I'm trying to use importC to help automate .di and targeted library production (for the Nvidia libs that dcompute taps into).

As you'll note if you try this yourself, the -Hf output of an importC run looks pretty interesting but has some, obvious, systemic problems.  For example:

  typedef SomeOpaqueStructYoureNotSupposedToSeeInside *HandleT;
becomes
  alias HandleT = struct SomeOpaqueStructYoureNotSupposedToSeeInside*;

I wondered if others have experience generating bindings with importC and suggestions they'd care to share.

I will note that I've had some luck using importC directly, with a little fiddling, but it looks like it could also be a supported/robust 99%+ binding generator going forward with some work.

On Wednesday, 6 September 2023 at 00:53:40 UTC, Bruce Carneal wrote:
> I'm trying to use importC to help automate .di and targeted library production (for the Nvidia libs that dcompute taps into).
>
> As you'll note if you try this yourself, the -Hf output of an importC run looks pretty interesting but has some, obvious, systemic problems.  For example:
>
>   typedef SomeOpaqueStructYoureNotSupposedToSeeInside *HandleT;
> becomes
>   alias HandleT = struct SomeOpaqueStructYoureNotSupposedToSeeInside*;
>
> I wondered if others have experience generating bindings with importC and suggestions they'd care to share.
>
> I will note that I've had some luck using importC directly, with a little fiddling, but it looks like it could also be a supported/robust 99%+ binding generator going forward with some work.

Sorry, typo, left out "struct" in the C typedef:
  typedef struct SomeOpaqueStructYoureNotSupposedToSeeInside *HandleT;

On Wednesday, 6 September 2023 at 00:55:57 UTC, Bruce Carneal wrote:
> On Wednesday, 6 September 2023 at 00:53:40 UTC, Bruce Carneal wrote:
>> ...
>> I will note that I've had some luck using importC directly, with a little fiddling, but it looks like it could also be a supported/robust 99%+ binding generator going forward with some work.
>

Here "some work" may actually be a Herculean challenge but I think not since the errors all seem to be of the form: "the information is in the compiler but was not processed properly for -Hf".

In any event, commentary from those in the know would be most appreciated.

On Wednesday, 6 September 2023 at 01:14:47 UTC, Bruce Carneal wrote:
> On Wednesday, 6 September 2023 at 00:55:57 UTC, Bruce Carneal wrote:
>> On Wednesday, 6 September 2023 at 00:53:40 UTC, Bruce Carneal wrote:
>>> ...
>>> I will note that I've had some luck using importC directly, with a little fiddling, but it looks like it could also be a supported/robust 99%+ binding generator going forward with some work.
>>
>
> Here "some work" may actually be a Herculean challenge but I think not since the errors all seem to be of the form: "the information is in the compiler but was not processed properly for -Hf".
>
> In any event, commentary from those in the know would be most appreciated.

I am using this process to update the ODBC bindings in etc.c.odbc, however, I currently have two active bugs against ImportC because when I try to generate a DI binding from an H file that has an tagged enum in it I get a segfault. But if you comment out the offending bits it actually works pretty much as you'd expect.

On Sunday, 10 September 2023 at 01:45:58 UTC, Adam Wilson wrote:
> On Wednesday, 6 September 2023 at 01:14:47 UTC, Bruce Carneal wrote:
...
>>> On Wednesday, 6 September 2023 at 00:53:40 UTC, Bruce Carneal wrote:
>>>> ...
>>>> I will note that I've had some luck using importC directly, with a little fiddling, ...
>>
>> Here "some work" may actually be a Herculean challenge but I think not since the errors all seem to be of the form: "the information is in the compiler but was not processed properly for -Hf".
>>
>> In any event, commentary from those in the know would be most appreciated.
>
> I am using this process to update the ODBC bindings in etc.c.odbc, however, I currently have two active bugs against ImportC because when I try to generate a DI binding from an H file that has an tagged enum in it I get a segfault. But if you comment out the offending bits it actually works pretty much as you'd expect.

Good to hear.  I'll look at the -Hf path again down the road but went back to using importC directly for #include <cuda.h> bindings.  Everything works AFAICT from my testing apart from importC ignoring #defines that should be function aliases for functions being declared elsewhere in the file, as in:

#define someStableFuncName actualFuncNameOfFunctionDefinedInThisFile
...

I'm not sure where I should report this "bug" in importC or even if it is considered a bug, but I've worked around it with a simple helper program, importC having done the heavy lifting.

On Sunday, 10 September 2023 at 05:15:59 UTC, Bruce Carneal wrote:
> I'm not sure where I should report this "bug" in importC or even if it is considered a bug, but I've worked around it with a simple helper program, importC having done the heavy lifting.

I would file the report. Walter is looking for bugs in ImportC and either it's already been fixed and it's not an issue to file a dup, or it hasn't and Walter can make it work right.