/* trimset.sas D. Brockman 070730 remove leading and trailing missing obs. */

* tested 070730 ;

%macro trimset(iset,oset,vlist);
 
/*
* iset     = input set  
* oset     = output set
* vlist    = list of selected vars separated by spaces. 
*
* trimset returns in global macro var greturn the number of obs in oset.
*
* trimset deletes from the dataset any leading or trailing observations 
* for which all the vars in vlist have missing values. trimset doesn't
* remove the all-missing obs if there is a preceding observation with 
* some of these values not missing and if there is a subsequent observation
* with some of these values not missing. 
*
* Take care to assure an appropriate sort order of the input data set.
*
* Lore: Inspired by a data set that, through intended calculations, 
* surprisingly turned out to have many initial observations 
* missing for variables of interest, though uselessly present in the data set.
*
*/

%global greturn ;

%let dsid=%sysfunc(open(&iset));
%let nobs=%sysfunc(attrn(&dsid,nobs));
%let rc=%sysfunc(close(&dsid));
%let fkeep=%eval(&nobs+1);           * init first record to keep;
%let Lkeep=0;                        * init first record to keep;

* count the vars in our list ;
%let nv=0;                           * init count;
%do %while(%scan(&vlist,%eval(&nv+1)) ne  ) ;   * loop the vars ;
  %let nv=%eval(&nv+1);                         * incr count ;
  %end ;                                        * end loop the vars ;

%if (&nobs gt 0 ) %then %do ;        * if at least one obs in iset ;

  data _null_ ;
    set &iset ;
    retain fkeepL4L LkeepL4L;          * dont lose values found ;

    array varsL4L[&nv] &vlist ;        * set array on our var list ;

    if (_N_=1) then do ;                 * only on first iteration ;
      fkeepL4L=&fkeep ;                  * number of first observation to keep ;
      LkeepL4L=&Lkeep ;                  * number of last observation to keep ;
      end ;

    missL4L=1 ;                        * assume all missing ;
    do iL4L=1 to &nv ;                 * loop the var list ;
      missL4L=missing(varsL4L[iL4L]) ; * is it missing? ;
      if (not missL4L) then leave;     * we found a value ;
      end;                             * end loop the var list ;

    if ( (not missL4L) and (fkeepL4L>_N_) ) then fkeepL4L=_N_ ;  * first ?;
    if ( (not missL4L) and (LkeepL4L<_N_) ) then LkeepL4L=_N_ ;  * Last ?;
    
    call symput('fkeep',fkeepL4L) ;    * number of first some-not-missing obs;
    call symput('Lkeep',LkeepL4L) ;    * number of Last some-not-missing obs;
 
  run;

  %end;                                * end if nobs > 0 ;


data &oset ;    * create output set;
  set &iset (firstobs=&fkeep obs=&Lkeep) ;
run;

%let greturn=%eval(&Lkeep-&fkeep+1);   * number of obs in oset;
%mend trimset ;