Arity errors

[neumino]

For

r.table("test").zip(1)

The server will send back this error:

RqlCompileError: Expected 1 argument(s) but found 2. in:
r.db("test").table("test").zip(1)

The error we send back is not super-friendly because

• There is no backtrace
• We do not report the name of the function that was called with the wrong number of arguments
• The server suppose that everything uses the notation r.zip(r.table('test'), 1)

About the last point, r.and is a valid syntax, so we can't just suppose that we always chain methods. But if the function that triggered the error was reported, the driver could catch the error and update the number of arguments depending of the method.

Putting in polish.

[neumino]

Actually, we could have one term for r.add and one for datum.add.
That way:

• it solve the problem of the number of arguments
• we can build better backtraces
• it would avoid strange things like r.add(1)

Same thing for the other commands, especially for tableCreate, tableDrop and table.

[mlucy]

That sounds really bad to me. The terms sent to the server should represent an AST, which should be independent of what people typed to produce it. We'd end up having two of every term, and some languages (like Clojure or C) would only ever send half of them, then on the server we'd turn them into one "real" term.

I would rather make the server return structured error messages ({type: 'type_error', term: 'add', arg: '3', message: ..., bt: ...}) so that drivers can choose to render them in driver-specific ways (since the driver knows whether you were using infix or postfix notation to generate the AST).

[coffeemug]

I would rather make the server return structured error messages ({type: 'type_error', term: 'add', arg: '3', message: ..., bt: ...}) so that drivers can choose to render them in driver-specific ways (since the driver knows whether you were using infix or postfix notation to generate the AST).

👍

[neumino]

r.add(1) and r.expr(1).add() compile to the same query.

So if the driver wants to make the distinction between the two notations, it has to build a second "query" that keeps track of the notation used, which is cumbersome to do.

That being said, I'm ok just sending back a structured error message for now. That would make errors way more readable.