A lot of tests verify the client sends something like this (borrowed from test 221):

<verify>
<strippart>
s/^Accept-Encoding: .*/Accept-Encoding: xxx/
</strippart>
<protocol>
GET /%TESTNUMBER HTTP/1.1
Host: %HOSTIP:%HTTPPORT
User-Agent: curl/%VERSION
Accept: */*
Accept-Encoding: xxx

</protocol>
<errorcode>
61
</errorcode>
</verify>

The client actually sends Accept-Encoding: deflate, gzip[CR][LF]. This gets replaced by the <strippart> bit to Accept-Encoding: xxx[LF] - the .* matches, and removes, the carriage return. However, the <protocol> section compensates for this by omitting the carriage return from the expected client message, only on that Accept-Encoding line. I've taken a photo to make this clear - the ^M is a carriage return.

This is mostly ok, except that runtests.pl then calls subNewlines on Accept-Encoding: xxx[LF], which restores the carriage return and fails the test, for Hyper specifically.

I think the fix is to make the strippart matching better - match on a space character, a comma, or [a-zA-Z], but don't match on carriage returns, and then modify the <protocol> bits to include the carriage return, which is what actually gets sent by the client.