[RFC] Paren-less access to scalar extension methods ('a,b'->arr as sugar for 'a,b'->arr()) View as Markdown

Hi,

This follows up on the Extension Methods / Scalar Extension Methods RFCs (currently in draft). I'd like to propose allowing extension methods on scalar values to be invoked through property-fetch syntax, with the parentheses omitted.

Proposal: when a property read on a scalar fails, fall back to the extension method registry and invoke the method with its default arguments. ->name becomes sugar for ->name().

extension string $s {
    public function arr(string $separator = ','): array {
        return explode($separator, $s);
    }
}

'a,b'->arr; // ['a', 'b'] <- new
'a,b'->arr(); // ['a', 'b'] <- already works
'a|b'->arr('|'); // ['a', 'b'] <- already works

Only methods callable with zero arguments would be eligible. 'abc'->needsArg would throw ArgumentCountError. Unknown names keep today's behavior.

Why this is cheap here: the main use case for scalar extension methods is derived values (length, upper, slug, toArray), which read like properties. And first-class callables are already disabled for scalar receivers (a by-value receiver cannot be bound into a Closure), so $s->arr cannot be interpreted as "return a Closure" - the only remaining meaning is "call it and return the value". That removes the usual ambiguity that makes this expensive in C#.

Scope: scalar receivers only (string, int, float, bool, array). Object receivers are excluded on purpose: a missing property on an object is usually a typo, and turning that warning into a successful call would weaken a useful diagnostic. Scalars have no property table today, so nothing is shadowed.

Backward compatibility: no parser change; ->name already compiles to ZEND_FETCH_OBJ_R. Today the affected expressions always emit "Warning: Attempt to read property "x" on string" and evaluate to null, which no correct code can rely on. Verified against a local build of hollyschilling/php-src@extension-methods-scalars (5e5112f):

'aaa'->length; // Warning: Attempt to read property "length" on string -> NULL
'a,b'->arr; // Warning: Attempt to read property "arr" on string -> NULL

These two lines are exactly what the proposal would change.

Open questions I'd like the list's input on:

A. When a method has a required parameter, throw ArgumentCountError (my preference; a silent null would be indistinguishable from a typo), or fall back to the existing warning?

B. Should isset() / ?? consult the registry, or stay untouched for v1?

C. Is excluding object receivers the right call, or is the resulting inconsistency ($str->upper works, $obj->upper does not) worse than losing the typo diagnostic on objects?

Happy to prototype against the extension-methods-scalars branch once (A) is settled.

Thanks,
h4acda

497161114
h4acda@qq.com