| Thread Links | Date Links | ||||
|---|---|---|---|---|---|
| Thread Prev | Thread Next | Thread Index | Date Prev | Date Next | Date Index |
Vincent & P1788 On compressed intervals §8.8.8 On 12 Dec 2012, at 14:36, Vincent Lefevre wrote: > You assume here that the threshold t is the same before and after > Empty has been produced. What I have in mind is that: > > 1. xx = Empty can be produced with t = trv in some program block. > 2. Empty survives across the program block. > 3. A new program block sets t = dac. > 4. In this program block, normalInterval(xx) is called. > > I don't think §8.8.8 forbids that yet. I wrote: > That is a VERY good point. I (and others till now) had not considered how the threshold is initially set, or under what conditions it can be changed. I know this is probably a minority interest. But for certain applications it may become an important way to get high performance in interval computations. So having started it, we had best get it right. I think 8.8.8 paragraph 6: > The threshold has the properties of an attribute logically associated with a program block, as in 754-2008 §4 (see also 754-2008 §2.1.7). It shall be possible to specify a constant value for it as in 754-2008 §4.1; dynamic mode specification should be supported as in 754-2008 §4.2. is misguided and should be removed entirely. Treating t as something one might change dynamically opens a can of worms, as Vincent points out above. Having a separate level 2 compressed type, per threshold as well as per parent type, seems to work better. I offer the attached revision of 8.8.8 for comments. John Pryce
Attachment:
compressedArithSnippet.pdf
Description: Adobe PDF document