Python 3.15: What's New, Release History & Support Lifecycle

What Is New in Python 3.15

Python 3.15 makes UTF-8 the default I/O encoding on all platforms (ending the long-standing platform-dependent behavior), ships a built-in high-frequency sampling profiler, introduces the PyBytesWriter C API for efficient buffer building, and improves AttributeError suggestions across nested members. Several long-deprecated items are removed.

What Does Defaulting to UTF-8 Mean in Practice?

Before Python 3.15, open("file.txt") used the platform's locale encoding -- UTF-8 on most Linux systems, but often Latin-1, CP1252, or another encoding on Windows or older macOS. This caused files written on one system to be read incorrectly on another depending purely on locale settings. PEP 686 makes UTF-8 the default everywhere when no encoding= argument is given.

# Before 3.15: encoding depends on platform locale
with open("data.txt") as f:  # could be UTF-8, Latin-1, CP1252...
    content = f.read()

# Python 3.15+: always UTF-8
with open("data.txt") as f:  # guaranteed UTF-8
    content = f.read()

# If you truly need the platform encoding:
import locale
with open("data.txt", encoding=locale.getpreferredencoding(False)) as f:
    content = f.read()

This breaks code that wrote Latin-1 or CP1252 encoded files and reads them back without specifying encoding=. The fix is explicit: always pass encoding="latin-1" (or whatever the file actually is). Code that already specifies encoding= is unaffected.

How Does the Sampling Profiler Work?

The new profiling.sampling module is a statistical profiler that interrupts execution at a configurable frequency (up to ~1,000,000 Hz) to record the current call stack. Unlike cProfile (which instruments every function call), sampling adds minimal overhead -- typically under 1% -- and can attach to a running process without restarting it.

import profiling.sampling as sp

profiler = sp.Sampler(interval=0.001)  # 1ms = 1000 Hz
profiler.start()
run_workload()
profiler.stop()

stats = profiler.get_stats()
stats.print_top(20)  # top 20 hotspots by sample count

Sampling profilers are the right tool for production profiling where cProfile's call-count instrumentation overhead is unacceptable. The profiler works with both GIL and free-threaded builds, and with the JIT enabled.

What Is the PyBytesWriter C API?

The PyBytesWriter API (PEP 782) provides C extensions with a way to incrementally build a bytes object without multiple allocations and copies. The old pattern was to allocate a large buffer, fill it, then resize -- or use PyBytes_FromStringAndSize on a complete buffer. PyBytesWriter manages a growable buffer internally and produces the final bytes object in one step.

// C extension code (conceptual)
PyBytesWriter *writer = PyBytesWriter_Create(1024);
PyBytesWriter_WriteBytes(writer, header, header_len);
PyBytesWriter_WriteBytes(writer, payload, payload_len);
PyObject *result = PyBytesWriter_Finish(writer);  // single allocation

This is relevant for C extension authors writing serializers, encoders, or protocol packers -- not for Python-level code, where bytearray and bytes.join() serve the same purpose.

What Was Removed or Deprecated in Python 3.15?

• Old NamedTuple keyword-argument syntax (using NamedTuple("Name", field=type) instead of class form) is removed -- use the class-based form: class Point(NamedTuple): x: int; y: int.
• Legacy zipimport methods that were already deprecated are removed.
• collections.abc.ByteString is removed from collections.abc.__all__ (was deprecated since 3.12) -- use Union[bytes, bytearray, memoryview] explicitly.
• Various undocumented C API functions replaced by the newer stable ABI equivalents.
• RLock() will silently ignore any positional/keyword arguments -- warnings issued in 3.14 become enforced in 3.15+.