Date: Thu, 12 Nov 2026 15:01:25 -0700
Hello,
I have explored this matter on this list before, and as I understand
it, there is 0 possibility that all the compiler vendors are going to
commit to anything at the ABI level. And it only really takes one of
the compiler vendors to kill a proposal. So ABI proposals seem to be a
non-starter.
Therefore, I am not trying to suggest anything at the ABI level. If
you want to take the logical next steps with me and talk about
bindings generation, I did just post a whole website spelling out how
that could go. But it was intentionally not part of the proposal. (And
yes, there is a small annoyance with name mangling in there.)
There is also a conversation that needs to happen with independent
vendors of bindings generators, as well, as I understand it. I'll have
to talk to them too. They are stakeholders in a different way.
On Thu, Nov 12, 2026 at 12:59 PM Tiago Freire <tmiguelf_at_[hidden]> wrote:
>
> I think you need to be a little bit more specific with the proposal.
> My understanding of language interop is "it's a dll/so and we agree at the elf/os level what names belong to what function and then we also agree at the machine instruction what is the calling convention, what is the structure of the data being passed (ABI).
>
> It's a difficult proposal since the language has intentional steered away from defining since the details are dependent on other systems not directly in the scope of the C++ language.
>
> Doing it as attributes is not the way the go, as it is my understanding that those are optional and compilers are free to ignore.
>
> The expressivity of the solution is also insufficient, and it will fall into the same pitfalls that make the status quo utterly broken and subject to endless discussions.
>
> For example, you append a [[=std::interop]] to a function with different overloads, now you are kind of stuck with having to define some name mangling rules in order to avoid name colisions.
>
> If instead we define a different type of expression to express an export table that forces to explicitly define what functions overloads get exported and what names they get as well as other optional details like a calling convention (i.e. export table) then this is not even a problem. The linker would be able to resolve all the necessary details.
> And the language doesn't even need to know what other types of language exist as long as it is documented somewhere that X language is using Y convention.
>
> And if you have an export table you will also need an import table, which can be combined with the same expression as the export one. But an import expression would also be able to solve another problem, which is the ability to explicitly define from which dll/so from where those symbols should be loaded as opposed to the status quo of having the linker trying to figure out on its own by searching every single library passed to it.
>
> What do you think?
>
>
>
> ________________________________
> From: Std-Proposals <std-proposals-bounces_at_[hidden]> on behalf of A Johnston via Std-Proposals <std-proposals_at_[hidden]>
> Sent: Wednesday, November 11, 2026 7:40:24 AM
> To: std-proposals_at_[hidden] <std-proposals_at_[hidden]>
> Cc: A Johnston <ajohnston54637_at_[hidden]>
> Subject: [std-proposals] [[=std::interop]] An Annotation for Interoperation with Targets
>
> Hello,
>
> This is a first draft of a proposal for [[=std::interop]], a standard
> annotation marking a declaration for exposure to a foreign execution
> environment. There is an optional facility to build up lists of
> targets. E.g. [[=std::interop("Python", "Rust")]].
>
> It is intended for use by binding generation tools, and there is a
> reference code generator with bindings generated from C++ in plain C,
> JavaScript (ECMAScript in the proposal), Python and Rust. The code
> generator is just example code, but the outputs would be worth looking
> at if you are curious about language bindings.
>
> I have not inquired about a proposal number, and am looking for early
> feedback on the proposal.pdf first.
>
> The proposal.pdf is here:
> https://whatchamacallem.github.io/std_interop/proposal.pdf
>
> And the website with the source code, generated code and tests is
> here: https://whatchamacallem.github.io/std_interop/index.html
>
>
> Regards,
>
> Adrian
> --
> Std-Proposals mailing list
> Std-Proposals_at_[hidden]
> https://lists.isocpp.org/mailman/listinfo.cgi/std-proposals
>
I have explored this matter on this list before, and as I understand
it, there is 0 possibility that all the compiler vendors are going to
commit to anything at the ABI level. And it only really takes one of
the compiler vendors to kill a proposal. So ABI proposals seem to be a
non-starter.
Therefore, I am not trying to suggest anything at the ABI level. If
you want to take the logical next steps with me and talk about
bindings generation, I did just post a whole website spelling out how
that could go. But it was intentionally not part of the proposal. (And
yes, there is a small annoyance with name mangling in there.)
There is also a conversation that needs to happen with independent
vendors of bindings generators, as well, as I understand it. I'll have
to talk to them too. They are stakeholders in a different way.
On Thu, Nov 12, 2026 at 12:59 PM Tiago Freire <tmiguelf_at_[hidden]> wrote:
>
> I think you need to be a little bit more specific with the proposal.
> My understanding of language interop is "it's a dll/so and we agree at the elf/os level what names belong to what function and then we also agree at the machine instruction what is the calling convention, what is the structure of the data being passed (ABI).
>
> It's a difficult proposal since the language has intentional steered away from defining since the details are dependent on other systems not directly in the scope of the C++ language.
>
> Doing it as attributes is not the way the go, as it is my understanding that those are optional and compilers are free to ignore.
>
> The expressivity of the solution is also insufficient, and it will fall into the same pitfalls that make the status quo utterly broken and subject to endless discussions.
>
> For example, you append a [[=std::interop]] to a function with different overloads, now you are kind of stuck with having to define some name mangling rules in order to avoid name colisions.
>
> If instead we define a different type of expression to express an export table that forces to explicitly define what functions overloads get exported and what names they get as well as other optional details like a calling convention (i.e. export table) then this is not even a problem. The linker would be able to resolve all the necessary details.
> And the language doesn't even need to know what other types of language exist as long as it is documented somewhere that X language is using Y convention.
>
> And if you have an export table you will also need an import table, which can be combined with the same expression as the export one. But an import expression would also be able to solve another problem, which is the ability to explicitly define from which dll/so from where those symbols should be loaded as opposed to the status quo of having the linker trying to figure out on its own by searching every single library passed to it.
>
> What do you think?
>
>
>
> ________________________________
> From: Std-Proposals <std-proposals-bounces_at_[hidden]> on behalf of A Johnston via Std-Proposals <std-proposals_at_[hidden]>
> Sent: Wednesday, November 11, 2026 7:40:24 AM
> To: std-proposals_at_[hidden] <std-proposals_at_[hidden]>
> Cc: A Johnston <ajohnston54637_at_[hidden]>
> Subject: [std-proposals] [[=std::interop]] An Annotation for Interoperation with Targets
>
> Hello,
>
> This is a first draft of a proposal for [[=std::interop]], a standard
> annotation marking a declaration for exposure to a foreign execution
> environment. There is an optional facility to build up lists of
> targets. E.g. [[=std::interop("Python", "Rust")]].
>
> It is intended for use by binding generation tools, and there is a
> reference code generator with bindings generated from C++ in plain C,
> JavaScript (ECMAScript in the proposal), Python and Rust. The code
> generator is just example code, but the outputs would be worth looking
> at if you are curious about language bindings.
>
> I have not inquired about a proposal number, and am looking for early
> feedback on the proposal.pdf first.
>
> The proposal.pdf is here:
> https://whatchamacallem.github.io/std_interop/proposal.pdf
>
> And the website with the source code, generated code and tests is
> here: https://whatchamacallem.github.io/std_interop/index.html
>
>
> Regards,
>
> Adrian
> --
> Std-Proposals mailing list
> Std-Proposals_at_[hidden]
> https://lists.isocpp.org/mailman/listinfo.cgi/std-proposals
>
Received on 2026-11-12 22:01:56
