From: Dan C. <dan...@co...> - 2005-11-14 16:25:22
|
Marco wrote: > As bad as the ++ and +- stuff looks, I usually write code for > CMUCL/SBCL and LW (and other ones) > Having this extra bit of compatibility with LW would help. Compatability among CLs is important, but if there is strong support for another representation scheme, perhaps we can get LW to support it as well. I generated the ++, +- mods for SBCL, CMUCL, and OpenMCL, and I've also floated the idea with Franz. A small advantage of ++, +- (other than current support in LW) is that it adds only a tiny change to the reader: an additional sign check after the exponent sign has been parsed. Christophe's 1.0/0.0 suggestion is clearly readable by humans, but it could cause problems for someone using an infix extension, and it becomes a bit less clear specifying single, double, ... floats explicitly. (1d+inf, -1s+inf might be clearer in this regard.) I also agree that it would be nice to represent all the information in a NaN (even if it is mostly ignored by another CL when reading the representation). Using a somewhat different scheme for NaNs than for infinities might make sense. My need is to address the (currently) bad situation of portably and readably expressing infinite values and (to a lesser extent) NaNs. My hope is that there can be enough agreement to get something reasonable in place. I felt the LW approach was a good place to start, so that's what I did... -- Dan |