[developers] [pet] unpacking

Yi Zhang yzhang at coli.uni-sb.de
Thu Jul 12 12:39:53 CEST 2007


well, in a sense, exhausting the edge limit is a crash, since the parser was
not able to find the solution within limited memory, and just aborted. In
other words, the parser did not even know whether there was a parse or not.
With selective unpacking (-packing=15), the parser spends much more time in
searching before hitting the limit.

Yi


On 7/12/07, Sergio Roa <sergior at coli.uni-sb.de> wrote:
>
> Hi Yi,
>
> On Thursday 12 July 2007, Yi Zhang wrote:
> > > I mean leaving it as default, so, the options now are:
> > >
> > >   cheap -limit=100000 -tok=fsr -nsolutions=10
> >
> > Oh, I just checked the code. It seems that if ``-packing'' is not given,
> > then packing is disabled (=0, no packing). If ``-packing'' is given
> without
> > ``=n'', then it's defaulted to 15.
> >
> > So in your case, you are running without packing. I just tried again,
> and
> > it leads to immediate out of memory failure. Therefore, what you got
> > without packing is a crash, rather than real results.
>
> How can  you know this  is a crash?  The verbose output I  get without
> packing is:
>
> 69133 types in 7.5 s
> edge limit exhausted (100000 pedges)
> id: 1
> trees: 0
> readings: 0
> words: 200
> words_pruned: 37
> mtcpu: 0
> first: -1
> tcpu: 14290
> utcpu: 0
> ftasks_fi: 16514325
> ftasks_qc: 4671458
> fsubs_fi: 0
> fsubs_qc: 0
> etasks: 175805
> stasks: 150097
> aedges: 50281
> pedges: 100000
> upedges: 0
> raedges: 0
> rpedges: 0
> medges: 0
> unifications_succ: 175421
> unifications_fail: 25708
> subsumptions_succ: 0
> subsumptions_fail: 0
> copies: 200660
> dyn_bytes: 908659104
> stat_bytes: 904933232
> p_dyn_bytes: 0
> p_nstat_bytes: 0
> cycles: 0
> fssize: 1
> unify_cost_succ: 409
> unify_cost_fail: 109
> equivalent: 0
> proactive: 0
> retroactive: 0
> frozen: 0
> failures: 0
>
> I thought it means that the edge limit is exhausted.
>
> --
> Sergio.
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.delph-in.net/archives/developers/attachments/20070712/1b279b66/attachment.html>


More information about the developers mailing list