On emscripten, if the request is read in chunks, sometimes it fails to read the whole amount. This is because the code in response.py assumes that if a read() call to the underlying request stream doesn't fill the buffer it is given, this must be the last buffer in the stream, which isn't the case sometimes, because the chunks we read are those given by javascript, which may be of arbitrary size.

I have a fix, which I'll make a PR for in a mo.