Do you have a microbenchmark for what specific thing is being optimized? I'd recommend using pyperf and posting the results to an issue at bugs.python.org.

From the Dev Guide:

• Do not do cosmetic changes to unrelated code in the same commit as some feature/bugfix.

It seems like most of the diff is filled with these. I think the general guideline is to keep the diff as small as possible. I would think that the PR should limit changing cosmetics (const qualifiers, delayed declarations, whitespace, braces, changing ifs to asserts) to the couple of functions that the PR is to focus on. Finally, the change to .gitignore looks unrelated to the suggestion.

Is there anything positive accomplished by moving behavior from public functions like PyDict_SetItem and PyDict_GetItemWithError into new identical static-scope functions? If anything, I think the existing usage of the public functions is good because it gives examples of how the API is to be used outside of this file.

@sweeneyde: I removed all the minor and cosmetic changes.

About PyDict_SetItem and the other one, I followed the example of setobject.c, where all the public functions calls the static internal functions. IMHO it's more clear. Furthermore, should it save you some nanoseconds (static objects are allocated in the stack)?

@Marco-Sulla Please create bpo issue and post benchmark&result to it.

@methane I already created it: https://bugs.python.org/issue41835