On Wed, Apr 20, 2011 at 11:46:32AM +0100, Nick Fortescue wrote: > On 20 April 2011 11:45, Gilles Sadowski <[email protected]> wrote: > > Hello Nick. > > > >> MullerSolver() and other Univariate solvers (NewtonSolver, > >> SecantSolver) have had their no argument constructor deprecated. In > >> the deprecation comment it says "to be removed in 3.0". I can see no > >> mention in the javadocs for commons-math 2.2 or 3.0 of any replacement > >> mechanism for constructing, and the method still exists in the 3.0 > >> javadoc. > >> I can think of 4 possibilities: > >> > >> 1) The deprecation of this method was a mistake and added at the same > >> time as the constructor which took a function. > >> 2) The proposal is that construction be moved to a factory class, but > >> the Factory class has not been added yet. > >> 3) There is an alternative way of constructing a MullerSolver I > >> haven't found yet. > >> 4) Deprecation has been decided with no migration route yet decided upon. > >> > >> Can anyone tell me which of these is correct, or a fifth option, and > >> tell me how code which uses "new MullerSolver()" should be changed? > >> > >> I really hope it isn't 4! My guess is 2, as SecantSolver has > >> UnivariateRealSolverFactoryImpl.newSecantColver(). While I might > >> disagree with this design (force the use of the Factory) at least > >> there is the hope of migration. > > > > It's not 2) because "UnivariateRealSolverFactory" has been removed in the > > development code. Cf. https://issues.apache.org/jira/browse/MATH-439 > > > > The indication of future removal was a mistake: A default constructor still > > exist (setting a default value for the absolute accuracy) but it does not > > behave as in 2.2 (where there was also a default "maximum iteration count", > > which is now replaced with a "maximum evaluation count" to be passed to the > > "solve" method). > > > > > > Best regards, > > Gilles > > Thanks for clarifying Gilles. Is there any chance of getting the > @deprecated out of a 2.2 release? We try and keep our deprecated > warning count down in our codebase, and this has hit us with a lot. > Similarly for Secant, Newton etc?
Unfortunately, there probably won't be any 2.2.x release. [The 2.2 release has already much delayed the release of the upcoming 3.0.] > Should I add this to the Math-439 > issue? I think that it would be clearer to open a new issue. Sorry for the inconvenience, Gilles --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
