Re: [eclipse-clp-users] bug version 6?

Vicenç Gomez wrote:
> Dear all,
> 
> I've experienced the following problem when upgrading to version 6:
> Using the following code:
> 
> ...
> prob: eplex_solver_setup(min( X1_2*C1_2 + ...  + X93_4*C93_4)),
> ...
> 
> where I ommited part of the string I pass as argument to the min predicate,
> I get this error :
> 
> Goal failed unexpectedly: current_content(hash(256, 263, [a(151) -> 431,
> a(152) -> 432, a(153) -> 433, a(154) -> 434, a(155) -> 435, a(156) -> 436,
> a(157) -> 437, a(158) -> 438, a(159) -> 439, a(160) -> 440, a(161) -> 441,
> a(162) -> 442, a(163) -> 443, a(164) -> 444, a(165) -> 445, a(...) -> 446,
> ... -> ..., ...]), a(3), _699378)
> Aborting execution ...
> 
> Looks like the length of the string passed as argument to min is too large
> (it has more than 3000 characters) since after proper shortening
> everything goes fine again.
> 
> Is there a maximum length for the argument string?  This error didn't
> occur in the previous version of Eclipse and I don't know how to make the
> string shorter while keeping all the constraints.

Hi Vicenç,

the ECLiPSe 6 compiler is still lacking some overflow handling when the
compiled code needs too many registers.  This tends to be only a problem
when compiling generated code like yours.  The quickest workaround for
you is probably to declare the offending predicate as dynamic, i.e.

:- dynamic remeses/3.



In general, I would recommend against generating such large flat models
(i.e. with all variables mentioned explicitly) as ECLiPSe source code:
it is complicated to generate, slow to compile, and only executed once.

A better way is to use a structure similar to
http://www.eclipse-clp.org/examples/apc_lp.ecl.txt
i.e. have a piece of generic (data-independent) code to set up the model
(model/5), and generate only the data part (data/3, which contains only
numbers, not variables).  The variables are then only generated as
arrays or lists at runtime.


-- Joachim