Date:         Tue, 2 Jun 1998 13:01:39 -0400
Reply-To:     "Hudson, Spencer" <shudson@VIROPHARMA.COM>
Sender:       "SAS(r) Discussion" <SAS-L@UGA.CC.UGA.EDU>
From:         "Hudson, Spencer" <shudson@VIROPHARMA.COM>
Subject:      FW: SAS Trap: ordering of FORMAT= in ATTRIB statement
Content-Type: text/plain; charset="iso-8859-1"

Here ! Here !
 
Can you imagine the chaos if the following made x character of length 3?
 
proc format ;
 value onetwo 1 = 'One' 2 = 'Two';
run;
 
data x;
 format x onetwo.;
 etc, etc;
run;
 
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
~
Spencer Hudson, Ph.D.
Director, Biostatistics and Clinical Data Management
ViroPharma, Inc
405 Eagleview Blvd
Exton, PA 19341
Phone:  (610) 458-7300 ext 154
FAX:    (610) 458-7380
email:  shudson@viropharma.com
 
 
 
                -----Original Message-----
        From:   Berryhill, Timothy [SMTP:TWB2@PGE.COM]
<mailto:[SMTP:TWB2@PGE.COM]>
        Sent:   Tuesday, June 02, 1998 11:50 AM
        Subject:        Re: SAS Trap: ordering of FORMAT= in ATTRIB
statement
 
        Frankly, I find myself agreeing with Tim Church's original
disgruntlement.  If I formatted a numeric variable as 7.3, I would not
expect it to become length 7.  If I formatted it BINARY32, I would not
expect it to become length 32.  We have many formats which translate a
single character code to a meaningful label-I would not expect assigning
one of those to set the length of the base value.
        I see a reason why this happens, so it may be a feature rather
than a bug:
        when the first mention of a numeric variable is in a format
statement, the length can default to 8.  There is no comfortable default
for the length of a character variable.  I cannot support 200 as the
default, but anything less has problems, too.  Some hapless programmer
just chose the longest formatted value as the default.
        Tim Berryhill - Contract Programmer and General Wizard
        TWB2@PGE.COM <mailto:TWB2@PGE.COM>  or
http://www.aartwolf.com/twb.html <http://www.aartwolf.com/twb.html>
        Frequently at Pacific Gas & Electric Co., San Francisco
        The correlation coefficient between their views and
        my postings is slightly less than 0
        > ----------
        > From: martin trollope[SMTP:martint@HOLLARD.CO.ZA]
<mailto:[SMTP:martint@HOLLARD.CO.ZA]>
        > Reply To:     martin trollope
        > Sent: Tuesday, June 02, 1998 5:27 AM
        > To:   SAS-L@UGA.CC.UGA.EDU <mailto:SAS-L@UGA.CC.UGA.EDU>
        > Subject:      Re: SAS Trap: ordering of FORMAT= in ATTRIB
statement
        >
        > Point taken, but...
        >
        > Formats have a default length, in this case 18. Using $bug. is
the
        > equivalent of $bug18. Thus, in your second datastep, the first
reference
        > to the field is effectively 'Format testvar $bug18.;' hence
the length is
        > 18.
        >
        > Sorry, I didn't mean to come across as if I was lecturing you
at all.
        >
        > Martin
        > ----------
        > From: Tim Churches[SMTP:tchurch@ibm.net]
<mailto:[SMTP:tchurch@ibm.net]>
        > Sent: 02 June 1998 01:48
        > To:   martin trollope
        > Cc:   SAS-L@VM.MARIST.EDU <mailto:SAS-L@VM.MARIST.EDU>
        > Subject:      Re: SAS Trap: ordering of FORMAT= in ATTRIB
statement
        >
        > martin trollope wrote:
        > > This is not really news to me, although I've never seen the
concept in >
        > quite this form on the ATTRIB statement. One of those funny
'hidden'
        > > things that SAS does all by itself (and often causes hours
of >
        > frustrated
        > > debugging) is set the length of the variable to the length
it has at >
        > its
        > > first appearance in the datastep. And not, as I have
discovered to my
        > > chagrin, the first EXECUTED statement, either, but the first
statement
        > > when reading from top to bottom of the datastep.
        >
        > Martin,
        >
        > Yes, I was aware of that, but in my example, the only value
being
        > assigned to the variable testvar in both data steps is "0000",
and the
        > length of testvar is explicitly assigned in the ATTRIB
statement. So why
        > does testvar have a length of 18, not 4, in the second
dataset? Where
        > does the length of 18 come from? Actually, 18 is the length of
the
        > format label 'This must be a bug'. But nowhere is the string
        > 'This must be a bug' assigned to the variable testvar! The
only thing
        > being done is that the format $bug. which contains that string
is
        > assigned as the default output format for the testvar
variable, and that
        > assignation is in the same ATTRIB statement that assigns a
length of 4
        > to the testvar variable. So it seems that the length of a
character
        > variable is dependent not just on the order of statements
within a
        > dataset, but also on the order of parameters within a
statement, and
        > furthermore, the variable length is assigned on the basis of
the maximum
        > length of a label in a default output format assigned to that
variable,
        > not, as I previously thought, the length of the first value
assigned to
        > that variable.
        > Just to repeat the offending code:
        >
        > ************************* ;
        > proc format ;
        >  value $bug
        >  '0000' = 'This must be a bug'
        >   other =  'Or maybe a feature' ;
        >  run ;
        >
        > data test1 ;
        >  attrib testvar length=$4 format=$bug.  ;
        >  testvar = '0000' ;
        >  run ;
        >
        > data test2 ;
        >  attrib testvar format=$bug. length=$4 ;
        >  testvar = '0000' ;
        >  run ;
        >
        > proc contents data=test1 ;
        >  run ;
        >
        > proc contents data=test2 ;
        >  run ;
        > ************************* ;
        >
        > Tim Churches
        >