[tsayin4]

The buffer that a TextIOWrapper sits on top of reads ahead, so after a read its
position is not the position of the text stream. truncate() with no argument passed
nothing through to the buffer, so the stream was resized to the read-ahead position:

with open("f", "w") as f:
    f.write("abcde")

f = open("f", "r+")
f.read(2)
f.tell()        # 2
f.truncate()    # 5 -- the file is left untouched

With a file bigger than one chunk it truncates to 8192 instead. io.BufferedRandom
gets this right, and so does _pyio.TextIOWrapper, which passes tell() to the
buffer. The test added here fails on the C implementation and passes on the Python one
before the fix, so no _pyio change is needed -- this only brings _io into line.

Refs gh-161275.

🤖 Generated with Claude Code

[python-cla-bot]

The following commit authors need to sign the Contributor License Agreement:

• hilmitunasayin@icloud.com

[cmaloney]

Please read and review the https://devguide.python.org/, in particular the parts around using AI and the lifecycle of a PR (https://devguide.python.org/getting-started/pull-request-lifecycle/). Don't force push to branches which are in Python PRs: https://devguide.python.org/getting-started/pull-request-lifecycle/#don-t-force-push

[tsayin4]

That force push was mine and it won't happen again. I was fixing an unresolvable
:meth: target in the NEWS entry and amended the commit instead of adding one. I've read
the AI tools page and the PR lifecycle page now.

On disclosure: I use Claude Code, and it's noted at the bottom of the PR description. The
bug came out of diffing _io against _pyio behaviour; I reproduced it on 3.10, 3.11 and
3.13 and read the code on main before writing anything. truncate() hands pos straight
to the buffer, and after a read the buffer sits wherever its read-ahead ended, so on a
5 byte file read(2) then truncate() leaves the file at 5. _pyio passes tell()
through, which is why the test fails on _io and passes on _pyio untouched.

Happy to go through any of it. Also, if you'd rather I close the two _pyio PRs while
steering-council#372 is open, say so and I'll close them.