`gzip.GzipFile` creates reference cycle that requires a deep garbage collection cycle to cleanup.

[fowczarek]

Bug report

Bug description:

During debugging memory buildup, I've noticed gzip.GzipFile holds a reference to itself (Cycle: GzipFile._buffer->BufferedWriter._raw->_WriteBufferStream.gzip_file-> GzipFile). This cycle prevents memory from being freed until the garbage collector runs a deep cleanup cycle (generation=2).

Steps to reproduce

1. Disable garbage collection temporarily to make sure we are the ones who catch it
2. Set the garbage collector's debug level to DEBUG_LEAK
3. Open GzipFile.
4. Force garbage collection and look at its output

import gc
import gzip
import io

gc.collect()
gc.disable()
gc.set_debug(gc.DEBUG_LEAK)

with io.BytesIO() as buffer:
    with gzip.GzipFile(mode="wb", fileobj=buffer):
        pass

gc.collect()
gc.set_debug(0)

Potential solution

class _WriteBufferStream(io.RawIOBase):
    ...

    def __del__(self):
        del self.gzip_file

CPython versions tested on:

3.12

Operating systems tested on:

macOS

Linked PRs

• gh-129640: Add tests for reference cycle in gzip.GzipFile #130916

[cmaloney]

I think this is the equivalent to gh-129726? (Hadn't seen this issue until just now). If so, has a fix which has been backported to 3.12. Similar concept, but broke refloop by adding a weakref (PR: #130055)

[Mr-Sunglasses]

This Issue is fixed in (PR: #1300550) by @cmaloney . The (PR: #130916) add's some tests for it.

[hauntsaninja]

Additional test PRs still welcome, but going ahead and closing this since the issue is resolved by #130055 (and backported)