Reduce unnecessary work in FileIO #159446

Feature or enhancement

Reduce number of instructions run and runtime. This is generally removing unnecessary work such as a TypeError always being constructed when FileIO is given a string filename. A second method is reducing GIL acquisitions in common paths. Measuring with the callgrind tool on my x64 linux desktop on an LTO build FileIO(path).readall() executes 23% fewer instructions and runs over 5% faster reading small files in full with these changes:

• When FileIO is constructed on a filename string currently a TypeError is raised then caught because the constructor calls PyLong_AsInt then checks for an error rather than a fast inline _PyIndex_Check.
• In FileIO.readall() do a single long GIL release rather than reacquiring the GIL for each read(). PyErr_CheckSignals is still checked at the same frequency.

Has this already been discussed elsewhere?

This is a minor feature, which does not need previous discussion elsewhere

Links to previous discussion of this feature:

No response

Linked PRs

• gh-159446: Avoid TypeError on FileIO construction #159480
• gh-159446: Release the GIL less in FileIO open and readall #159482