* Hoehle, Joerg-Cyril [2006-01-17 14:24+0100] writes:
Hi,
I thought about alternatives and rejected them. Probably this patch is just fine as is and it should not raise an error when Emacs uses UTF-8 or anything and clisp was only built with 8bit strings --without-unicode.
Hmm.. Emacs shouldn't use utf-8 if CLISP can't handle it. If Emacs encodes a string which includes non-ascii chars with utf-8 and, say, CLISP tries to decode it with iso-8859-1, it will lead to errors pretty quickly (because the packet length field isn't interpreted correctly and the payload of the packet will be truncated).
The user has to choose an encoding that actually works and it's better if we detect bad combinations as early as possible than to proceed as long as possible, because at that point we can no longer print reasonable error messages.
It would be nice if SLIME could ask CLISP what encodings are available and than switch to one of those. But that doesn't work in practice because not all Lisp implementations can switch the encoding of streams on the fly. (Closing and re-opening the connection with a different encoding sounds like a terrible solution too.)
2006-01-14 Joerg Hoehle hoehle@users.sourceforge.net
- swank-clisp (find-encoding): work in a 8bit clisp w/o #-UNICODE.
--- swank-clisp.lisp.orig 2005-11-12 00:43:43.000000000 +0100 +++ swank-clisp.lisp 2006-01-14 13:17:08.000000000 +0100 @@ -122,7 +122,9 @@ (:iso-latin-1-unix "iso-8859-1") (:utf-8-unix "utf-8") (:euc-jp-unix "euc-jp"))))
- (ext:make-encoding :charset charset :line-terminator :unix)))
- (ext:make-encoding :charset (if (member :unicode *features*)
charset :default):line-terminator :unix)))(defimplementation accept-connection (socket &key (external-format :iso-latin-1-unix)
The idea in the present patch is that (untested) the same set of slime.fas files might possibly work in both builds of clisp: with and without unicode. But you'd need to compile from the 8bit version.
Is "iso-8859-1" available in CLISPs --without-unicode ? If it is, then the unpatched fasl files should also work with both builds. But the user has to choose latin-1 on the Emacs side.
Helmut.