Comment 2 for bug 2025748

It seems that python3-ubuntutools switched from using urllib.request.urlopen() to requests.get() when fetching sources using the pull() method. For the sake of providing a progress bar, both the old code and the new code "streamed" the data by asking for chunks at a time, incrementally saving those to disk, and then later checking the size and checksum. But the switch to requests now inadvertently _decompresses_ files, in the case for example of a .diff.gz. The uncompressed data gets written to a file named something.diff.gz, which of course is now larger than expected.

For validation purposes we require the original unmodified file to be written to disk.

I don't see a way of getting the requests library to handle transfer encoding but not content encoding without using undocumented API. As far as I can tell, the raw attribute would give me neither.

Perhaps "Accept-Encoding: identity" in the request would help? Would that be correct in all cases?