missing-field-initializers warning not reporting missing designated initializers

[ottojo]

In the following code, the missing-field-initializers does not produce a warning that b is not explicitly initialized.

Changing CheckForMissingFields to True here:

llvm-project/clang/lib/Sema/SemaInit.cpp

Lines 2167 to 2169 in a9a60f2

	// Disable check for missing fields when designators are used.
	// This matches gcc behaviour.
	CheckForMissingFields = false;

seems to produce the expected warning.

The comment suggests

// Disable check for missing fields when designators are used.
// This matches gcc behaviour.

However, GCC does produce the warning in that case.
Comparison with GCC: https://godbolt.org/z/WeYdY11q3

struct ExampleStruct {
    int a;
    int b;
};

ExampleStruct buildExampleStruct(int a) {
    ExampleStruct e{
            .a=a
    };
    return e;
}

Build flags: -Werror=missing-field-initializers -std=c++20

Expected behavior:

main.cpp:10:5: error: missing field 'b' initializer [-Werror,-Wmissing-field-initializers]

matching GCCs

<source>:9:5: error: missing initializer for member 'ExampleStruct::b' [-Werror=missing-field-initializers]

Actual behavior: Compiles without warning/error.

[dbelousoff]

Is there any updates/workarounds for this?

[shafik]

I don't see a workaround, it looks out of date at this point. Someone needs to remove the CheckForMissingFields = false; but this is a potentially breaking change since there may be folks that using this flag that won't expect the warning to start popping up. Although it should.

[ottojo]

I see how this would be a breaking change, although i guess if someone enabled Wextra and Wall Werror they would expect some breakage when updating the compiler...

Anyways, i felt like hacking at this a bit and maybe introducing a dedicated switch for this might be an idea? I think the warning is useful, and i would certainly enable it in my projects (although discovery of this flag might be a problem...)
An attempt of an implementation is here, which seems to work in the above test at least 😄 https://github.com/ottojo/llvm-project/pull/2/files

[gracicot]

I'm actually hit by this issue and I actually missed a bug in my code because I assumed the warning was working on Clang too.

[shafik]

@AaronBallman sounds like missing this case is not ideal, wdyt about making this change?

[AaronBallman]

@AaronBallman sounds like missing this case is not ideal, wdyt about making this change?

I'm comfortable with the change. If it turns out to be highly disruptive, we can always put the diagnostic under a new warning flag grouped under -Wmissing-field-initializers so that users can selectively disable this (but hopefully we won't need such a measure).

[Fznamznon]

https://reviews.llvm.org/D157879