[developers] RMRS characterization differences

[Bernd Kiefer]

Hi all,

i will happily revert the change Ben mentions below. However, this will
break characterization in some of the other 4739 configurations PET can
be run in. As you can see from the changelog, this was a fix for some
other problem that occured with characterization.

I have no time to look into this problem systematically at the moment,
so my only option is to simply revert the change. After that, i will
not work on anything that concerns characterization in PET again for
the following reasons:

- this mechanism is so dependent on things being done at the right time
  and in the right order that given the many ways PET can be used, it
  is almost impossible to do it correctly for all configurations. It
  really interacts badly with restriction, unfilling and the like
- i have never seen a specification for the functionality that was much
  more specific than "there are some empty places in the feature
  structure ..." and "the output has to look this way ..."
- the whole thing is so brittle and makes so many assumptions that if i
  had had a say in the matter, i would have refused to implement it
  like this in the first place
- 6 out of 7 requests/bug reports i worked on for PET in the last time
  had to do with characterization, which took some of my time and
  confirms my initial concerns

To sum up: I would like to ask the people who created this
functionality to also take over its implementation in PET.

Best,
        Bernd


> The bug in PET reported by Sergio below apparently originates in SVN 
> revision 328 due to beki01 (Bernd). The changelog is shown below. Bernd, 
> please could you fix this?
> 
> Thanks,
> - Ben
> 
> ===================================================================
> --- CHANGELOG   (revision 327)
> +++ CHANGELOG   (revision 328)
> @@ -1,4 +1,6 @@
>  v0.99.??
> +  - changed init() of tLexItem, such that characterization during 
> exhaustive
> +    unpacking works correctly
>    - Moved functionality of fresh_constraint_p into cached_constraint_of
>    - Default for memory allocation is now NO mmap, because of the notorious
>      problems with the ecl-based modules. Works fine for cheap and flop.
> Index: configure.ac
> ===================================================================


-- 
----------------------------------------------------------------------
Bernd Kiefer       DFKI GmbH, Stuhlsatzenhausweg, D-66123 Saarbruecken
kiefer at dfki.de     +49-681/302-5301 (phone)     +49-681/3025338  (fax)
----------------------------------------------------------------------
Deutsches Forschungszentrum fuer Kuenstliche Intelligenz GmbH
Trippstadter Strasse 122, D-67663 Kaiserslautern, Germany
Geschaeftsfuehrung: Prof. Dr. Dr. h.c. mult. Wolfgang Wahlster (Vor-
                    sitzender), Dr. Walter Olthoff
Vorsitzender des Aufsichtsrats: Prof. Dr. h.c. Hans A. Aukes
Amtsgericht Kaiserslautern, HRB 2313