Ambiguous grammar?

Robert-Egan

I have two tokens in my SQL grammar, SchemaName and TableName.

If I have the TableName in the grammar before the SchemaName, my "DROP SCHEMA <SchemaName>" production fails with this error message
Encountered an error at (or somewhere around) input:1:13 Was expecting one of the following: SchemaName Found string "dbo.test" of type TableName
BUT, if I switch the order of the token definitions in the grammar, it works. Now, I haven't tested DROP TABLE <TaleName> yet, but I have a hunch that if I did, it would fail now that it found string "dbo,test" of type SchemaName.

So, what am I doing wrong, and how can I do it right?

revusky

Well... I don't think your grammar is ambiguous. It's not doing what you think it should, but that is another matter...

Let's see... you're defining two token types in your lexical grammar, SchemaName and TableName. Are these the exact same regular expression pattern or just similar, i.e. not exactly the same but with significant overlap?

One thing that seems clear is that the string "dbo.test" matches both ShemaName and TableName but it will be taken to be one or the other based on which pattern occurs first. This is the basic rule. The first match wins. (Actually, it's the longer match wins, then if they are the same length, the first match, the one that occurs earlier wins.)

So, anyway, the string "dbo.test" will always be matched as TableName and never as SchemaName. So the syntactic rule of DROP SCHEMA <SchemaName> will never work because you will have <TableName> instead of <SchemaName>. And that is what is going on.

I guess one question is whether you really need two token types SchemaName and TableName. If you just had one type, TableOrSchemaName or maybe just Name, then that would work surely.

On the other hand, if, for whatever reason, you want to have those two token types, you could have:

      <DROP> <SCHEMA> DEACTIVATE_TOKENS TABLE_NAME (<SchemaName>)

And since you deactivate the TABLE_NAMEtoken type, the string "dbo.test" will now be matched as a SCHEMA_NAME token type.

That is another possibility.

Robert-Egan

Thanks. I think I need to revisit my initial approach. I originally ran into a situation where "dbo.test" would not pass at all unless there was space on either side of the period "dbo . test". So I though perhaps instead of SchemaName and TableName being productions, they should be tokens. I guess this missed the mark as well.

It seems I need to learn more about the two states than I currently do. I will continue to poke at various combinations, but would welcome any recommended reading. I don't really want to stumble on a design that only works by accident but is actually brittle.