Patch: Promoting characters to 31 bits (for discussion)

Stephen> The appended patch swaps the Lisp integer with the Lisp Stephen> character representations, so that characters have 31 Stephen> bits of precision. [...] Stephen> Second, I tracked down all uses of Lisp_Type_* and am Stephen> fairly confident that these are correct in the patch. I Stephen> have this nagging feeling I'm missing something that I Stephen> should be verifying, though. OK, one obvious thing I missed is use of the VALBITS macros. I'll chase those down. Also, Morioka-san mentioned `char-int' will break. Actually, this isn't true, in the sense that we've never had characters as big as 29 bits so _all_ existing uses are still available. The main problem I can see is that someone might have a huge "character set" with integer-like properties that they would want to put into UCS-4 private space. Not likely, IMO. This set would have more than 65536 members, otherwise they can use private space planes in Group 0, well below the 29-bit barrier. For compatibility, I propose using a hack like if ((unsigned int) (ch >> 1) < 2**29) make-int (ch >> 1) else XCONS (make-int ((ch >> 1) & (2**29 - 1)), make-int (ch >> 30)) to implement char-int, and convert to bignums when we get them. (Maybe that should throw an exception, instead?) For the foreseeable future, only people using private space in Groups 60 to 7F will have to be concerned about it; we've barely begun thinking about Planes 2 and 3, let alone Group 1 AFAIK. (We might put the entire 19-bit Mule character set up in private space, as a compatibility hack---but we'd want to do that in C, not in Lisp.) -- University of Tsukuba Tennodai 1-1-1 Tsukuba 305-8573 JAPAN Institute of Policy and Planning Sciences Tel/fax: +81 (298) 53-5091 __________________________________________________________________________ __________________________________________________________________________ What are those two straight lines for? "Free software rules."