Hi,
I toyed a bit with pulling out the JavaCC generated code into its own
compilation tree. I got a working version, but it comes with a price,
so I want do discuss this first before I go ahead and actually
implement it.
The following files will be generated by JJTree:
CharStream.java
JJTParserState.java (JJTree)
Node.java (JJTree)
Parser.java
ParserConstants.java
ParserException.java
ParserTokenManager.java
ParserTreeConstants.java (JJTree)
ParserVisitor.java (JJTree)
Token.java
TokenMgrError.java
The following problems exist:
- While it is possible to use different packages for JJTree and JavaCC, it
actually does not work. Parser.java will use JJTParserState extensively
and that class is package protected. And there seems to be no way around
this.
That also puts the ParserVisitor in the ...parser package. And this is a
problem. In the currently (modified) grammar, the ParserVisitor is part
of the node package. And it must be there because I can't seem to be able
to put an "import org.apache.velocity.parser.node.*" statement into this
autogenerated file and I don't seem to be able to split the JJTParserState
into ...parser and the ParserVisitor into ...parser.node
So in the end, the parser must be in one package. And this means, everything
from the parser.node package (all the AST nodes) would have to go into this
package, too. This is a matter of a simple Refactoring (which eclipse does
happily for me), but it is quite a change (though it doesn't seem to be
really user visible) which I want to discuss first.
- The current "Node.java" interface is an extended version of the original
Node. If we want to keep the auto-generated files clean, we must just rename
it to VelocityNode and extend the autogenerated Node interface.
We also lose jjtOpen, jjtClose, jjtSetParent, jjtGetParent, jjtAddChild
jjtGetChild, jjtGetNumChildren, jjtAccept from this VelocityNode
interface as they are defined in the generated Node.
- in SimpleNode, the jjtGetParent changes its Signature from
VelocityNode to Node, Same goes for jjtSetParent(Node n), jjtGetChild(int i)
and jjtAddChild(Node n, int i)
now there are a bazillion errors with jjtGetChild and returning the
wrong type for all the parser nodes. So we add a getChild(int i) method to
SimpleNode and VelocityNode, which does the same as jjtGetChild(int i) but
returns a VelocityNode instead of a Node. Replace all jjtGetChild()
occurences in the VelocityCode (can also be done with Eclipse
Refactoring).
- As the ParserVisitor class fell behind the changes in the Parser because
it was not autogenerated, the visitors in parser.visitor now have a number
of missing methods. Add visit() for ASTEscapedDirective, ASTEscape, ASTMap,
ASTIntegerRange and ASTStop to NodeViewVisitor and BaseVisitor. This is
IMHO a logic bug BTW which would've been caught by ParserVisitor being
autogenerated instead of manually.
- Adding code to the ParseException wasn't really a good idea. Factor the
template code out into an TemplateParseException, also cleaning up the
code a bit. There are some hints in the JavaCC manual about customizing the
error messages, I will dig into this.
That's it. Lots of code shuffling, not actually many changes. In the end,
IMHO we will benefit from these changes in the long run.
The resulting engine does pass all the tests so at least it does not
crash and burn immediately. Some of that shuffling (the ParseException
bit) I will put in anyway; while the change to display the
templateName is nice, it is not actually intuitive or really readable
code. :-)
If no-one objects loudly, I will apply these changes tomorrow.
Best regards
Henning
--
Dipl.-Inf. (Univ.) Henning P. Schmiedehausen INTERMETA GmbH
[EMAIL PROTECTED] +49 9131 50 654 0 http://www.intermeta.de/
RedHat Certified Engineer -- Jakarta Turbine Development -- hero for hire
Linux, Java, perl, Solaris -- Consulting, Training, Development
4 - 8 - 15 - 16 - 23 - 42
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
