Re: [eclipse-clp-users] Segmentation fault when loading large source file

Yes, this is generated (I can't imagine how crazy one should be to
type it by hand :-D)

I used Python to generate the ECLiPSe program. I was very short on
time (it was a programming contest) and I didn't feel confident enough
to transform the data to constraints in ECLiPSe itself. In the end I
solved this particular input (there were ten inputs from very small to
this huge one) by generating a program for B-Prolog (while ECLiPSe was
much faster for all other inputs, probably because of OSI CLP solver
vs GLPK in B-Prolog).

The task itself is really just a system of linear equations.

For example, the simplest input was:

5 6
1.638 1.483 1.779 1.304 1.108
4 1 0.2955
1 2 0.479
1 3 0.2417
2 3 0.2139
4 5 0.451
4 3 0.0398

It means that there are 5 variables P1..P5 and 6 data lines describing
relations between variables. So
P1 = 1.638 + 0.0479 * P2 (corresponds to '1 2 0.479' in the input) +
0.02417 * P3 (corresponds to '1 3 0.2417')
P2 = 1.483 + 0.2139 * P3 (corresponds '2 3 0.2139')
etc.

The original problem description can be found on the page 13 of
http://ch24.org/static/archive/2013/2013_preec.pdf

It can be assumed that the data is preprocessed to be easily read in Prolog:
[5, 6].
[1.638, 1.483, 1.779, 1.304, 1.108].
[4, 1, 0.2955].
[1, 2, 0.479].
[1, 3, 0.2417].
[2, 3, 0.2139].
[4, 5, 0.451].
[4, 3, 0.0398].

The output (space-separated values of Ps) for the sample small input
should be (the precision should be at least 1E-3):
2.9606142599 1.8635281 1.779 2.74937371380045 1.108

I would greatly appreciate if you can show how to solve the problem in ECLiPSe.

Sergii.

On Mon, Feb 18, 2013 at 2:45 PM, Joachim Schimpf <jschimpf_at_...311...> wrote:
> On 18/02/2013 20:38, Sergii Dymchenko wrote:
>> Hi,
>>
>> When try to load very large source file (almost 10 Mb) I get
>> "Segmentation fault" error.
>>
>> Version 6.0 #199 (x86_64_linux), Fri Nov  2 22:47 2012
>>
>> $ eclipse -g 5000000 -b R10.ecl
>> Segmentation fault
>>
>> Is there some kind of hard limit on source file size?
>
> There is no limit on source file size.  However, your file contains a
> single big clause, and the parser chokes on that (which is due to the
> hardware stack, and not affected by any ECLiPSe command line setting).
>
> Your clause is mainly a big conjunction of constraints, looking like:
>
> main :-
>      ...
>      P[1] $= 0.971700,
>      P[2] $= 1.797000 + P[37500] * 0.032760 + P[15936] * 0.005254 + ...,
>      ...
>      P[50000] $= 0.782900 + P[20370] * 0.400100 + P[31487] * 0.073090 + ...,
>      ...
>
> I assume you have generated this rather than typed it in ;)
>
> This program-generation technique seems very popular with our
> users, but it is the wrong thing to do.  In a programming language
> like ECLiPSe, which supports metaprogramming, you never have to
> generate a textual program.
>
> If the program that generates this clause is also written in
> ECLiPSe, then it probably contains a loop like this
>
>      ( for(I,1,50000) do
>           generate_constraint(I, Constraint),
>           write(Constraint), writeln(",")
>      )
>
> The thing to do is to get rid of the intermediate file, and
> execute each constraint as soon as you have generated it:
>
>      ( for(I,1,50000) do
>           generate_constraint(I, Constraint),
>           call(Constraint)
>      )
>
> In other words, just have a program that sets up the constraints
> directly from your data files that contain the coefficients (or
> wherever you get the numbers from).
>
> This of course only works if your program generator was already
> written in ECLiPSe.  If not (and you don't want to rewrite it
> in ECLiPSe), let me know, and I'll suggest another technique!
>
>
> -- Joachim
>
> ------------------------------------------------------------------------------
> The Go Parallel Website, sponsored by Intel - in partnership with Geeknet,
> is your hub for all things parallel software development, from weekly thought
> leadership blogs to news, videos, case studies, tutorials, tech docs,
> whitepapers, evaluation guides, and opinion stories. Check out the most
> recent posts - join the conversation now. http://goparallel.sourceforge.net/
> _______________________________________________
> ECLiPSe-CLP-Users mailing list
> ECLiPSe-CLP-Users_at_lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/eclipse-clp-users