[[nodiscard]] messages

[AlexGuteniev]

Fixes #271
Fixes #1742

[fsb4000]

You can also add: "Fixes #1742"

Oops, I was wrong:

STL/stl/inc/yvals_core.h

Line 427 in 0fbb82d

#elif __has_cpp_attribute(nodiscard) >= 201907L

[AlexGuteniev]

Oops, I was wrong

Why, it still fixes #1742 where possible

[fsb4000]

Why, it still fixes #1742 where possible

You are correct.
I tested on godbolt: https://godbolt.org/z/r1fPM3b9K
So __has_cpp_attribute(nodiscard) >= 201907L is true for /std:c++14

[StephanTLavavej]

Thanks for this major diagnostic improvement 😻 - and apologies for taking so very long to review this! I've pushed a bunch of changes, renaming/rephrasing stuff and updating a few more locations to use the message macros. Please let me know if anything looks undesirable. (I was probably most aggressive with rephrasing the assume_aligned message; I thought it was better to follow your launder pattern and not describe in too much detail how the compiler may or may not optimize.)

[AlexGuteniev]

Please let me know if anything looks undesirable.

I intentionally did not place full stop at the end of each message for consistency with other ranges. Consider this program:

[[nodiscard("don't discard. the value is useful")]]
bool lock()
{
    return false;
}

int main()
{
    int i;
    lock();
}

The warnings (/W4):

ConsoleApplication14.cpp(19,9): warning C4858: discarding return value: don't discard. the value is useful
ConsoleApplication14.cpp(18,9): warning C4101: 'i': unreferenced local variable

Do you still want to have it?

[StephanTLavavej]

(capturing Discord discussion here for the record)

The compiler definitely prefers short phrases without periods for most of its warning messages. However, longer messages (typically static analysis warnings) are sometimes full sentences with periods, like https://godbolt.org/z/v9ansc3fa :

<source>(3) : warning C28251: Inconsistent annotation for 'new': this instance has no annotations.
See c:\users\containeradministrator\appdata\local\temp\compiler-explorer-compiler2022728-2624-10iodxs.ufwa\predefined c++ types (compiler internal)(33).
The first user-provided annotation on this built-in function is at line c:\data\msvc\14.33.31629\include\vcruntime_new.h(48). 

And our [[deprecated("message")]] explanations, which are the closest precedent for [[nodiscard("message")]], have periods:

STL/stl/inc/yvals_core.h

Lines 935 to 939 in 8ca7bd3

	[[deprecated("warning STL4014: " \
	"std::result_of and std::result_of_t are deprecated in C++17. " \
	"They are superseded by std::invoke_result and std::invoke_result_t. " \
	"You can define _SILENCE_CXX17_RESULT_OF_DEPRECATION_WARNING " \
	"or _SILENCE_ALL_CXX17_DEPRECATION_WARNINGS to suppress this warning.")]]

Thus while there are arguments for both ways, we've concluded that keeping the periods would be best.

[StephanTLavavej]

I'm speculatively mirroring this to the MSVC-internal repo - please notify me if any further changes are pushed.

[StephanTLavavej]

Thanks again for enhancing the STL's favorite warning! 😻 ⚠️ 🚀