RE: [xml-dev] Accurate line and column numbers for elements and attribut

["Klotz, Leigh" <Leigh.Klotz@xerox.com>]

Thank you for your quick and detailed answer, John.  I think you're
saying that because of attribute value normaliation rules, it's
inconvenient for parser authors to report attribute line number and
ill-defined for column number, because of the space normaliation rules
mean navigating beyond the column is difficult.  

For the record, with saxon9 I'm able to use the AugmentedSource and
obtain line numbers for attributes retrieved from XPath expressions,
just not column numbers as they're not in the NodeInfo API.  (NodeInfo
gives warnings about how vague line numbers can be, especailly for
SAX-constructed documents.)

            AugmentedSource source =
AugmentedSource.makeAugmentedSource(new StreamSource(new
File(filename)));
            source.setLineNumbering(true);
            Configuration config = new Configuration();
            config.setLineNumbering(true);
            StaticQueryContext sqc = new StaticQueryContext(config);
            DocumentInfo doc = sqc.buildDocument(source);
            XPathEvaluator xpath = new XPathEvaluator(doc);
            List nodeList = (List)xpath.evaluate("/data/*/@line", doc,
XPathConstants.NODESET);
            int len = nodeList.size();
            for (int i = 0; i < len; i++) {
                Object item = nodeList.get(i);
                if (item instanceof NodeInfo) {
                    NodeInfo nodeInfo = (NodeInfo)item;
                    System.out.println(nodeInfo.getDisplayName() + " " +
nodeInfo.getLineNumber());
                } else {
                    System.out.println(item);
                }
            }

Another reason it may not be implemented is that using SAX to handle
validation errors is 80% of the solution because SAX will give you the
correct answer in its own errors.

Thank you,
Leigh.
 

-----Original Message-----
From: John Snelson [mailto:john.snelson@ORACLE.COM] 
Sent: Friday, June 06, 2008 10:44 AM
To: Klotz, Leigh
Cc: xml-dev@lists.xml.org
Subject: Re: [xml-dev] Accurate line and column numbers for elements and
attributes in Java

Hi Leigh,

There's a reason that most XML parsers don't report accurate line and
column numbers - the XML specification makes it difficult in a number of
ways. For instance, the namespaces spec basically requires that all
attributes of an element be parsed before the element can be reported,
which often means that line and column information for the attributes is
lost. Similarly, the XML spec applies attribute value normalization to
each attribute value, meaning that the original position of characters
in the source document is lost.

What a lot of XML tools end up doing is reporting a line number and then
doing something like quoting the attribute value marked with the
position of the error. This is quite unsatisfactory when you're used to
accurate line and column numbers from other tools - especially for
programming languages.

In writing my own parser FAXPP [1] (in C, so not great for you), one of
my aims was to be able to use it to report accurate line and column
numbers when parsing XSLT. In order to do that it has an option to turn
off attribute value normalization - but of course the application then
has to perform the same operation later to be described as a conformant
XML parser.

John

[1] http://faxpp.sourceforge.net/

Klotz, Leigh wrote:
> I'm developing an XML representation for a templating system which can
> report errors at both parse and at execution time.
> 
> The existing system does a good job of reporting input file line
numbers
> on its element- and attribute-like constructs, and I'd like to provide
> that same functionality with the XML representation.
> 
> Most SAX2 implementations (including the default one with recent Sun
> Java systems) offer start and end line numbers of tags, but provide
only
> the ending column number of element content.
> 
> While I'm able to make use of this information, the most critical need
> for column numbers comes in attribute values, so that errors in
> attribute-value template expressions can be reported with column
numbers
> accurate in the original input stream.  SAX2 is no help here, as the
> attributes spring forth fully formed as if from Zeus's forehead.
> 
> I've taken a look at Piccolo, which has a commented-out set of column
> number info (on elements only), and at Micahel Kay's suggestion of
using
> the Saxon AugmentedSource (
>
http://www.nabble.com/-xml-dev--Line-number-of-a-node-to7990605.html#a79
> 90605 ) but again this provides only line numbers, not column numbers,
> and not for attributes.
> 
> Does anyone have experience or recommendations for a non-validating,
> namespace-aware XML parser in Java which supports or can easily be
made
> to support accurate beginning line and column numbers of both elements
> and attributes (and maybe text)?  The solution needs to be released
> under a BSD license.
> 
> Thank you,
> Leigh.

-- 
John Snelson, Oracle Corporation            http://snelson.org.uk/john
Berkeley DB XML:            http://oracle.com/database/berkeley-db/xml
XQilla:                                  http://xqilla.sourceforge.net