R & SPlus XML Parsers

Latest version: XML_1.6-3.tar.gz

R Package

Latest News

  • Support for S4/Splus5 for the Tree Parsing
  • Event driven parsing and function callbacks not yet added for S4/Splus5. Requires mutable state and hence integration of the CORBA/Java/XML driver interface for this kind of thing.
  • Fix of some trivial bugs.
  • Overview

    XML is one of several facilities we are investigating for enhancing communication between applications. The ability to easily transfer data between applications is enhanced if the data is self-describing. One can imagine datasets being well-formed documents that specify a DTD specifying their format. For example, Statlib datasets would contain information about the variables, number of observations, levels of factors, etc. as well as meta-data about the source of the data, precision, additional commands for specific systems to assist in interpreting the data. This format allows readers to perform single passes by announcing their dimensions, types, etc. at the top of the document. Additionally, entities in the DTD and document can contain code for specific systems. These then act as "portable" methods. (Security is an issue, but somewhat orthogonal to the parsing mechanism.)

    Data can also be exchanged dynamically with other systems that use XML. For example, Office, Oracle, Lotus Notes, browsers, HTTP servers, etc.

    The markup language for mathematics - MathML - will be important in the research end of statistics, and also to some extend the applied data analysis for specifying models, etc. See Math ML Example and you can fetch the DTD from mmlents.zip

    Scalable Vector Graphics is an XML based format for specifying graphics descriptions that can be scaled easily without distortion. We may be using it (or an extension of it) in Omegahat to represent plots. The DTD is available from here.

    Since XML is similar to HTML, we can encourage people to use this type of format for different inputs. We have effectively used for defining options with potentially more complicated structures than simple name-value pairs. Hierarchical structures are easily handled by XML. Plot descriptions can be described in this way, and indeed we intend to do this in Omegahat.

    This XML-approach is in contrast to a simple ASCII or native object dump which relies on the receiving system or user to understand the format. (Communicating via the S4 object ASCII dump format was used effectively to communicate between Java and S4, but was heavily dependent on the parsing facilities being migrated to Java, and any other system engaging in such communication.)
    In contrast with the embedded Java facilities and CORBA packages for R and S, XML is a more static representation of data rather than a live object offering methods.

    In addition to providing an environment neutral form of persistence, XML can be used for configuration files, plot template descriptions, documentation, etc.

    The aim of providing facilities in R and S for reading XML at the user level is to encourage users to consider the development of DTDs for statistical data and concepts. If we can "standardize" on some basic descriptions, we can exchange data easily between numerous systems, including spreadsheets, etc. These DTDs, coupled with Java interface classes and IDL modules create an integrated framework for open network computing in multiple ways and at multiple user levels. We strongly encourage people to actively make their DTDs available to others.

    In the future, we will develop facilities for writing objects from R, S and Omegahat in XML format using the DTDs we develop. A general mechanism for having validated output filters can be created. See Writing XML

    Package Overview

    This is a small, but sufficient, collection of both C routines and R and S functions for reading the contents of XML files for processing in these two environments. There are two general styles of XML parsers.
  • Document/Tree-based.
  • Here, the entire XML document is read and a tree constructed containing the different elements in the document. At this point, we process the elements by traversing the tree and generating a user-level representation of the nodes. We allow the user to specify functions that are called for different types of nodes so that she can customize the resulting tree as it is being constructed.
  • Event driven.
  • This style involves reading the XML elements in the document one at a time and invoking different user level functions/methods that correspond to the type of element - a tag, text, entity, CData, etc. The methods are responsible for processing the information locally and building the appropriate data structure from all of them. Thus, no tree need be constructed and traversed post-construction. This reduces the memory used in reading the document and provides much greater control over the parsing.

    Rather than offering one of these styles, we provide functions that work both ways for R. In S, we currently only support the document/tree-based approach. xmlTreeParse() is the tree based version which generates an internal tree and then converts it to a list of lists in R/S. This is uses the XML library from Daniel Veillard of W3.org.

    The second function, xmlEventParse(), is event driven. The user specifies a collection of R/S-level functions, in addition to the file name, and the parser invokes the appropriate function as new XML elements are encountered. The C-level parser we use is Expat developed by Jim Clark.

    Unless you have very large XML documents, if you want to experiment with just one parser, use the first of these, i.e the document-based one. That is the simplest to use, sacrificing control of the creation of the data structures and potential memory growth.

