Re: How do I bisect unbounded intervals?
P 1788,
On Jan 12, 2012, at 8:11 PM, Vincent Lefevre wrote:
> On 2012-01-12 19:40:28 -0600, Nate Hayes wrote:
>> Baker Kearfott wrote:
>>
>>> P-1788:
>>>
>>> Although I find this topic interesting and am glad it is
>>> being discussed, I now wish to express a personal opinion
>>> about its relevance to P-1788. Namely, in my own
>>> view, I think specifying a "bisection point" in a
>>> branch and bound algorithm is too application specific to
>>> be a part of our standard. I think specifying "midpoint"
>>> is well within the scope, in which case we may wish to specify
>>> what is returned for unbounded intervals (just so it will
>>> be known and standard across platforms; however, IMO specifying
>>> "random" in such cases won't help in portability and reproducibility).
>>> However, specifying "bisection point" different from midpoint, and
>>> designed to be good for branch and bound algorithms, seems a bit
>>> of a stretch of the scope.
>
> The standard could include informative text about bisection,
> so that the programmer doesn't misuse some operation provided
> by the standard.
Assuming, of course, the programmer reads the standard.
Sorry. While disclaimer signs on ladders or in manual may satisfy lawyers, I wonder about their effectiveness at forestalling dangerous behavior.
Variant of KISS: You can't sell a program without a manual, and you can't sell a program that NEEDs a manual.
I prefer we try REALLY hard to avoid functions that deliver surprising results.
George