P1208R6 <source_location>

[StephanTLavavej]

P1208R6 <source_location>

LWG-3396 Clarify point of reference for source_location::current() (DE 169)

This needs compiler support.

The IDE's list of extensionless headers has been updated for VS 2019 16.5.

Feature-test macro as of WG21-N4842:
#define __cpp_lib_source_location 201907L

[StephanTLavavej]

Microsoft-internal MSVC-PR-229114 implemented intrinsics for VS 2019 16.6 Preview 2:

const char* __builtin_FILE();
const char* __builtin_FUNCTION();
int __builtin_LINE();
int __builtin_COLUMN();

[SuperWig]

@StephanTLavavej Is this not also blocked on consteval?

[StephanTLavavej]

Aha, consteval didn't exist when I first looked at source_location years ago, but it's there in the paper now. I'm unsure as to how much of a blocker this is; thanks for bringing it up!

[SuperWig]

I'm unsure as to how much of a blocker this is

Isn't not being able to compile (including tests, which from what I've read makes Billy sad) a pretty big blocker?

Having it be constexpr instead doesn't seem like a correct thing to do as that's not what the paper says 😕.

---

So I have this implemented (with constexpr) and with the following

constexpr void f() {
    constexpr auto x = source_location::current();
    static_assert(x.line() == 21);
    static_assert(x.column() == 41);
    static_assert(x.function_name() == "f");
    static_assert(x.file_name() == "D:\\Projects\\STL Testing\\STL Testing\\STL Testing.cpp");
}

The static_asserts for foo_name() fail though regular asserts pass. Seeing as it also fails by using the builtins directly would that be a bug with __builtin_FUNCTION() and __builtin_FILE()?

---

Question, what is the value of the following supposed to be?

struct s1
{
    s1(source_location x = source_location::current()) : loc(x) {}
    source_location loc;
};

struct s2
{
    s1 s;
};

int main()
{
    s2 x;
    assert(x.s.loc.line() == ???);
    assert(x.s.loc.column() == ???);
}

Because I'm getting 1 for the column and for the line, the line of } for main. (Same with int and __builtin_LINE()). Interestingly in the non-isolated version of this test the line was for the line after the declaration of x (also the line of the assert).

---

More quirks detected! __builtin_COLUMN() is giving the character number instead of the column.

	auto x = source_location::current();

So using tabs here gives 28 for the column number when I expected 31.

[StephanTLavavej]

Thanks for the investigation; we'll need to talk to the compiler devs again.

[SuperWig]

One question I forgot to ask was, where is the value of the column supposed to be?

source_location::current();
                 ^
source_location::current();
^

Because I got the first but kinda expected the latter.

[AdamBucior]

One question I forgot to ask was, where is the value of the column supposed to be?

source_location::current();
                 ^
source_location::current();
^

Because I got the first but kinda expected the latter.

Paper describes the column value as implementation-defined, so I suppose it's OK. Also both of these values seem fine to me as I can't imagine a situation in which it'd matter.

[SuperWig]

Yeah, I saw that. I guess I was more or less asking if it was intentional for it to be that position and/or if one was preferred, rather than asking which is "correct".

[StephanTLavavej]

VS 2019 16.6 Preview 2 shipped on March 26, 2020.