Date:         Tue, 23 May 2000 10:48:01 -0400
Reply-To:     WHITLOI1 <WHITLOI1@WESTAT.COM>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         WHITLOI1 <WHITLOI1@WESTAT.COM>
Subject:      Re: Macro Problem On Input
Comments: To: Kurt Beckman <kurt_beckman@EARTHLINK.NET>,
          frank.mwaniki@PHARMA.NOVARTIS.COM
Content-Type: text/plain; charset=US-ASCII

Subject: Re: Macro Problem On Input
Summary: Test it.
Respondent: Ian Whitlock <whitloi1@westat.com>

In response to a question about macro and an INPUT statement Kurt
Beckman <kurt_beckman@EARTHLINK.NET> responded.

>   On 22 May 00 22:48:34 GMT, frank.mwaniki@PHARMA.NOVARTIS.COM
>   (<Frank Mwaniki>) wrote:
>
>   >.....
>   >
>   >        INPUT
>   >           @ 014 %DO I = 1 %TO 27;
>   >                CAPN&I    3.
>   >                %END;
>   >           @;
>   >        INPUT
>   >           @ 095 %DO I = 1 %TO 27;
>   >                XGXN&I    3.
>   >                %END;
>   >           @;
>   > etc....
>
>   I try to avoid using macro language to do something that can be
>   done in open code.  If you are reading a large amount of data,
>   the difference between this solution and one done with open code
>   can be significant.

In general remarks like the above should include some form of
quantification and statement of systems information for which the
claim is made.

I decided to test the claim on a PC running Windows 95 SAS V8 TS1M0.
In general, I agree with the statement - avoid macro when it is easy
to avoid.  However, I agree because of the issue of code clarity,
not execution efficiency.  If it were my task, I would probably have
used Jack Hamilton's <JackHamilton@FIRSTHEALTH.COM> suggestion (had
it occurred to me).

>  input @1   area $char6.
>        @7   id   $char7.
>        @14  (capn1-capn27) (3.)
>        @95  (xgxn1-xgxn27) (3.)
   ...

To me the execution efficiency issues are

   1) What does it cost to use trailing @'s?
   2) What does it cost to use a well designed macro?
   3) What does it cost to use PL1 edit style input?

To find out I ran the following program

   filename q "c:\junk\test.dat" ;

   data _null_ ;
      out = repeat ( "12345" , 19 ) ;
      file q ;
      do i = 1 to 200000 ;
         put i z10. out ;
      end ;
   run ;

   %macro getvars ( dummy ) ;

      %local i ;
      input id $char10. @ ;
      %do i = 1 %to 100 ;
          input x&i 1. @ ;
      %end ;

   %mend  getvars ;

   %macro vlist ( dummy ) ;

      %local i ;
      id $char10.
      %do i = 1 %to 100 ;
          x&i 1.
      %end ;

   %mend  vlist ;

   data w ;
      infile q truncover obs = 10  ;
      %getvars ()
   run ;

   data w ;
      infile q truncover ;
      %getvars ()
   run ;

   proc datasets lib = work nolist ;
      delete w ;
   quit ;

   data w2 ;
      infile q truncover obs = 10 ;
      input %vlist ;
   run ;

   data w2 ;
      infile q truncover ;
      input %vlist ;
   run ;

   proc datasets lib = work nolist ;
      delete w2 ;
   quit ;

   data w3 ;
      infile q truncover obs = 10 ;
      input id $char10. (x1-x100) (1.) ;
   run ;

   data w3 ;
      infile q truncover ;
      input id $char5. (x1-x100) (1.) ;
   run ;

   proc datasets lib = work nolist ;
      delete w3 ;
   quit ;

Note that each INPUT data step is preceded by the same code reading
10 observations, since I wanted to eliminate the noise created by
the previous step.

Here are the results for two executions.

   Macro with 100 trailing @'s  1:28.65  1:28.10
   Macro well designed          1:01.22  1:01.12
   Pl1 edit style no macro      1.01.18  1:00.59

It appears that the claim to avoid trailing @'s has some validity at
least when you generate 100 of them.  However the claim that one
should avoid macro to save execution time appears to hold little
merit in this case.

Ian Whitlock