Re: [soapbuilders] The Interop tests and BDG (was : some questions/observations re: BDG)
- Dave Winer wrote:
> 3. I just sent an email to Keith suggesting that we have a phone talk.
Boy that feels good to get off my chest.
Can I make a suggestion? How about I arrange a teleconference for 10 am
PDT / 1 pm EDT on Monday, April 2nd? I do believe that we could all
benefit from a higher bandwidth and synchronous conversation than e-mail
permits. And I still do believe that a face-to-face would be productive.
I still have a large room reserved for April 19th and 20th...and will even
spring for lunch... ;-)
- Sam Ruby
- jake wrote:
> Do you have any suggestions about how we should deal with this issue, inuse David Gay's I/O libraries (http://netlib.bell-labs.com/netlib/fp/).
> terms of real-world interop?
if that's not possible, make sure you write out floating point values
with enough precision (17 digits for doubles, 9 for floats). also doc-
ument that your implementation may not be fully compliant wrt. FP
> Should an implementation ignore FP errors beyond a certain precision?imo, that's up to the application, not the stack - but test code that
checks floating point values needs to be aware of this.
> Should it simply fault, unless IEEE 754 compliance is known for certain?you mean XSD schema compliance -- the schema spec goes beyond
IEEE 754 in this case. I'm pretty sure all modern C libraries comply
with IEEE 754, but very few comply with the schema spec.
> (I think not, since as you point out, many people won't have any easyagreed.
> way to know if their implementation is compliant or not.)