Does the solution have to be a regex? I would have thought this would be easier to accomplish by using something like XML::Twig to slurp up the XML, and then walking through the resulting data structure and picking out the appropriate node.

/<para0[^>]*?>
 (?: \s* (?: <title .*?<\/title>
            |<para .*?<\/para>
            |<applic .*?<\/applic>
            |<capgrp .*?<\/capgrp>
            |<subpara1 .*?<\/subpara1>
            |<caution .*?<\/caution>  ) )*?
 \s* (<warning .*? <\/warning>)
 (?: \s* (?: <title .*? <\/title>
            |<para .*? <\/para>
            |<applic .*? <\/applic>
            |<capgrp .*? <\/capgrp>
            |<subpara1 .*? <\/subpara1>
            |<caution .*? <\/caution>
            |<warning .*? <\/warning>  ) )*?
 \s* <\/para0>                             /sx
[download]

The code works as follows:

1. Find an element starting with <para0>, which may have attributes
2. Non-grouping match any number of title, para, applic, capgrp, subpara1, or caution tags
3. Match and capture your warning tag
4. Non-grouping match any number of title, para, applic, capgrp, subpara1, caution or warning tags
5. Close the search with the closing </para> tag

Note that your entire XML must be in a single string (not an array) and should be executed with the /s modifier.

Update: Some additional notes. This assumes that your XML is well-formed. It assumes you have at least one warning element in your para0 element. And most importantly, if there are 1st generation tags which are not accounted for, .*? can jump to unexpected locations, meaning you will not get what you expected. Compare that to ikegami's solution, which will just work.

Update 2: At shmem's suggestion, I added an /x modifier and reformatted the regex to make it (maybe) easier to follow.

THANK YOU!

For the record, as a developer that manipulates XML a lot!!!, I EMPHATICALLY AGREE regex is not the way to go and I'm not happy that I need to rely on regex to solve my problem. Unfortunately, I am in a situation where portions of both XML and SGML documents need to be modified and the modifications must be restricted to those specifically made (e.g., entity references cannot be resolved). The current implementation uses string-based manipulations to ensure this and regexes to retrieve values and identify replacement sections. I simply need to retrieve one more thing and therefore need a regex to do so.

My hope would be that at some time I'd be able to find a solution that allows us to use XPath, etc. yet still be able to re-generate the *exact* same XML/SGML originally parsed.

THANKS AGAIN! You are a life-saver!

Try parsing the XML instead, using one of the XML modules at CPAN. RegExps don't work too well with hierarchical data formats.

Alexander

The regex needs to return the first warning element directly parented by a para0 element (no grandchildren allowed).

As an XPath, that would be

/descendant::para0/child::warning[1]
[download]

In shorthand:

//para0/warning[1]
[download]

Other than "don't do it with REs", I have a couple of thoughts...

You can "fix" your RE using the highly experimental extended pattern "(?>pattern)". This prevents your patterns matching the non-warning tags spanning into other tags. The modified pattern is:

<para0[^>]*>\s*(?><applic[^>]*>.*?</applic\s*>)?\s*(?><title[^>]*>.*?<
+/title\s*>)?\s*(?><capgrp[^>]*>.*?</capgrp\s*>)?\s*(<warning[^>]*>\s*
+.*?\s*</warning\s*>)
[download]

I don't understand your criteria: you say you want to match "the first warning element directly parented by a para0 element". I see two such elements in your sample data:

  <warning>
   <?warningcaution MOD_A ### ID_1?>
   <para>moda id1 warning</para>
  </warning>
[download]

and

  <warning>
   <?warningcaution MOD_C ### ID_3?>
   <para>modc id3 warning</para>
  </warning>
[download]

So, I don't understand why you say "If warning MOD_A/ID_1 is removed, my expected response is no match." instead of matching the second warning directly parented by para0. I am misunderstanding something or you have a contradiction in your requirements.

Another RE you might consider is the following:

<para0[^>]*?>(?>\s*<([^ >]*?).*?>.*?</\1(\s[^>]*)?>)*?\s*(<warning>.*?
+</warning>)
[download]

This assumes that none of the tags directly under the para0 tag nest within themselves but it does not need to be modified if the set of possible tags changes. Note that this pattern has two capture buffers, so your warning will be in $2 if there is a match.