Well, I guess one of my motivations for opening this thread is to ask whether the kind of usage I’m talking about above is really that crucial or common. It’s really most vital in cases where you want to do random access on very large buffers such as with MMap. (Though I’m still very confused about why the performance seems to matter much… any CPU time should be overwhelmed by actual hardware IO time, but in casual observation it still seemed to matter. I suppose this will come a bigger deal in the future as we see more of things like NVMe and Optane. I really should do a comprehensive benchmark of memory mapped data…)
I would think that IPC would be the most performance critical case, and I’m imagining that the devs might say that we should use read. However, I would think that even in those cases reinterpret and the like would open up a lot of design options and you wouldn’t want to be limited to read because it requires being extremely selective about when to copy data.
It might be helpful if you give some examples. I’m pretty new to this stuff so, like I said above, not too sure if what I’m talking about is a universal requirement.