Re: [Quantlib-dev] long and possibly inifiniteobserver/observablenotifying loops

[DU V. DE V. F. G. P. <fra...@ca...>]

Hi all,

After further inquiries I have found out  what was going wrong in =
notification chains. Indeed, I discovered that the main culprit was the =
CapStripper. A drawing being worth more than thousand words (especially =
if your command of English is poor ;-) I have attached a picture showing =
dependencies between objects for the stripping of 3M volatilities. Let =
me point out that is only one part of the problem because this =
CapStripper is observed by another one which strips the 6M libor caps...
The fix proposal I tried to described last time is to make LazyObjects =
notifying their observers only once until they are recomputed. ( I hope =
it is clear enough this time).
This solves the problem quite efficiently (apart from the fact that all =
objects inheriting from TermStructure notify their observers twice =
systematically). Yet I can't refrain myself from suggesting further =
design improvements. What about removing all market data dependencies =
from instruments and indexes definitions ?  Not only it would simplify =
notifications chains, but it would also decouple Instrument/Indexes =
implementation from market data implementations.
Regards,
Fran=E7ois