0005834: Confused by Marshal type safety

Notes
(0008532) gasche (developer) 2012-11-19 10:05	The current behavior of Marshal with respect to typing is rather simple: functions of the Marshal module have a type of the form (foo -> 'a), where foo is the input type (string, in_channel, etc.). This means that their return type is fully polymorphic, just as the return type of (raise : exn -> 'a) for example. You can do anything with a value of such a type and the type-checker can't stop you. The advice to add an explicit type annotation is precisely there to refine the static information of the program and allow for more static verification. About warning 20: when an expression produced has a fully polymorphic type such as ('a), then in particular you can instantiate it to a function type of the form ('b -> 'c). But if it was produced without magic (Marshal or Obj) you know, statically, that this function will not use its argument. Indeed, to use its argument a function must evaluates to something of the form (fun foo -> ...), and therefore be type-checked as a function type (_ -> _) rather than a fully polymorphic type. Warning 20 is there to tell the programmer: "you're doing something weird by trying to pass an argument to something that I know isn't a regular function". # let rec loop arg = loop arg;; val loop : 'a -> 'b = <fun> # loop () 1;; Warning 20: this argument will not be used by the function. Interrupted. # (raise Exit) 1;; Warning 20: this argument will not be used by the function. Exception: Pervasives.Exit. Of course, in the context of Marshal, this warning isn't always appropriate because it may be the case that what you unmarshalled really is a function. It's only a way for the compiler to tell you it doesn't have any static knowledge about this function -- so annotating it with some type information would be a good idea, if only to provide better static information about the function return type. To sum up, this warning is useful in other circumstances and doesn't hurt in this case. Besides it would be technically difficult/complicated to special-case the result of unmarshaling (that would require a non-local test) to disable the warning. I don't expect this behavior to change in a medium-term future.

Issue History
Date Modified	Username	Field	Change
2012-11-19 07:00	warwick	New Issue
2012-11-19 10:05	gasche	Note Added: 0008532
2012-11-19 10:05	gasche	Status	new => resolved
2012-11-19 10:05	gasche	Resolution	open => no change required
2012-11-19 10:05	gasche	Assigned To	=> gasche

Relationships