Strange behaviour when concatenating datasets with missing columns

[gabonzo]

Hi,

I have the following problem: I have two datasets test1 and test2, and want to concatenate them to create a third dataset test3.

test1 has only column A present, while test2 has columns A and B. In test3, I want both columns to be there and, if B has a missing value, then B must be a copy of A.

Example:

data test1;
length A $16;
input A $;
datalines;
ABBEY
ABBOTSFORD
KALAMAZOO
OUAGADOUGOU
;
run;
data test2;
length A $16 B $16;
input A $ B $;
datalines;
OKEFENOKEE OKEFENOKEE
ALBUQUERQUE ALBUQUERQUE
;
run;

data test3;
set test1 test2;
if missing(B) then B=A;
run;

The table test3 should (IMHO) look like this:

ABBEY	ABBEY
ABBOTSFORD	ABBOTSFORD
KALAMAZOO	KALAMAZOO
OUAGADOUGOU	OUAGADOUGOU
OKEFENOKEE	OKEFENOKEE
ALBUQUERQUE	ALBUQUERQUE

Instead, it looks like this:

ABBEY	ABBEY
ABBOTSFORD	ABBEY
KALAMAZOO	ABBEY
OUAGADOUGOU	ABBEY
OKEFENOKEE	OKEFENOKEE
ALBUQUERQUE	ALBUQUERQUE

I am using Base SAS 9.4.01 M7, I have verified that this behaviour exists since M2 at least.

Did I get anything wrong?

[s_lassen · Accepted solution]

What happens is that the B variable, which is read (later) from another input dataset, is not reset to missing when rereading the A dataset.

The easy way out in this case is to check what table you are reading from:

data test3;
set test1(in=in1) test2;
if in1 then B=A;
run;

Alternatively, you can explicitly set the B variable missing before the SET statement:

data test3;
length A B $16;
B='';
set test1 test2;
if missing(B) then B=A;
run;

[gabonzo]

Ok, thanks for the explanation. So in practice it's like I had wrote RETAIN for column B.

Is this the expected behaviour or a bug?

Because, however I accept your workaround, I don't think I want to use it (or remember I must use it) every time I have to do something as simple as a variable assignment!

[Kurt_Bremser]

Using an IN= variable is always the preferred way to identify from which dataset data is read.

And the behavior is not a bug, it is well documented and is like this since the beginning of SAS. Variables coming from any incoming dataset are automatically retained. Otherwise a one-to-many MERGE would not work.

[Tom]

---

@gabonzo wrote:

Ok, thanks for the explanation. So in practice it's like I had wrote RETAIN for column B.

Is this the expected behaviour or a bug?

Because, however I accept your workaround, I don't think I want to use it (or remember I must use it) every time I have to do something as simple as a variable assignment!

---

All variables that are coming from input datasets are "retained".  Actually they are just not set to missing at the start of a new iteration.  Without this 1 to Many MERGE would not work.