I'm looking at the pytorch codebase, and I see multiple anonymous namespaces in the same scope like https://github.com/pytorch/pytorch/blob/master/aten/src/ATen/native/quantized/cpu/qpool.cpp#L23 and https://github.com/pytorch/pytorch/blob/master/aten/src/ATen/native/quantized/cpu/qpool.cpp#L226
What is the point of this design? Couldn't we just combine them?
Couldn't we just combine them?
All unnamed namespaces in one parent namespace and within the same translation unit are the same namespace. They can be combined or separated just like named namespaces can be.
What is the point of this design?
Since I'm not the author of the linked code, I cannot tell why they wrote the program like they did. It could be a design choice, aesthetic preference, an incidental artifact of refactoring or any other reason.
But I can describe an advantage. It is difficult for a human programmer to perceive vertically long scopes (at least without aid of a visualisation tool that an IDE may have). At a glance, when you are in middle of a long file, it's impossible to tell whether you are inside the scope that was started at the beginning of the file or perhaps some other scope. It can be convenient to be able to scroll up only a few lines to find the beginning of the scope. Splitting namespaces into smaller pieces allows that.
This applies to namespaces in general; not only to unnamed namespaces.