    In R, the collection of functions is usually a closure and it can manipulate local data. In S, these are usually a list of functions. In order to handle mutable state, one should use the interface driver mechanism. The closure approach is described in more detail in Docs/Outline.nw and the R document in man/.


    Tree-Based Parsing

    The first thing we do is perform a simple parse of a document. Make certain to specify the correct location of the file.
    >  x <-  xmlTreeParse("data/test.xml")
    > names(x)
    [1] "file"     "version"  "children"
    > x$file
    [1] "data/test.xml"
    > names(x$children[[1]])
    [1] "name"       "attributes" "children"   "value"     

    Now we turn our attention to manipulating the previously generated tree. We can do this in R/S using the following version of treeApply.

    treeApply <- function(x, func, post=NULL, pre=NULL, ...) {
     ans <- NULL
     value <- func(x)
       ans <- list(value=value)
       # If there are any children, do a recursive apply on those also.
       # If the result is non-null
     if( length(x[["children"]]) > 0 ) {
       tmp <- lapply(x[["children"]], treeApply, func, ...)
       if(length(tmp) > 0)
         ans$children - tmp
      # invoke the post-processing of children hook.
     if(length(post)) {
    Armed with this version of apply(), we can start doing some processing of the tree. First, lets display the type of each node in the tree.
    v <- treeApply(x, function(x) cat(class(x),"\n"))
    A slightly more interesting example is to produce a graphical display of the tree. I use PStricks for this purpose. We define a node function that produces the relevant TeX commands and also a post function to tidy up the groups.
    foo <-  function(x) {
         label <- ifelse(length(x$value), x$value, ifelse(length(x$name), x$name,"doc"))
         if(length(x[["children"]])==0) {
          } else {
     post <- function(x) if(length(x$children) > 0) cat("}\n")
    treeApply(x, foo, post= post)
    The result is
    \Tr{ A comment }%
    \Tr{ }%
    \Tr{test entity}%
    \Tr{print "This is some more PHP code being executed."; }%

    Note that the post function is more naturally done in an event-driven parser, via the endElement handler.

    Another example is that this document has been carefully constructed to be parseable by the xmlTreeParse function.

     v <- xmlTreeParse("index.html")

    Event-driven Examples

    The event-driven style is essentially a filtering mechanism. It provides lower level control over the processing of the elements. Because both R and S do not have references, incremental processing is slightly more complicated than it is in languages such as C or Java. However, the event-driven style does allow us to avoid reading the entire document into memory at once and is ideal for situations when most of the document is not of interest, but a small number of nodes are important and their location in the document is necessary to understand.

    A simple example is where we gather all the character text in the document. In other words, we throw away the XML hierarchical structure and any nodes that are not simply character text.

    characterOnlyHandler <- function() {
      txt <- NULL
      text <- function(val,...) {
        txt <<- c(txt, val)
      getText <- function() { txt }
      return(list(text=text, getText=getText))
    z <- xmlEventParse("data/job.xml", characterOnlyHandler())
     [1] "  "                                                                   
     [2] "    "                                                                 
     [3] "      "                                                               
     [4] "      "                                                               
     [5] "GBackup"                                                              
     [6] "      "                                                               
     [7] "Development"                                                          
     [8] "      "                                                               
     [9] "        "                                                             
    [10] "Open"                                                                 
    [11] "        "                                                             
    [12] "Mon, 07 Jun 1999 20:27:45 -0400 MET DST"                              
    [13] "        "                                                             
    [14] "USD 0.00"                                                             
    [15] "      "                                                               
    [16] "      "                                                               
    [17] "        "                                                             
    [18] "        "                                                             
    [19] "      "                                                               
    [20] "      "                                                               
    [21] "        "                                                             
    [22] "Nathan Clemons"                                                       
    [23] "        "                                                             
    [24] "nathan@windsofstorm.net"                                              
    [25] "        "                                                             
    [26] "        "                                                             
    [27] "        "                                                             
    [28] "        "                                                             
    [29] "        "                                                             
    [30] "        "                                                             
    [31] "        "                                                             
    [32] "        "                                                             
    [33] "        "                                                             
    [34] "        "                                                             
    [35] "      "                                                               
    [36] "      "                                                               
    [37] "      The program should be released as free software, under the GPL."
    [38] "      "                                                               
    [39] "    "                                                                 
    [40] "  "                                                                   
    Note that we can discard the lines that are simply white space using the trim argument. This trims all text values. More granularity is needed here.
    z <- xmlEventParse("data/job.xml", characterOnlyHandler(), ignoreBlanks=T, trim=T)
    > z$getText()
    [1] "GBackup"                                                        
    [2] "Development"                                                    
    [3] "Open"                                                           
    [4] "Mon, 07 Jun 1999 20:27:45 -0400 MET DST"                        
    [5] "USD 0.00"                                                       
    [6] "Nathan Clemons"                                                 
    [7] "nathan@windsofstorm.net"                                        
    [8] "The program should be released as free software, under the GPL."

