io.TextIOWrapper.read does not flush the underlying write buffer when given a size argument

[blhsing]

Bug report

Bug description:

As reported by the StackOverflow question:
https://stackoverflow.com/questions/76142400/python-file-write-stuck-in-append-mode-when-read-has-byte-count-as-parameter

In the code below, when f.read is not given an argument, the pending content in the write buffer gets flushed so the output is ***s is a line as expected.

with open("new.txt", 'w+') as f:
    f.write("this is a line")

with open("new.txt", 'r+') as f:
    f.write("***")
    f.read()  # << note here
    f.seek(0)
    print(f.read()) # outputs: ***s is a line

But when f.read is given a size argument, it is apparent that the underlying write buffer is not flushed immediately, causing the pending content to be written at the end of the file, rendering an output of this is a line*** instead:

with open("new.txt", 'w+') as f:
    f.write("this is a line")

with open("new.txt", 'r+') as f:
    f.write("***")
    f.read(1)  # << note here
    f.seek(0)
    print(f.read()) # outputs: this is a line***

This issue occurs only with the C implementation of io.TextIOWrapper since the code above would work as expected if it is prepended with:

from _pyio import open

CPython versions tested on:

3.8, 3.11

Operating systems tested on:

Linux, Windows

Linked PRs

• gh-115059: Flush the underlying write buffer in io.BufferedRandom.read1() #115163
• [3.11] gh-115059: Flush the underlying write buffer in io.BufferedRandom.read1() (GH-115163) #115206
• [3.12] gh-115059: Flush the underlying write buffer in io.BufferedRandom.read1() (GH-115163) #115205
• gh-115059: Remove debugging code in test_io #115240
• [3.12] gh-115059: Remove debugging code in test_io (GH-115240) #115244
• [3.11] gh-115059: Remove debugging code in test_io (GH-115240) #115245

[serhiy-storchaka]

It is because the C implementation of io.TextIOWrapper.read() uses the read1() method of the underlying binary stream if it is available and the size is specified, and the C implementation of io.BufferedRandom.read1() did not flush the write buffer. #115163 fixes the latter.

[serhiy-storchaka]

Thank you for your report @blhsing. The issue is now fixed.