The next step

What is the next step in Antlrvsix to help improve a grammar? To answer that, I decided to go back and take a look at where this all began, with the comparison of the Java grammars in Antlr.

Java9 is derived from version 9 of the Java spec. Parr’s Java is derived from an unknown, earlier version of the spec, but it is much faster in parsing that the other grammar.

How can one compare these grammars? A line-by-line diff of the files could be done, but it is a terrible way to compare the grammars because they are very different: one is a partitioned grammar, the other a combined grammar; the formatting is different; the rules in each grammar are ordered differently; and, the rules differ considerably between the grammars. In order to compare the grammars, a much smarter diff is needed.

What I’m planning is to do a rule-by-rule comparison, working with rules in a DFS order from the start rule. I’ll need to know how the rules differ and why. Then, using Antlrvsix, apply a transformation to make the rules the same, and continue the diff with the next rule in the DFS ordering.

So, I first applied a DFS reorder of the grammars using the Antlrvsix extension, then a reformat (it turns out Codebuff has some bugs). Starting with the start rules, compilationUnit, the difference has already started: Parr decided to unfold ordinaryCompilation and remove the rule.

I could edit the grammar manually for the rule, but I decided to save time and implement the fold transformation right now. And, in order to make this reproducible, I want Antlrvsix to take a sequence of transformations from a file and apply them.

–Ken

Posted in Tip