`give` statement with arbitrary expression produces invalid story file

[Natrium729]

The compiler accepts arbitrary expressions for attributes in the give statement. However, it sometimes produces an invalid story file in that case, without emitting an error or a warning.

For example, the following works, giving the attribute with ID 2 to the room:

Attribute locked;

Object obj;

[ Main;
	print "Give bug^";
	give obj 1 + 1;
];

(Compiled with inform6 -~S~D -c test.inf)

However, if we replace 1 + 1 with a parenthesised expression such as (1 + 1) or even (locked), Quixe will show this error message when running the story : "Error: Unknown opcode #0 at pc=201". Gargoyle (git) also complains with a different error message: "Fatal error: too many (48) locals at 0x14 (pc = 0x512)".

On one hand, that situation shouldn't happen often since authors aren't likely to use expression in that place. On the other hand, the compiler should never produce an invalid file without complaining.

I'm not sure if accepting arbitrary expressions in that place is a good idea since it leads to ambiguous syntax. But I'm not really sure what should be done.

[erkyrath]

Expressions should be valid there. This is a code generation bug.

(In the line give obj 1 + 1, the expression gets folded to a constant at compile time. That's why the bug doesn't show up.)

[erkyrath]

Whoops, hang on, I'm wrong about what happened.

The line give obj (1); is parsed as give X; where X is a function call obj(1). That's the ambiguity you were referring to. It's confusing because give X; is legal, according to the language def -- it gives the object X zero attributes.

Yeah, it's not obvious how to even detect this. We could warn on give X; but then the same problem occurs with give obj (1) light;.

Maybe the call syntax X(Y) can do some static type-checking.

[erkyrath]

On the other hand, the compiler should never produce an invalid file without complaining.

This is not a rule at all, I'm afraid. There are an infinite number of ways to produce an invalid story file. Calling X() where X is not a function is just one of them.

[Natrium729]

Ah, I didn't know it was possible to have a give statement without attributes. (And I also totally forgot about constant folding.)

In that case, yeah, the syntax is ambiguous but the compiler isn't wrong either, so I guess it's not a bug after all. (Although I would have thought interpreters would display an error message in the lines of "tried to call something that's not a routine".)

I suppose that, as you said, it should be possible to type-check routine calls, but that's not extremely important either.

[erkyrath]

#172 is the routine-call type check.

line 12: Warning:  In function call, expected Routine but found Object "obj"
>  give obj (1+1);

Of course this warning only appears when obj is an Object literal. If it's a local variable, the compiler can't help you.