#308 hfst-proc --weight-classes seems to just pick the first analysis

[Kevin Brubeck Unhammer]

As

    $ echo allaskuvllain guollebivdu| hfst-proc -W sme-nob.automorf.hfst |tr '/ ' '\n' 
    ^allaskuvllain
    alas<n><sem_body><sggencmp><cmp>+kuvlla<n><pl><loc>~10~
    alas<n><sem_body><sggencmp><cmp>+kuvlla<n><sg><com>~10~
    allaskuvla<n><sem_edu_org><pl><loc>~0~
    allaskuvla<n><sem_edu_org><sg><com>~0~
    allat<adj><attr><cmp>+skuvla<n><sem_edu_org><pl><loc>~10~
    allat<adj><attr><cmp>+skuvla<n><sem_edu_org><sg><com>~10~$
    ^guollebivdu
    guollebivdu<n><sem_act><sg><nom>~0~
    guolli<n><sem_ani><sgnomcmp><cmp>+bivdu<n><sem_act><sg><nom>~10~$

shows, this fst has weights that should lower the priority of any analysis with a <cmp> tag.

However, when we try to use hfst-proc's --weight-classes 1 to just select the best analyses, it seems to just pick the first appearing weight-class instead:

$ echo allaskuvllain guollebivdu| hfst-proc --weight-classes 1 sme-nob.automorf.hfst |tr '/ ' '\n' 
^allaskuvllain
alas<n><sem_body><sggencmp><cmp>+kuvlla<n><pl><loc>
alas<n><sem_body><sggencmp><cmp>+kuvlla<n><sg><com>$
^guollebivdu
guollebivdu<n><sem_act><sg><nom>$

(so for each word, it picks the first set of analyses that have the same weight?)

[Kevin Brubeck Unhammer]

hfst-proc/formatter.cc says

  for(int i=0;i<maxAnalyses && it != new_finals.end();i++,it++) {
      // this condition filters out > maxWeightClasses
      if(dynamic_cast<const LookupPathW*>(*it) != NULL) { // if we actually have weights
      Weight current_weight = dynamic_cast<const LookupPathW*>(*it)->get_weight();
      if (classes_found == -1) { // we're just starting
          classes_found = 1;
          last_weight_class = current_weight;
      } else if (last_weight_class != current_weight) { // we might want to ignore the rest due to weight classes
          last_weight_class = current_weight;
          ++classes_found;
      }
      if (classes_found > maxWeightClasses) {
          break; // don't insert any more
      }
      }
      clipped_finals.insert(*it);

So it should've sorted the analyses by weight first.

[Kevin Brubeck Unhammer]

https://github.com/unhammer/hfst3/compare/git-svn...master fixes this

[Kevin Brubeck Unhammer]

should be fixed in -r4426