Why is semicolon required in this single line Standard ML code: Int.toString 5?

Question

I have a file foo.sml with a single line of Standard ML code:

Int.toString 5

This runs fine in SML/NJ but not in MLton:

$ cat foo.sml
Int.toString 5

$ sml < foo.sml
Standard ML of New Jersey (64-bit) v110.99.5 [built: Thu Mar 14 17:56:03 2024]
- = [autoloading]
[library $SMLNJ-BASIS/basis.cm is stable]
[library $SMLNJ-BASIS/(basis.cm):basis-common.cm is stable]
[autoloading done]
val it = "5" : strin

$ mlton foo.sml
Error: foo.sml 3.0.
  Syntax error found at EOF.
Error: foo.sml 3.0-3.0.
  Parse error.

If I add a semicolon, the problem gets resolved:

$ cat foo.sml
Int.toString 5;

$ sml < foo.sml
Standard ML of New Jersey (64-bit) v110.99.5 [built: Thu Mar 14 17:56:03 2024]
- [autoloading]
[library $SMLNJ-BASIS/basis.cm is stable]
[library $SMLNJ-BASIS/(basis.cm):basis-common.cm is stable]
[autoloading done]
val it = "5" : string
- 

$ mlton foo.sml
$ ./foo
$

Why is semicolon necessary in this single line code for MLton to compile it successfully?

Answer 1

The syntax of SML programs is defined as follows (The Definition of Standard ML, page 64):

𝑝𝑟𝑜𝑔𝑟𝑎𝑚 ::= 𝑡𝑜𝑝𝑑𝑒𝑐 `;` 𝑝𝑟𝑜𝑔𝑟𝑎𝑚?

Hence, a program must be terminated with a semicolon. So MLton is correct, while SML/NJ is more liberal than the standard and allows the final semicolon before the end-of-file to be omitted.

Edit: There is the added complication that using a bare expression at the toplevel is a so-called derived form, i.e., syntactic sugar, and that is again only defined with a semicolon (page 72, Fig. 18). In general, if the semicolon was omitted between two such expressions, then this syntax would become ambiguous (because x x is an application). Hence, I suspect that MLton's parser requires the semicolon only around such expressions. (Technically, I suppose it parses a file as either a program, or just a 𝑡𝑜𝑝𝑑𝑒𝑐.)