[PPL-devel] a bug in PPL Java bindings

[Enea Zaffanella]

On 02/14/2012 10:18 AM, Gianluca Amato wrote:
> Hi,
> I'm here again with another problem concerning boxes and Java bindings.
> I found by chance a sequence of operations which gives a wrong result.

Hello Gianluca.

Thanks for reporting the issue.

This is a misbehavior in method Box<ITV>::upper_bound_assign() which is 
caused by the wrong handling of flags recording box emptiness.
We have now fixed it on git master (I also added a C++ variant of your 
testcase to our test suite).

As witnessed by the observation you performed when trying minor 
variations of your program, the error had to do with the laziness of the 
computation (e.g., the error was disappearing if printing out 
intermediate results).

Please confirm us that things are now OK also on the Java side.

Cheers,
Enea.


> It is difficult to describe what happens, therefore I prepared a small
> Java program exhibiting the bad behavior. The program should print
>
> A in 0, B in 1
> A in 0, B in 1
> A in 0, B in 4
> false
> A in 0, B in 4
>
> but it prints:
>
> A in 0, B in 1
> A in 0, B in 1
> A in 0, B in 4
> false
> A in 0, B in [1, 4]
>
> Cheers,
> --gianluca