[antlr-interest] Embedding expressions in plain text -- how to?

[Gavin Lambert]

At 08:59 11/11/2007, Rob Greene wrote:
>How should I go about embedding an expression language in plain 
>text.  The expression is wrapped in '${' and '}' but may occur 
>zero to many times within plain text.  When in plain text, I want 
>to keep the whitespace, but within the expression, it should be 
>ignored.
>
>I've currently placed a boolean into a ThreadLocal and check that 
>before calling skip() in my whitespace rule.  But, that causes 
>issues with the actual expressions.
>
>For instance, "${true}" returns Boolean.TRUE while "${ true }" 
>returns a null.  If I take out the test in WS, I can have those 
>spaces within an expression, but the plain text loses all white 
>space.

The way I'd do it would be to create multiple lexers/parsers.  (It 
is possible to do it in a single one, but I think it's cleaner to 
separate it out.)

At the first level, you just need a lexer-only grammar to separate 
plain text and expressions.  You need to make the whole ${...} 
construct a single token (see the STRING rules in the wiki for an 
example), and everything else a different token.

Once you've done that, in your app code loop through all the 
returned expressions and then run another lexer/parser over them 
to work out what they actually mean :)

(If you do want to combine it all into one operation, then look up 
the "island grammar" examples.  Same basic idea.)