form as the name of the single element in the integer vector. The possible values are "Internal_General", "External_General_Parsed", "External_General_Unparsed", "Internal_Parameter", "External_Parameter", "Internal_Predefined".

If we are dealing with an internal entity, the content will be the string containing the value of the entity. If we are dealing with an external entity, then content will be a character vector of length 0, i.e. empty. Instead, either or both of the system and public arguments will be non-empty and identify the location of the external content. system will be a string containing a URI, if non-empty, and public corresponds to the PUBLIC identifier used to identify content using an SGML-like approach. The use of PUBLIC identifiers is less common.


a logical value indicating whether text elements made up entirely of white space should be included in the resulting ‘tree’.


logical value indicating whether the callback functions in ‘handlers’ should be invoked with contextual information about the parser and the position in the tree, such as node depth, path indices for the node relative the root, etc. If this is True, each callback function should support ....


a logical value. If this is TRUE, when the SAX parser signals an event for the start of an XML element, it will first look for an element in the list of handler functions whose name matches (exactly) the name of the XML element. If such an element is found, that function is invoked. Otherwise, the generic startElement handler function is invoked. The benefit of this is that the author of the handler functions can write node-specific handlers for the different element names in a document and not have to establish a mechanism to invoke these functions within the startElement function. This is done by the XML package directly.

If the value is FALSE, then the startElement handler function will be called without any effort to find a node-specific handler. If there are no node-specific handlers, specifying FALSE for this parameter will make the computations very slightly faster.


logical value indicating that the first argument, ‘file’, should be treated as the XML text to parse, not the name of a file. This allows the contents of documents to be retrieved from different sources (e.g. HTTP servers, XML-RPC, etc.) and still use this parser.


whether to strip white space from the beginning and end of text strings.


a logical value indicating whether to use the expat SAX parser, or to default to the libxml. If this is TRUE, the library must have been compiled with support for expat. See supportsExpat.


indicates whether the file argument refers to a URL (accessible via ftp or http) or a regular file on the system. If asText is TRUE, this should not be specified.


an optional S object that is passed to the callbacks and can be modified to communicate state between the callbacks. If this is given, the callbacks should accept an argument named .state and it should return an object that will be used as the updated value of this state object. The new value can be any S object and will be passed to the next callback where again it will be updated by that functions return value, and so on. If this not specified in the call to xmlEventParse, no .state argument is passed to the callbacks. This makes the interface compatible with previous releases.


logical value indicating whether to substitute entity references with their text directly. This should be left as False. The text still appears as the value of the node, but there is more information about its source, allowing the parse to be reversed with full reference information.


an integer value which should be either 1 or 2. This specifies which SAX interface to use in the C code. The essential difference is the number of arguments passed to the startElement handler function(s). Under SAX 2, in addition to the name of the element and the named-attributes vector, two additional arguments are provided. The first identifies the namespace of the element. This is a named character vector of length 1, with the value being the URI of the namespace and the name being the prefix that identifies that namespace within the document. For example, xmlns:r="http://www.r-project.org" would be passed as c(r = "http://www.r-project.org"). If there is no prefix because the namespace is being used as the default, the result of calling names on the string is "". The second additional argument (the fourth in total) gives the collection of all the namespaces defined within this element. Again, this is a named character vector.


Currently, this has no effect as the libxml2 parser uses a document structure to do validation. a logical indicating whether to use a validating parser or not, or in other words check the contents against the DTD specification. If this is true, warning messages will be displayed about errors in the DTD and/or document, but the parsing will proceed except for the presence of terminal errors.


a named list of functions. Each element identifies an XML element name. If an XML element of that name is encountered in the SAX stream, the stream is processed until the end of that element and an internal node (see xmlTreeParse and its useInternalNodes parameter) is created. The function in our branches list corresponding to this XML element is then invoked with the (internal) node as the only argument. This allows one to use the DOM model on a sub-tree of the entire document and thus use both SAX and DOM together to get the efficiency of SAX and the simpler programming model of DOM.

Note that the branches mechanism works top-down and does not work for nested tags. If one specifies an element name in the branches argument, e.g. myNode, and there is a nested myNode instance within a branch, the branches handler will not be called for that nested instance. If there is an instance where this is problematic, please contact the maintainer of this package.

One can cause the parser to collect a branch without identifying the node within the branches list. Specifically, within a regular start-element handler, one can return a function whose class is SAXBranchFunction. The SAX parser recognizes this and collects up the branch starting at the current node being processed and when it is complete, invokes this function. This allows us to dynamically determine which nodes to treat as branches rather than just matching names. This is necessary when a node name has different meanings in different parts of the XML hierarchy, e.g. dict in an iTunes song list.

See the file itunesSax2.R inthe examples for an example of this.

This is a two step process. In the future, we might make it so that the R function handling the start-element event could directly collect the branch and continue its operations without having to call another function asynchronously.


a logical value indicating whether to use the newer format for identifying general element function handlers with the '.' prefix, e.g. .text, .comment, .startElement. If this is FALSE, then the older format text, comment, startElement, ... are used. This causes problems when there are indeed nodes named text or comment or startElement as a node-specific handler are confused with the corresponding general handler of the same name. Using TRUE means that your list of handlers should have names that use the '.' prefix for these general element handlers. This is the preferred way to write new code.


a function that is called when an XML error is encountered. This is called with 6 arguments and is described in xmlTreeParse.


a logical value or identifier for a C routine that controls whether we register finalizers on the intenal node.


a character string (scalar) giving the encoding for the document. This is optional as the document should contain its own encoding information. However, if it doesn't, the caller can specify this for the parser.


