[PPL-devel] [Fink-devel] ppl-0.10.2 check thorough test fail

[David Fang]

Hi,

>>> I agree with Roberto that all of the symptoms suggest that Apple gcc-4.2
>>> is miscompiling something (and we would really appreciate if someone can
>>> try out using a newer gcc).
>>>
>>> The first differences above are related to a very simple MIP problem
>>> test. Can you please provide us with the output of the following command:
>>>
>>> $ cd <PATH_TO_PPL_BUILD>/demos/ppl_lpsol
>>> $ ./ppl_lpsol -s -p1 -v 4 -c -M
>>> <PATH_TO_PPL_SRC>/demos/ppl_lpsol/examples/ex1.mps
>>>
>>> This will also show the problem as read from the input, which should be
>>> something like:
>>> =============
>>> Integer variables:
>>> x1 x2
>>> Constraints:
>>> -2*x1 - x2 >= -5
>>> 4*x1 - 4*x2 >= -5
>>> x1 >= 0
>>> x2 >= 0
>>> Objective function:
>>> x1 - 2*x2
>>> Maximizing.
>>> =============
>>
>> Hi,
>>     On 10.6/i686, apple-gcc-4.2, gmp-5.0.1, I get:
>>
>> fang at fangbook 4> ./ppl_lpsol -s -p1 -v 4 -c -M
>> ../../../demos/ppl_lpsol/examples/ex1.mps
>> Integer variables:
>> x1 x2
>> Constraints:
>> -2*x1 - x2 >= -5
>> 4*x1 - 4*x2 >= -5
>> x1 >= 0
>> -x1 >= -1
>> x2 >= 0
>> -x2 >= -1
>> Objective function:
>> x1 - 2*x2
>> Maximizing.
>> Optimum value: 1
>> Optimum location:
>> x1 = 1
>> x2 = 0
>>
>> Different indeed!  Does this indicate an earlier parse error?
>>
>> Fang
>
>
> Well ... ppl_lpsol uses glpk to parse the input file.
> Apparently, the installed glpk is parsing the input file differently.
> Which version of glpk do you have?
>
> If I am to guess, I would say that the installed glpk is flagging the
> two variables as being *binary* variables, rather than *integer* ones.
> But this is just a guess, I will have to check the documentation for the
> input file format as well as the changelog for glpk.

I have glpk-4.44:

fang at fangbook 6> fink list -t glpk
Information about 10198 packages read in 9 seconds.
  i      glpk    4.44-1  GNU Linear Programming Kit
  i      glpk-dev        4.44-1  GNU Linear Programming Kit
         glpk-java       4.19-1  Java bindings for GLPK
  i      glpk-shlibs     4.44-1  GNU Linear Programming Kit


> Apart from this, ppl_lpsol seems to be correctly doing its job in
> optimizing the problem (and use of option -c seems to confirm that glpk
> is computing the very same optimal value).
>
> Enea.

That's good to know that ppl is doing its part of the work 'correctly'.
Anything else I can test for you?

Fang


David Fang
http://www.csl.cornell.edu/~fang/
http://www.achronix.com/