`gzip._GzipReader.read()` with sometimes uninitialized variable

[maurycy]

Bug report

Bug description:

This one is easy to spot but hard to reproduce:

cpython/Lib/gzip.py

Line 589 in d7dbde8

if buf == b"":

buf is initialized only when self._decompressor.needs_input:

cpython/Lib/gzip.py

Line 577 in d7dbde8

buf = self._fp.read(READ_BUFFER_SIZE)

It exists since gh-95534 (#97664).

I spent a lot of time trying to reproduce this bug using the standard zlib decompressor without any success:

import gzip
import io

class EvilDecomp:
    def __init__(self, **kwargs):
        self.needs_input = False
        self.unused_data = b""
        self.eof = False
    def decompress(self, data, max_length):
        self.needs_input = True
        return b""

gzip.zlib._ZlibDecompressor = EvilDecomp

with gzip.GzipFile(
    fileobj=io.BytesIO(gzip.compress(b"hello world")), mode="rb"
) as f:
    f.read(1)

results in:

22:16:40.678647000PM CEST maurycy@gimel /Users/maurycy/src/cpython % ./python.exe meow.py
Traceback (most recent call last):
  File "/Users/maurycy/src/cpython/meow.py", line 18, in <module>
    f.read(1)
    ~~~~~~^^^
  File "/Users/maurycy/src/cpython/Lib/gzip.py", line 349, in read
    return self._buffer.read(size)
           ~~~~~~~~~~~~~~~~~^^^^^^
  File "/Users/maurycy/src/cpython/Lib/compression/_common/_streams.py", line 68, in readinto
    data = self.read(len(byte_view))
  File "/Users/maurycy/src/cpython/Lib/gzip.py", line 589, in read
    if buf == b"":
       ^^^
UnboundLocalError: cannot access local variable 'buf' where it is not associated with a value

CPython versions tested on:

CPython main branch

Operating systems tested on:

macOS

Linked PRs

• gh-137571: Fix UnboundLocalError in gzip._GzipReader.read() #137572

[ZeroIntensity]

Just for clarification, does this cause an actual problem in practice? The fix seems simple enough that it's probably worth doing, but I want to make sure this isn't just from fuzzing the private API.

[maurycy]

@ZeroIntensity Nope! Neither. I was reviewing various modules to see how similar are various streams across the stdlib. Many classes provide streaming, but they employ different strategies (some copy, some use memoryview etc.), and I'm curious if there's any opportunity to contribute. That's how I found gh-137524 also.