[fs.op.funcs] Remove empty parens when referring to functions by name

[jwakely]

As per the Specification Style Guidelines.

https://github.com/cplusplus/draft/wiki/Specification-Style-Guidelines#describing-function-calls

[jensmaurer]

The specification style guidelines expressly say that "call X" is fine, where X has parentheses (= is an expression). It's just "calling the function Y" needs to have parentheses removed.

Maybe we want to fine-tune the rules a little, but the changes proposed here don't seem to be "obviously in line" to me.

[jwakely]

The specification style guidelines expressly say that "call X" is fine, where X has parentheses (= is an expression).

Yes but it has to be a valid expression, so "calls create_directory()" is wrong, because it needs arguments.

It's just "calling the function Y" needs to have parentheses removed.

I don't see a distinction between "calls f()" and "calling the function f()" in the case where f() is not a valid expression. They're both just "naming" a function, so should not have parentheses.

[tkoeppe]

So, we accept "calls f(1, 2, 3)" (instead of "invokes f with arguments 1, 2, 3"), but if we don't give arguments, then it must be "calls f", not "calls f()", right? We can make that clearer in the wiki, but that seems like a good rule.

[jwakely]

Yes, that's what I thought the rule was. I see that the wiki is actually ambiguous and can be interpreted differently, but that different interpretation seems useless to me. A rule that allows "call sin()" but not "calling the function sin()" seems like a completely pointless rule. You cannot "call sin()" because that's not the name of a function, and calling one of the overloads of sin would require an argument to make it a valid expression.

[jwakely]

If we do want to allow "call sin()" then IMHO we might as well strike the rule entirely, and just allow empty parens on any function name! (This would match POSIX, which intentionally puts empty parens when referring to any function, but I guess that can be argued to be OK in ISO C89 where fopen() declares a function without saying what its parameters are).

[tkoeppe]

Let's fix the wiki instead!

[jensmaurer]

"call sin()" should not be valid. If the wiki needs fixing, so be it.

[tkoeppe]

Actually, is the wiki really ambiguous? It says (emphasis mine),

"""
When we say "Calls blah", blah should be an expression that results in a function call. Eg, "calls operator new(size)" or "calls `current_exception() and stores the returned value".
"""

and since sin() is not an expression that results in a function call, it already doesn't seem to be allowed? We can certainly clarify the wording a bit, but the intention seems clear already.

[jwakely]

Yeah I checked it again yesterday and cross-referenced it with #2150, and the wiki seems pretty clear. And all the changes in this PR are cases where foo() is not a valid expression, and does not disambiguate an overloaded function by naming the foo(void) overload.

[tkoeppe]

OK, great, thanks!