GZipDecoder Dropping Chunks

[aaronst]

My team has identified an issue with httpx._decoders.GZipDecoder that results in only the first chunk of data being decompressed and returned by Response.iter_bytes(). To reproduce this, we dumped the stream into a list chunks and wrote the following:

from httpx._decoders import GZipDecoder
from urllib3.response import GzipDecoder

chunks = [...]

httpx_decoder = GZipDecoder()
httpx_decoded = b""

urllib3_decoder = GzipDecoder()
urllib3_decoded = b""

for chunk in chunks:
    httpx_decoded += httpx_decoder.decode(chunk)
    urllib3_decoded += urllib3_decoder.decompress(chunk)

print(f"encoded: {len(b''.join(chunks))}")
print(f"httpx: {len(httpx_decoded)}")
print(f"urllib3: {len(urllib3_decoded)}")

"""
➜ python gzip_decoding_poc.py
encoded: 145314
httpx: 580
urllib3: 398739
"""

For httpx, the first chunk decodes fine but the rest decode to an empty byte-string. This doesn't throw any errors, the result is just a much smaller body returned from any of the Response methods, except for .iter_raw() since it doesn't decode.
I can't share the actual chunks here, the data comes from a paid third-party service, but can do so privately as needed. It's not clear to us if the encoding sent from the server is "bad", but urllib3 handles it just fine and it would be awesome if httpx could too!

[aaronst]

Digging further into urllib3, I found this PR. It seems like the problem might be that httpx doesn't currently handle gzip payloads that contain multiple members?

[aaronst]

@tomchristie since I saw another thread on GZipDecoder.