Should more template arguments be propagated as members?

[steffenlarsen]

Most templated SYCL classes already expose a number of member aliases and constexpr variables for communicating meta information about the given instance of the classes. Examples of these are the member aliases value_type and allocator_type of sycl::buffer propagating the T and AllocatorT template parameters respectively and the constexpr member variable address_space of sycl::multi_ptr which propagates the value of the Space template parameter.

However, some of the template arguments are not propagated through such an interface. Examples of these are the AccessMode template parameter of sycl::accessor and Dimensions template parameter of sycl::accessor, sycl::buffer, sycl::nd_range, etc.

Having more of these as members could potentially simplify SFINAE and the use of if constexpr, for example if a user writes code specializing for a certain access mode of accessors. Should the specification extend the interface with more members like this?

[keryell]

Yes, of course. I have this as triSYCL extensions, like probably other implementations do.

[fraggamuffin]

regular template parameters are doable and easier to get from the class
parameter pack is harder
OK to draft