Regular expression to match ">", "<", "&" chars that appear inside XML nodes - php

Question

I'm trying to write a regular expression using the PCRE library in PHP.
I need a regex to match only &, > and < chars that exist within string part of any XML node and not the tag declaration themselves.
Input XML:
<pnode>
<cnode>This string contains > and < and & chars.</cnode>
</pnode>
The idea is to to a search and replace these chars and convert them to XML entities equivalents.
If I was to convert the entire XML to entities the XML would look like this:
Entire XML converted to entities
<pnode>
<cnode>This string contains > and < and & chars.</cnode>
</pnode>
I need it to look like this:
Correct XML
<pnode>
<cnode>This string contains > and &lt and & chars.</cnode>
</pnode>
I have tried to write a regular expression to match these chars using look-ahaead but I don't know enough to get this to work. My attempt (currently only attempting to match > symbols):
/>(?=[^<]*<)/g
Just to make it clear the XML I'm trying to fix comes from a 3rd party and they seem unable to fix it their end hence my attempt to fix it.

Answer 1

In the end I've opted to use the Tidy library in PHP. The code I used is shown below:
// Specify configuration
$config = array(
'input-xml' => true,
'show-warnings' => false,
'numeric-entities' => true,
'output-xml' => true);
$tidy = new tidy();
$tidy->parseFile('feed.xml', $config, 'latin1');
$tidy->cleanRepair()
This works perfectly correcting all the encoding errors and converting invalid characters to XML entities.

Answer 2

Classic example of garbage in, garbage out. The real solution is to fix the broken XML exporter, but obviously that's out of the scope of your problem. Sounds like you might have to manually parse the XML, run htmlentites() on the contents, then put the XML tags back.

Answer 3

I'm reasonably certain it's simply not possible. You need something that keeps track of nesting, and there's no way to get a regular expression to track nesting. Your choices are to fix the text first (when you probably can use an RE) or use something that's at least vaguely like an XML parser, specifically to the extent of keeping track of how the tags are nested.
There's a reason XML demands that these characters be escaped though -- without that, you can only guess about whether something is really a tag or not. For example, given something like:
<tag>Text containing < and > characters</tag>
you and I can probably guess that the result should be: ...containing < and >... but I'm pretty sure the XML specification allows the extra whitespace, so officially "< and >" should be treated as a tag. You could, I suppose, assume that anything that looks like an un-matched tag really isn't intended to be a tag, but that's going to take some work too.

Answer 4

Would it be possible to intercept the text before it tries to become part of your XML? A few ounces of prevention might be worth pounds of cure.

Answer 5

This should do it for ampersands:
/(\s+)(&)(\s+)/gim
This means you're only looking for those characters when they have whitespace characters on both sides.
Just make sure the replacement expression is "$1$2amp;$3";
The others would go like this, with their replacement expressions on the right
/(\s+)(>)(\s+)/gim "$1>$2"
/(\s+)(<)(\s+)/gim "$1<$2"

Answer 6

As stated by others, regular expressions don't do well with hierarchical data. Besides, if the data is improperly formatted, you can't guarantee that you'll get it right. Consider:
<xml>
<tag>Something<br/>Something Else</tag>
</xml>
Is that <br/> supposed to read <br/>? There's no way to know because it's validly formatted XML.
If you have arbitrary data that you wish to include in your XML tree, consider using a <![CDATA[ ... ]]> block instead. It's treated the same as a text node, and the only thing you don't have to escape is the character sequence ]]>.

Answer 7

What you have there is not, of course, XML. In XML, the characters '<' and '&' may not occur (unescaped) inside text: only inside a comment, CDATA section, or processing instruction. Actually, '>' can occur in text, except as part of the string ']]>'. In well-formed XML, literal '<' and '&' characters signal the start of markup: '<' signals the start of a start tag, end tag, or empty element tag, and '&' signals the start of an entity reference. In both these cases, the next character may NOT be whitespace. So using an RE like Robusto's suggestion would find all such occurrences. You might also need to catch corner cases like '<<', '<\', or '&<'. In this case you don't need to try to parse your input, an RE will work fine.
If the source contains strings like '<something ' where 'something' matches the production for a Name:
Name ::= NameStartChar (NameChar)*
Then you have more of a problem. You are going to have to (try to) parse your input as if it were real XML, and detect the error cases of malformed Names, non-matching start & end tags, malformed attributes, and undefined entity references (to name a few). Unfortunately the error condition isn't guaranteed to happen at the location of the error.
Your best bet may be to use an RE to catch 90% of the error and fix the rest manually. You need to look for a '<' or '&' followed by anything other than a NameStartChar