Regular expression for extracting tag attributes

Although the advice not to parse HTML via regexp is valid, here's a expression that does pretty much what you asked:

/
   \G                     # start where the last match left off
   (?>                    # begin non-backtracking expression
       .*?                # *anything* until...
       <[Aa]\b            # an anchor tag
    )??                   # but look ahead to see that the rest of the expression
                          #    does not match.
    \s+                   # at least one space
    ( \p{Alpha}           # Our first capture, starting with one alpha
      \p{Alnum}*          # followed by any number of alphanumeric characters
    )                     # end capture #1
    (?: \s* = \s*         # a group starting with a '=', possibly surrounded by spaces.
        (?: (['"])        # capture a single quote character
            (.*?)         # anything else
             \2           # which ever quote character we captured before
        |   ( [^>\s'"]+ ) # any number of non-( '>', space, quote ) chars
        )                 # end group
     )?                   # attribute value was optional
/msx;

"But wait," you might say. "What about *comments?!?!" Okay, then you can replace the . in the non-backtracking section with: (It also handles CDATA sections.)

(?:[^<]|<[^!]|<![^-\[]|<!\[(?!CDATA)|<!\[CDATA\[.*?\]\]>|<!--(?:[^-]|-[^-])*-->)

• Also if you wanted to run a substitution under Perl 5.10 (and I think PCRE), you can put \K right before the attribute name and not have to worry about capturing all the stuff you want to skip over.

Just to agree with everyone else: don't parse HTML using regexp.

It isn't possible to create an expression that will pick out attributes for even a correct piece of HTML, never mind all the possible malformed variants. Your regexp is already pretty much unreadable even without trying to cope with the invalid lack of quotes; chase further into the horror of real-world HTML and you will drive yourself crazy with an unmaintainable blob of unreliable expressions.

There are existing libraries to either read broken HTML, or correct it into valid XHTML which you can then easily devour with an XML parser. Use them.