This is now implemented using the libxml parser. Originally, this was implemented via the Expat XML parser by Jim Clark (http://www.jclark.com/).


The return value is the ‘handlers’ argument. It is assumed that this is a closure and that the callback functions have manipulated variables local to it and that the caller knows how to extract this.


The libxml parser can read URLs via http or ftp. It does not require the support of wget as used in other parts of R, but uses its own facilities to connect to remote servers.

The idea for the hybrid SAX/DOM mode where we consume tokens in the stream to create an entire node for a sub-tree of the document was first suggested to me by Seth Falcon at the Fred Hutchinson Cancer Research Center. It is similar to the XML::Twig module in Perl by Michel Rodriguez.


Duncan Temple Lang


http://www.w3.org/XML/, http://www.jclark.com/xml/

See Also

xmlTreeParse xmlStopParser XMLParserContextFunction


 fileName <- system.file("exampleData", "mtcars.xml", package="XML")

   # Print the name of each XML tag encountered at the beginning of each
   # tag.
   # Uses the libxml SAX parser.
                list(startElement=function(name, attrs){
                useTagName=FALSE, addContext = FALSE)

## Not run: 
  # Parse the text rather than a file or URL by reading the URL's contents
  # and making it a single string. Then call xmlEventParse
xmlURL <- "https://www.omegahat.net/Scripts/Data/mtcars.xml"
xmlText <- paste(scan(xmlURL, what="",sep="\n"),"\n",collapse="\n")
xmlEventParse(xmlText, asText=TRUE)

## End(Not run)

    # Using a state object to share mutable data across callbacks
f <- system.file("exampleData", "gnumeric.xml", package = "XML")
zz <- xmlEventParse(f,
                    handlers = list(startElement=function(name, atts, .state) {
                                                     .state = .state + 1
                                                 }), state = 0)

    # Illustrate the startDocument and endDocument handlers.
               handlers = list(startDocument = function() {
                                                 cat("Starting document\n")
                               endDocument = function() {
                                                 cat("ending document\n")
               saxVersion = 2)

if(libxmlVersion()$major >= 2) {

 startElement = function(x, ...) cat(x, "\n")

 xmlEventParse(ff <- file(f), handlers = list(startElement = startElement))

 # Parse with a function providing the input as needed.
 xmlConnection = 
  function(con) {

     con = file(con, "r")
   if(isOpen(con, "r"))
     open(con, "r")

   function(len) {

     if(len < 0) {

      x = character(0)
      tmp = ""
    while(length(tmp) > 0 && nchar(tmp) == 0) {
      tmp = readLines(con, 1)
      if(length(tmp) == 0)
      if(nchar(tmp) == 0)
        x = append(x, "\n")
        x = tmp
    if(length(tmp) == 0)
    x = paste(x, collapse="")


 ## this leaves a connection open
 ## xmlConnection would need amending to return the connection.
 ff = xmlConnection(f)
 xmlEventParse(ff, handlers = list(startElement = startElement))

  # Parse from a connection. Each time the parser needs more input, it
  # calls readLines(<con>, 1)
 xmlEventParse(ff <-file(f),  handlers = list(startElement = startElement))

  # using SAX 2
 h = list(startElement = function(name, attrs, namespace, allNamespaces){ 
                                 cat("Starting", name,"\n")
          endElement = function(name, uri) {
                          cat("Finishing", name, "\n")
 xmlEventParse(system.file("exampleData", "namespaces.xml", package="XML"),
               handlers = h, saxVersion = 2)

 # This example is not very realistic but illustrates how to use the
 # branches argument. It forces the creation of complete nodes for
 # elements named <b> and extracts the id attribute.
 # This could be done directly on the startElement, but this just
 # illustrates the mechanism.
 filename = system.file("exampleData", "branch.xml", package="XML")
 b.counter = function() {
                nodes <- character()
                f = function(node) { nodes <<- c(nodes, xmlGetAttr(node, "id"))}
                list(b = f, nodes = function() nodes)

  b = b.counter()
  invisible(xmlEventParse(filename, branches = b["b"]))

  filename = system.file("exampleData", "branch.xml", package="XML")
  invisible(xmlEventParse(filename, branches = list(b = function(node) {
  invisible(xmlEventParse(filename, branches = list(b = function(node) {

  # Stopping the parser mid-way and an example of using XMLParserContextFunction.

  startElement =
  function(ctxt, name, attrs, ...)  {
      if(name == "rewriteURI") {
           cat("Terminating parser\n")
  class(startElement) = "XMLParserContextFunction"  
  endElement =
  function(name, ...) 
    cat("ending", name, "\n")

  fileName = system.file("exampleData", "catalog.xml", package = "XML")
  xmlEventParse(fileName, handlers = list(startElement = startElement,
                                          endElement = endElement))

XML documentation built on Dec. 4, 2022, 5:06 p.m.

Related to xmlEventParse in XML...