StreamResponse + HTTP/1.0 = corrupted file #476
Description
I'm using StreamResponse to serve files. My code is
from aiohttp import web
response = web.StreamResponse()
response.enable_chunked_encoding()
response.content_length = statinfo.st_size
response.content_type = 'application/octet-stream'
response.start(request)
with open(filename, 'rb') as f:
while True:
chunk = f.read(CHUNK_SIZE)
if not chunk:
return response
response.write(chunk)
yield from response.drain()
If I'm using wget to download the file then it works beautifully. But when I'm using Python urllib to download the file, my download is corrupted. The difference seems to be that wget can HTTP/1.1, but urllib only does HTTP/1.0. The network trace is
Host: ...
User-Agent: Python-urllib/1.17
HTTP/1.0 200 OK
CONTENT-LENGTH: 815206
CONTENT-TYPE: application/octet-stream
TRANSFER-ENCODING: chunked
CONNECTION: close
DATE: Thu, 27 Aug 2015 17:31:25 GMT
SERVER: Python/3.4 aiohttp/0.17.2
10000
... binary data ...
and Python saves everything starting at (and including) the 10000 into the file.
I think this is because TRANSFER-ENCODING: chunked is HTTP/1.1 only, and a HTTP/1.0 client is entirely correct in ignoring it. Aiohttp should not send chunk sizes to HTTP/1.0 clients as they almost certainly end up in the data stream.