Then of course, using transfer-encoding for anything but chunked is very rarely used. And never by browsers for example.

But I will agree that the RFC says what you're saying. Maybe I should take this to some HTTP peeps.

... the reason for this is that the Content-Length is the size of the compressed data, which means that when the transfer decoding changes the size, the amount may/will differ from the given size in the header.

Also detailed in the Content-Length section.

I could perhaps add that when I added transfer-encoding support to curl, lots of years ago, the (live) server end we used provided content-lengths the way curl now supports. Now I need to ponder on how to proceed from here.

Thanks a lot @seanmonstar for your quick answers.

FYI: the next step went here: curl/curl#7643