I could be misunderstanding your requirements, but you can use the hash object to output to datasets arbitrarily, without having to use macros. The main downside is that the dataset must be sorted by the group number, but you should be able to to this while maintaining the relative order within each group if that's important...
The below example generates some random data at the start, and appends a group number (filenum) field as it goes. This field is included in the ultimate output, but if required we should be able to remove it.
data Test;
format key 8. x1-x4 y1-y4 21.3 filenum 8.;
array n
x1-x4 y1-y4;
filenum = 0;
do key = 1 to 100000;
do i = 1 to dim(n);
n = round(ranuni(-1)*50000,0.001);
end;
filenum + 1;
if filenum > 5 then filenum = 1;
output;
end;
drop i;
run;
proc sort data=Test;
by filenum;
run;
proc sql;
select
name
into :fieldlist separated by "','"
from
sashelp.vcolumn
where
libname='WORK' /*change to your library*/
and memname='TEST' /*change to dataset*/
;
quit;
data _null_;
call symput("fieldlist","'&fieldlist.'");
run;
%put &fieldlist.;
data _null_;
if 0 then set Test; /*set up PDV with formats, but load no data*/
declare hash h (ordered:'a');
h.defineKey(&fieldlist.);
h.defineData(&fieldlist.);
h.defineDone();
do until (last.filenum);
set Test;
by filenum;
h.add();
end;
h.output(dataset: 'group' || strip(put(filenum,8.)));
run;
If you want more groups split out, then just increase the value that filenum resets at. As long as the "output" line in the final data step constructs a valid sas dataset name, then you can use this method to output any number of datasets you like, without specifying them in code.
Of course your analysis will then require a macro to run your proc means (etc.) against all of those datasets, but hopefully this will be tidier for creating the datasets in the first place.