    Much as we did with the tree-based parser, we can construct a display of the structure of the document using the event driven parser.

                     list(startElement = function(x,...){
                          endElement   = function(x,...)
    Note that we use a list of functions rather than a closure in this example. This is because we do not have data that persists across function calls.

    Parsing the mtcars.xml file (or generally files using the DTD used by that file) can be done via the event parser in the following manner. First we define a closure with methods for handling the different tags of interest. Rather than using startElement and looking at the name of the tag/element, we will instruct the xmlEventParse to look for a method whose name is the same as the tag, before defaulting to use the startElement() method. As with most event driven material, the logic is different and may seem complicated. The idea is that we will see the dataset tag first. So we define a function with this name. The dataset tag will have attributes that we store to attach to the data frame that we construct from reading the entire XML structure. Of special interest in this list is the number of records. We store this separately, converting it to an integer, so that when we find the number of variables, we can allocate the array.

    The next we do is define a method for handling the variables element. There we find the number of variables. Note that if the DTD didn't provide this count, we could defer the computation of variables and the allocation of the array until we saw the end of the variables tag. This would allow the user to avoid having to specify the number of variables explicitly.

    As we encounter each variable element, we expect the next text element to be the name of the variable. So, within the variable() method, we set the flag expectingVariableName to be true. Then in the text() function, we interpret the value as either a variable name if expectingVariableName is true, or as the value of a record if not. In the former case, we append the value to the list of variable names in varNames. We need to set the value expectingVariableName to false when we have enough. We do this when the length of varNames equals the number of columns in data, computed from the count attribute.

    A different way to do this is to have an endElement() function which set expectingVariableName to false when the element being ended was variables. Again, this is a choice and different implementations will have advantages with respect to robustness, error handling, etc.

    The text() function handles the case where we are not expecting the name of a variable, but instead interpret the string as the value of a record. To do this, we have to convert the collection of numbers separeted by white space to a numeric vector. We do this by splitting the string by white space and the converting each entry to a numeric value. We assign the resulting numeric vector to the matrix data in the current row. The index of the record is stored in currentRecord. This is incremented by the record method. (We could do this in text() also, but this is more interesting.)

    We will ignore issues where the values are separated across lines, contain strings, etc. The latter is orthogonal to the event driven XML parsing. The former (partial record per line) can be handled by computing the number seen so far for this record and storing this across calls to text() and adding to the appropriate columns.

     handler <- function() {
      data <- NULL
        # Private or local variables used to store information across 
        # method calls from the event parser
      numRecords <- 0
      varNames <- NULL
      meta <- NULL
      currentRecord <- 0
      expectingVariableName <- F
      rowNames <- NULL
       # read the attributes from the dataset
      dataset <- function(x,atts) {
        numRecords <<- as.integer(atts[["numRecords"]])
          # store these so that we can put these as attributes
          # on data when we create it.
        meta <<- atts
      variables <- function(x, atts) {
          # From the DTD, we expect a count attribute telling us the number
          # of variables.
        data <<- matrix(0., numRecords, as.integer(atts[["count"]]))
          #  set the XML attributes from the dataset element as R
          #  attributes of the data.
        attributes(data) <<- c(attributes(data),meta)
      # when we see the start of a variable tag, then we are expecting
      # its name next, so handle text accordingly.
      variable <- function(x,...) {
         expectingVariableName <<- T
      record <- function(x,atts) {
          # advance the current record index.
        currentRecord <<- currentRecord + 1
        rowNames <<- c(rowNames, atts[["id"]])
      text <- function(x,...) {
       if(x == "")
       if(expectingVariableName) {
         varNames <<- c(varNames, x)  
         if(length(varNames) >= ncol(data)) {
             expectingVariableName <<- F
             dimnames(data) <<- list(NULL, varNames)
       } else {
          e <- gsub("[ \t]*",",",x)
          vals <- sapply(strsplit(e,",")[[1]], as.numeric)
          data[currentRecord,] <<- vals
        # Called at the end of each tag.
      endElement <- function(x,...) {
       if(x == "dataset") {
             # set the row names for the matrix.
           dimnames(data)[[1]]  <<- rowNames
       return(list(variable = variable,
                   variables = variables,
                   text  = text,
                   record= record,
                   endElement = endElement,
                   data = function() {data },
                   rowNames = function() rowNames
    A more robust version of this that handles rownames and produces a data frame rather than a is given in the function dataFrameEvents

