| Thread Links | Date Links | ||||
|---|---|---|---|---|---|
| Thread Prev | Thread Next | Thread Index | Date Prev | Date Next | Date Index |
P1788 I circulate a revised version of the I/O clause, as a friendly amendment to the original version. Thanks mainly to Dmitry, Michel, Ned, Vincent. Main changes from the original July 28 version: 1. For output, the requirements on behaviour of the conversion specifier cs are more detailed. (a) If it is "invalid" or makes an unsatisfiable request, an enclosing interval literal shall still be output. (b) An implementation may extend the syntax, allowing an error message string embedded in the literal, e.g. if case (a) happens. (c) The features list explicitly says various features of uncertain form shall be supported. 2. Exact text representation. (a) For a 754-conforming (inf-sup) interval xx = [l,u], conversion of l & is now defined in a separate §12.4.1 "Conversion of 754 numbers to strings". This allows freedom to user & implementer, while ensuring for 754 numbers (which I think is desirable) that exact string output of a finite 754 number lets one extract its (sign,exponent,significand) representation and whether it is normal or subnormal. (b) The relation between exactToInterval and textToInterval is said more clearly. 3. In effect 12.4.1 also specifies, for number literals, the "portable syntax" that Dmitry has called for. As such, it really belongs in the "Interval literals" subclause 11.11.1. Shortly, I'll work on moving it to there, with a formal grammar provided by Dmitry. That makes 11.11.1 so lengthy that I think it should then be a subclause, 11.11, instead of a subsubclause. Thus the current 11.11 will become 11.12. However, these planned changes won't affect the content, which is what you are to vote on. Best wishes John Pryce
Attachment:
20130815IOclause.pdf
Description: Adobe PDF document