[Coin-discuss] Reset (was OsiOsl bug ?)
Tim Hultberg
thh at mat.ua.pt
Mon Jun 2 11:07:02 EDT 2003
>From my simple user point of view, things which are explicitly set different
than the default via the OSI interface should be kept round loadProblem and
nothing else. But I imagine that there are good reasons for this not being
so.
Anyway I can now do what I want with the reset function. However what I really
wanted to be able to do, and still cant, is to use the problem modification
methods (f.ex setObjCoeff) for MIP problems. I realize that in the case of
MIP there will not be any advantage in terms of execution speed by using
problem modification instead of a reset followed by loading the modified
data, but it would make implementation of problem modification within FLOPC++
easier not having to implement to different shemes. (Alternatively I could
prohibit the use problem modifacation for MIPs)
Tim Hultberg
PS I just downloaded the lastest tarball (2June), which broke anything with
this error (even for pure LPs):
Program received signal SIGSEGV, Segmentation fault.
0x4037a153 in OsiClpSolverInterface::setInteger (this=0x809d050, indices=0x0,
len=1072693248) at OsiClpSolverInterface.cpp:803
803 integerInformation_[indices[i]]=1;
(Same thing with Osl and Glpk)
On Friday 30 May 2003 19:08, John J Forrest wrote:
> Tim,
>
> It would be very tricky to decide what should be kept round loadProblem -
> e.g. should minimization be kept. If nothing is kept then it should be a
> constructor.
>
> As you say the one clean solution is to have a base empty model which has
> settings you want to keep e.g. minimization and then clone from that.
>
> However this may not always be best solution. Laci suggested that we add
> a "reset" function to interfaces which would set back as if default
> constructor. It would not involve any more coding as the current default
> constructor coding would be copied to reset() and the default constructor
> just call reset.
>
> Comments?
>
> John Forrest
--
Tim Hultberg
More information about the Coin-discuss
mailing list