Fix the clang issue due to empty structure in the Tracing Layer

[pratikbariintel]

Use void* instead of empty zer_get_default_context_params_t structure for parameter-less callbacks to fix Clang compilation.

Related-To: NEO-17790

Signed-off-by: Pratik Bari pratik.bari@intel.com

[joshuaranjan]

Why do we need void * params.
If empty params, if possible, lets skip this member variable

[pratikbariintel]

Currently all the callbacks defined have the same pattern. In case this field is not required we have to define a new callback pattern.

[rwmcguir]

So this data type is used in the exported function call:

ZE_APIEXPORT ze_result_t ZE_APICALL
zelTracerGetDefaultContextRegisterCallback(
zel_tracer_handle_t hTracer,
zel_tracer_reg_t callback_type,
zer_pfnGetDefaultContextCb_t pfnGetDefaultContextCb
);

So while it won't break immediate compilation or exported function ABI, the public struct type zer_pfnGetDefaultContextCb_t is changing in it's struct members. Thinking through this out loud.. The struct remains the same size, and void* also remains the same size.

The problem is the callback itself, the user must typecast it to match our callback data type, i.e. you see this yourself in the test content. It breaks how the user must define the actual callback API. This is my fear that, yes, this breaks ABI compatibility. Can instead of going this route, perhaps insert a placeholder void member; or something into the actual data structure, to avoid the clang warning?

[joshuaranjan]

Can instead of going this route, perhaps insert a placeholder void member; or something into the actual data structure, to avoid the clang warning?

That would change the size of the structure ? Which might again be an ABI issue.

[joshuaranjan]

LGTM
Please request Vishnu to review

[vishnu-khanth]

The function pointer signature is changed in tracing header.
Just thinking about whether this breaks backward compatibility