    Obtaining the Software


    The package is maintained as part of the Omegahat source tree, under the control of the Omega Project for Statistical computing. You can obtain the source in the following ways:
  • R package
  • The uncompiled, installable version as an R package. This is probably the easiest to install as at the end you can simply invoke library(XML).
    You can use the GNUmakefiles in libxml and expat to configure each of those distributions appropriately. (Basically, these build shared libraries.)
  • Basic source
  • (What is this?)
    The raw source. This compiles the same way as the R package. It just cannot be loaded as easily.
  • via anonymous CVS
  • The most up-to-date raw source.
    There are no binaries for Unix. If there is a need, please ask.

    This software is known to run on both Linux (RedHat 6.1) and Solaris (2.6).

    To run the R functions, you will need to install either or both of the following packages.

    See Installing 3rd party software.


    The code, documentation, etc. is released under the terms of the GNU General Public License and the owner of the copyright is the Omega Project for Statistical Computing.

    The goal is to share this code with an S4/Splus5 version. In order to keep the programming interfaces consistent, we would appreciate being notified of changes.


    The package, also known as chapter, can be configured to use either of the XML parsing styles discussed in the above, or both. The event-based parser uses the Expat library by Jim Clark . The tree/document-based parser uses libxml from Daniel Veillard. You can use either or both of these. First install whichever of these you will use, and make sure to build them as shared libraries. See below for some assistance in doing this.

    Having decided to use either libxml and/or expat, you must specify their locations. Edit the GNUmakefile, and uncomment the line defining LIBXML and/or LIBEXPAT as appropriate. Change the value on the right hand side of the = sign to the location of these directories.

    Next, you need to specify whether you are building for R or S4/Splus5. You can do this via the variable LANGUAGE in the GNUmakefile. It defaults to R. All of these can be specified on the command line such as:

      make LIBXML=$HOME/libxml-1.7.3 LIBEXPAT=$HOME/expat LANGUAGE=R CC=gcc

    Installing from XML_1.6-3.tar.gz

    1. Untar the XML_1.6-3.tar.gz file in the appropriate directory, probably one of the library/ directories your R distribution searches for libraries. (See library(), R_LIBS, etc.)
    2. cd XML
    3. Invoke make, specifying the different values for the 3rd party distributions, etc. on the command line.
    4.        make LIBXML=$HOME/libxml-1.7.3 LIBEXPAT=$HOME/expat LANGUAGE=R CC=gcc

    I have installed using the makefiles here and the GNUmakefile.admin in the omegahat source tree version of this. That however relies on some other makefiles in the R section of the Omegahat tree. If any one else wishes to package this, please send me the changes I can make them available to others. Of course you can use it by just attaching the chapter and using dyn.load().
    Some of this would be easier if we used either the R or S4/Splus5 package installation facilities. However, I do not have time at the moment to handle both cases in the common code.

    Make sure to specify the location of the library path. Use the environment variable LD_LIBRARY_PATH to include the location of the libxml distribution and also the lib directory in the expat distribution.

      setenv LD_LIBRARY_PATH ${LIBXML}:${LIBEXPAT}/lib
    or, in bash
      export LD_LIBRARY_PATH=${LIBXML}:${LIBEXPAT}/lib

    Microsoft Windows

    There is now a version of the package for Windows. One can install from source or download a binary, pre-compiled version of the package from Brian Ripley's R windows package builds.

    Installing From Binary

    Change directory to the location in which you want to install the library. This is usually R_HOME/library.
    Download the zip file XML_1.6-3.zip.
    Unzip the contents of the zip file.
    unzip XML_1.6-3.zip
    Download the libxml2-2.4.13 distribution for Windows created by Igor Zlatkovic.
    You will possibly need the iconv libraries also.
    Install the libxml2 (and iconv) libraries into a directory and add that to your PATH.
    Run R and load the library using library(XML)!

    Installing From Source on Windows

    To install from source, you can follow these steps.

