GzipFile.seek makes invalid write if buffer is not flushed in Python 3.12rc1

[effigies]

Bug report

Checklist

• I am confident this is a bug in CPython, not a bug in a third-party project
• I have searched the CPython issue tracker,
  and am confident this bug has not been reported before

CPython versions tested on:

3.12

Operating systems tested on:

Linux

Output from running 'python -VV' on the command line:

Python 3.12.0rc1 (main, Aug 16 2023, 05:03:59) [GCC 12.2.0]

A clear and concise description of the bug:

I have code that writes out sections of a data file in chunks, and uses seeks to ensure that the position is correct before writing.

In the following example, I write 5 bytes, seek to position 5 and write five more bytes. If I flush the buffer, the result is as expected. If I do not, 5 null bytes are written between the two groups of intended bytes.

#!/usr/bin/env python

import io
import gzip


for flush in (True, False):
    data = io.BytesIO()
    gzip_writer = gzip.GzipFile(fileobj=data, mode='wb')
    gzip_writer.write(b'abcde')

    # If the buffer isn't flushed, seek works from unchanged offset
    if flush and hasattr(gzip_writer, '_buffer'):
        gzip_writer._buffer.flush()

    gzip_writer.seek(5)
    gzip_writer.write(b'fghij')
    gzip_writer.close()
    
    # Recover result
    data.seek(0)
    gzip_reader = gzip.GzipFile(fileobj=data, mode='rb')
    result = gzip_reader.read()

    print(f'{flush=}: {result}')

In the case where I seek but don't tell, I get spurious \x00 bytes:

flush=True: b'abcdefghij'
flush=False: b'abcde\x00\x00\x00\x00\x00fghij'

Here is the output in Python 3.10.10:

flush=True: b'abcdefghij'
flush=False: b'abcdefghij'

Linked PRs

• gh-108111: Flush gzip write buffer before seeking, fixing bad writes #108341
• [3.12] gh-108111: Flush gzip write buffer before seeking, fixing bad writes (GH-108341) #108402

[effigies]

Looking at #101251, which is where I expect this was introduced, I suspect this is the fix:

     def seek(self, offset, whence=io.SEEK_SET):
         if self.mode == WRITE:
+            self._check_not_closed()
+            self._buffer.flush()
             if whence != io.SEEK_SET:
                 if whence == io.SEEK_CUR:

Here:

cpython/Lib/gzip.py

Lines 402 to 420 in daed54d

	def seek(self, offset, whence=io.SEEK_SET):
	if self.mode == WRITE:
	if whence != io.SEEK_SET:
	if whence == io.SEEK_CUR:
	offset = self.offset + offset
	else:
	raise ValueError('Seek from end not supported')
	if offset < self.offset:
	raise OSError('Negative seek in write mode')
	count = offset - self.offset
	chunk = b'\0' * self._buffer_size
	for i in range(count // self._buffer_size):
	self.write(chunk)
	self.write(b'\0' * (count % self._buffer_size))
	elif self.mode == READ:
	self._check_not_closed()
	return self._buffer.seek(offset, whence)
	return self.offset

[hugovk]

Thank you!