Speed up bytes.join() when all items are exact bytes #159191

Feature or enhancement

bytes.join() and bytearray.join() fill a Py_buffer and take a reference for every item, even when all items are exact bytes. With more than 10 items, that also means allocating the array of Py_buffers. On the free-threaded build the increfs are atomic when the items are shared between threads, and since gh-158803 they run while the list's lock is held.

When every item is exact bytes and the result is below the 1 MiB threshold at which join() releases the GIL, nothing can run Python code or suspend the critical section. So the items can be copied straight from the sequence, without a Py_buffer or a reference for each one. Other inputs keep the current path.

Release builds, b",".join(seq) with distinct 8-byte items:

Lists with non-bytes items take the general path, which can be a bit slower depending on the compiler's inlining (see the discussion in #159096).

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:

#158910, #159096

Linked PRs

• gh-159191: Add a fast path to bytes.join() for exact bytes items #159096