gh-129005: Remove copies from _pyio using take_bytes

[cmaloney]

Memory usage now matches that of _io for large files.

$ ./python -Wd -I -m test test_io.test_largefile -uall,largefile,walltime -M32G -v
== CPython 3.15.0a1+ (heads/gh129005_take_bytes-dirty:d8b88498ebf, Nov 13 2025, 15:43:51) [Clang 21.1.5 ]
== Linux-6.17.7-arch1-1-x86_64-with-glibc2.42 little-endian
== Python build: debug
== cwd: <build_dir>/build/test_python_worker_116910æ
== CPU count: 32
== encodings: locale=UTF-8 FS=utf-8
== resources: all
Using random seed: 365280675
0:00:00 load avg: 9.60 Run 1 test sequentially in a single process
0:00:00 load avg: 9.60 [1/1] test_io.test_largefile
test_large_read (test.test_io.test_largefile.CLargeFileTest.test_large_read) ... ... expected peak memory use: 2.3G
... process data size: 2.3G
ok
test_lseek (test.test_io.test_largefile.CLargeFileTest.test_lseek) ... ok
test_osstat (test.test_io.test_largefile.CLargeFileTest.test_osstat) ... ok
test_seek_read (test.test_io.test_largefile.CLargeFileTest.test_seek_read) ... ok
test_seek_readall (test.test_io.test_largefile.CLargeFileTest.test_seek_readall) ... ... expected peak memory use: 2.3G
... process data size: 2.3G
... process data size: 2.3G
ok
test_seekable (test.test_io.test_largefile.CLargeFileTest.test_seekable) ... ok
test_truncate (test.test_io.test_largefile.CLargeFileTest.test_truncate) ... ok
test_large_read (test.test_io.test_largefile.PyLargeFileTest.test_large_read) ... ... expected peak memory use: 2.3G
... process data size: 2.3G
ok
test_lseek (test.test_io.test_largefile.PyLargeFileTest.test_lseek) ... ok
test_osstat (test.test_io.test_largefile.PyLargeFileTest.test_osstat) ... ok
test_seek_read (test.test_io.test_largefile.PyLargeFileTest.test_seek_read) ... ok
test_seek_readall (test.test_io.test_largefile.PyLargeFileTest.test_seek_readall) ... ... expected peak memory use: 2.3G
... process data size: 2.3G
... process data size: 2.3G
ok
test_seekable (test.test_io.test_largefile.PyLargeFileTest.test_seekable) ... ok
test_truncate (test.test_io.test_largefile.PyLargeFileTest.test_truncate) ... ok
test_it (test.test_io.test_largefile.TestCopyfile.test_it) ... ok
test_it (test.test_io.test_largefile.TestSocketSendfile.test_it) ... ok
----------------------------------------------------------------------
Ran 16 tests in 6.950s
OK
0:00:06 load avg: 9.72 [1/1] test_io.test_largefile passed
== Tests result: SUCCESS ==
1 test OK.
Total duration: 7.0 sec
Total tests: run=16
Total test files: run=1/1
Result: SUCCESS

For reference; this test goes from 8.5sec overall duration on my AMD 64 bit Arch Linux box to 7 seconds.

• Issue: Reduce copies when reading files in pyio, match behavior of _io #129005

[cmaloney]

Optimization enabled by gh-139871 (Trying to keep that one issue referencing all the things which I use it for; planning to expand the What's New with a single "here's all the modules that got faster" list once optimize)

[vstinner]

Is there a difference between b.resize(n) and del b[n:]?

[cmaloney]

.resize() could grow while the del b[n:] never will. There's some code which does a = bytearray(b'12'); del a[3:] (intentionally del a slice past the end) as part of its buffer management. Generally I prefer resize because there's a lot less code involved than the slice code (and the slice code eventually calls resize anyways)

[vstinner]

LGTM