Oct 14, 2009

Updated Alpha release version, Oct 2009

I finally upgraded the download (Alpha-release) version and brought it in sync with latest sources. I removed some dead and not-used-yet code, did some code cleanup and a bit of refactoring. Aside this, there are a few changes and improvements that I would like to comment on.
The first important change is renaming SymbolTerminal to KeyTerm. I think that the name SymbolTerminal was misleading and a bit confusing, and it became more so as I faced a need for a terminal for scanning elements known as "symbols" in many programming languages; for example, in Ruby it is an identifier preceeded by a colon, like :name. The former SymbolTerminal, on the other hand, was created to recognize the language-defined keywords and special characters (operator, assignment, braces, etc). So the name change makes sense, and "key" prefix in KeyTerm comes from the "keyword". This name change caused obvious name changes to corresponding dictionaries, lists and fields. It also made sense to rename the  Symbol() method in the Grammar class - it is used to explicitly convert the first string in string-only BNF expressions and allow proper method overload to kick in. For example:

BinOp.Rule = Symbol("+") | "-" | "*" | "/";

I changed the name of the method to ToTerm(string), but I kept the old version Symbol() and decorated it with Obsolete attribute. The Symbol() method is heavily used in grammars, and leaving an old method for a while would allow to compile the old grammars without errors. The compiler will only give a warning that the method was deprecated and you should use ToTerm instead. Now the rule for BinOp should look like this:

BinOp.Rule = ToTerm("+") | "-" | "*" | "/";

Please make a change in your existing grammars.
Continuing with this theme, I'm thinking about better ways to handle keywords internally. It looks like this Grammar.CurrentGrammar thread-static property is more trouble than it's worth. The main problem is that keyword terminals can be created in static operator overloads (BnfTerm.Op_Pipe, Op_Plus etc). At the same time the terminal for a particular keyword should be instantiated once in a grammar, and with proper case sensitivity arrangement. Therefore, the grammar (an instance) should have a list/dictionary of keyword terminals which should be accessible from static operator overload methods. Currently it is done through this Grammar.CurrentGrammar property, but I'm wondering if there's a better way. Will see.
Another change is a new Grammar.RunSample virtual method. The purpose of this method is to allow custom implementations of running a sample code in the Grammar Explorer. The base implementation in Grammar class executes the code using Irony's AST interpreter - this works well for the expression evaluator grammar, as an example. The SearchGrammar (Google-to-SQL search syntax converter) overrides this method and provides it's own implementation - it converts the input string into SQL server syntax using its own custom method. Using the RunSample override allowed to get rid of a separate sample project I had before for testing the search grammar - now you can do it directly in the Grammar Explorer.
One more thing: the language flag SupportsInterpreter is renamed to CanRunSample. If this flag is set, the Grammar Explorer enables the Run button and calls the Grammar.RunSample method on click.
That's it for now - more stuff is coming. We are on track for release by the end of the year.
Enjoy it!

No comments:

Post a Comment