Reason for missing notes for incomplete-type errors in standard headers

Reason for missing notes for incomplete-type errors in standard headers

[Julian Lenz]

Hi everybody,

TL;DR:

What is the reason that `error: '<some class>' has incomplete type` does 
not give a note about where the forward declaration happened for 
standard library classes?

Full:

Recently, I forgot to include <array> but had <tuple> included. Due to 
this recent commit for gcc12

https://github.com/gcc-mirror/gcc/commit/261d5a4a459bd49942e53bc83334ccc7154a09d5

the `error: 'std::array<int,N>' has incomplete type` felt like a weird 
regression because all other major gcc versions as well as clang 
compiled just fine. After many hours of debugging, I finally realised 
that it was just a missing header and now the error message actually 
makes sense.

But when I experimented with this, I found that gcc does give you 
another `<filename>:<line number>:<column number>: note: declaration of 
'class SomeClass<int>'` with line number and everything. Now, I wonder 
why gcc doesn't do that for standard headers (at least not in my 
particular case). Is there a compelling argument for that? It feels like 
it must be intentional. If not, I would suggest adding that because I'm 
pretty sure it would have spared me a couple of hours just by realising 
quickly that it takes `std::array` from the wrong header.

Thank you in advance and keep up the great work!

Julian

Re: Reason for missing notes for incomplete-type errors in standard headers

[Jonathan Wakely]

On Fri, 5 Aug 2022 at 17:39, Julian Lenz via Gcc <gcc@gcc.gnu.org> wrote:
>
> Hi everybody,
>
> TL;DR:
>
> What is the reason that `error: '<some class>' has incomplete type` does
> not give a note about where the forward declaration happened for
> standard library classes?

Probably because the declaration happens in a system header.

Does -Wsystem-headers change it?

Re: Reason for missing notes for incomplete-type errors in standard headers

[Julian Lenz]

Hi Jonathan,

Thanks for the quick answer. No, it doesn't for the simplest possible case:

     #include <tuple>

     int main() { std::array<int,1> arr; };

But admittedly that would have been a surprise to me as I usually 
compile with -Wall -Wextra -Werror.

But I tried to replace `<array>` with `<tuple>` in yet another code 
(tiny but real-life, so not as trivial as above) with two appearances of 
`std::array` and there, one does yield the note.

I boiled it down to the following: The typename seems to not yield the 
note but the constructor does notify, i.e. in

     1 std::array<int,1> f() {

     2    return std::array<int,1>{{1}};

     3 }

Line 1 and 2 raise an error but only Line 2 comes with a note.

Best,

Julian


On 05/08/2022 17:55, Jonathan Wakely wrote:
> On Fri, 5 Aug 2022 at 17:39, Julian Lenz via Gcc <gcc@gcc.gnu.org> wrote:
>> Hi everybody,
>>
>> TL;DR:
>>
>> What is the reason that `error: '<some class>' has incomplete type` does
>> not give a note about where the forward declaration happened for
>> standard library classes?
> Probably because the declaration happens in a system header.
>
> Does -Wsystem-headers change it?