The grammar already specifies that tags can be next to one another.
I'm not entirely sure that Parse::RecDescent works the way I describe below, but most parser generators such as YACC and AntLR do.
First of all, the input is split in a stream of tokens. Tokens are specified by regular expressions and are supposed to be separated by some separator (mostly whitespace). Let's assume the input to be parsed looks like:
Now we would like to get the following tokens (quoted and separated by commas for legibility):
So as you already indicated, there are two types of tokens, tags and text, and they can be defined as you did.
Note the + though. Each of these token definitions will capture what we want them to capture, tags and text respectively. So by applying these definitions, we can split the input in the desired stream of tokens. This is phase one, the lexical analysis. If you're using YACC, you'll do this by using LEX.
Now for phase two: now we specify how the tokens can appear in the input stream so that we consider the input "valid", ie. conform to the grammar. This is quite simple in this case:
Now we're no longer trying to match characters, but rather tokens, so the input to this phase looks like:
This is a very simple grammar, so it isn't obvious here, but the result is in fact a (parse) tree that looks like:
INPUT / / | | \ \ TAG TEXT TAG TAG TEXT TAG(sorry for the rather lousy graphics ;-) If we just want to verify that the input satisfies the grammar, we're done. In general though, we want to do something with the parsed input, so we have to attach actions to the grammar nodes. Something like "add the content of the TEXT to some list" or whatever. This is the semantics of the grammar.
Now for "empty" tokens, it should now seem strange to have an empty token, each token is some meaningful entity in the input.
I hope this clarifies matters a bit, if not, don't hesitate to ask, -gjb-