A co-worker of mine had some problems today using Drakma to POST a STRING containing non-ASCII characters encoded as UTF-8. He was doing something equivalent to:
(http-request "http://zappa.com/favicon.ico" :method :post :content (concatenate 'string "hello" (string #\white_square) "world") :external-format-out :utf-8)
which ends up setting the Content-Length header value to 11, which is the LENGTH in characters of the string being sent.
The documentation says that "if content is a sequence, Drakma will use LENGTH to determine its length and will use the result for the 'Content-Length' header sent to the server," so Drakma is working as documented.
Also, I think I understand why this behavior is the default. You don't want to scan a content string in order to determine what its length will be in octets once it has been encoded. My co-worker could have used a vector of type (unsigned-byte 8) to hold his UTF-8 encoded data.
However, I think the current Drakma behavior may be a mistake. People who want high performance are probably manipulating encoded strings as vectors of (unsigned-byte 8). It's the casual users, those sending strings, who are the ones most likely to be bitten by the default behavior, but only when their strings contain non-ASCII characters.
bob