    File List

    DocParseparser using libxml.
    EventParse parser using expat.
    RSCommonFile that allows us to use the same code for R and S4/Splus5 by hiding the differences between these two via C pre-processor macros. This file is copied from $OMEGA_HOME/Interfaces/CORBA/CORBAConfig
    Utilsroutines shared by both files above for handling white space in text.
    RS_XML.hname space macro for routines used to avoid conflicts with routine names with other libraries.
    RSDTDRoutines for converting DTDs to user-level objects.
    GNUmakefilemakefile controlling the compilation of the shared libraries, etc.
    expat/ makefiles that can be copied into expat distribution to make shared libraries for use here.
    libxml/makefiles that can be copied into libxml distribution to make shared library
    Src/R/S functions for parsing XML documents/buffers.
    man/R documentation for R/S functions.
    Docs/document (in noweb) describing initial ideas.
    data/example functions, closure definitions, DTDs, etc that are not quite official functions.

    Installing libxml and/or Expat

    The following information helps in installing the 3rd party libraries. The approach is optional, but the need is to build shared libraries. GNU makefiles are provided (in the subdirectories expat/ and libxml/ of this distribution) to perform the necessary operations. A simple way to place these in the appropriate distribution is to give the command,
         make LIBEXPAT=/dir/subdir expat
         make LIBXML=/dir/subdir libxml
    These requires GNU make to be installed.

    These makefiles circumvents the regular Makefiles in the distributions.

    Installing Expat.

    1. Unzip the exapt.zip file. This will create a directory expat/.
    2. Copy the contents of the directory named expat within the directory where you are reading this installation file. There should be 4 files in total that are copied. Two of these GNUmakefile and GNUmakefile.lib go into expat/. There are two others, one in each of xmltok and xmlparse that should be copied to the corresponding directories in the expat distribution.
      You can do this via the command
      make LIBEXPAT=/wherever expat

      issued from this directory.

      Before doing this, you will have to edit these files to ensure that the correct values are used for compiling shared libraries. At present, there are settings for gcc and Solaris compilers. Edit the file expat/GNUmakefile.lib and comment out the settings that do not apply to your machine. Specifically, if you are using the GNU compiler (gcc), comment out the two lines for the Solaris compilers (the second settings for PIC_FLAG and PIC_LD_FLAG)

    3. Change directory to the expat/ distribution.
    4. Type make.

    Installing libxml

    The steps are similar to those for expat. However, when compiling this for use with Splus5/S4, there are additional steps. Please follow these or you will likely see segmentation faults, etc. due to conflicting symbols.
    1. Untar the libxml distribution, creating a directory called, say, libxml/.
    2. Copy the single GNUmakefile in the directory below this one (where you are reading this file) named libxml/ to the location you have installed the libxml distribution.
      You can do this via the command
      make LIBXML=/wherever libxml      

      You will have to edit these files to ensure that the correct values are used for compiling shared libraries. At present, there are settings for gcc and Solaris compilers.
    3. Change directory back to the libxml distribution.
    4. Type ./configure.
    5. Type make.
    An alternative to this involves the following steps. It has not been extensively tested at all.
    1. Apply the patch in the directory libxml/ to the libxml directory. This can be done via the commands.
        cd libxml
        make LIBXML=/wherever/installed  patch
    2. Append the following lines to either Makefile.in or Makefile in the libxml distribution (depending on whether you have alread configured that distribution and/or whether you want the changes to persist across reconfigurations).
      libxml.so : $(OBJS)
      	$(CC) $(SHARED_LD_FLAGS) -o $@  $(OBJS)
      	@echo $(CFLAGS)
    3. Now, in that libxml distribution directory, issue the command:
       make libxml.so CFLAGS="-fpic `make Cflags`" SHARED_LD_FLAGS=-shared 

    Compiling libxml for use with Splus5/S4

    Both S4 and libxml have a symbol attribute. Because of the way dynamically loaded code resolves symbols, the libxml facilities will use the one from S4, incorrectly. Until we determine the appropriate linker flags, please modify the three references to attribute in libxml before compiling the shared libraries. The following patch makes the changes. Apply them by invoking the
     make libxmlpatch
     (PWD=`pwd`; export PWD ; cd $(LIBXML) ; patch -f <  $(PWD)/libxml/PATCH.libxml)
    This works with the GNU patch.


    Recent Changes

    Duncan Temple Lang <duncan@wald.ucdavis.edu>
    Last modified: Mon Dec 13 21:28:37 EST 1999