Notes about COBOL Sort

Notes about COBOL Sort

The bare outline of COBOL sorting and merging is simple enough:

We have an input file (unsorted), and output file (sorted) and a workspace file (attached to SORTWK01).
We have an SD for the Workspace file.

We can sort with a statement such as:

         Sort Work-File
         On Ascending Key Work-Key
         Using Unsorted-File
         Giving Sorted-File.

Here the key field (Work-Key) is part of the data record for the workspace file (Work-File).

Available Options

We could say "Descending Key" instead of "Ascending Key".

We could have more than one key, as in:

         On Ascending Key Work-Key-1
           Descending Key Work-Key-2

We could add the optional phrase "With Duplicates in Order". This means that if two or more input records have the same key value, they should be in the same relative order in the sorted file (where they will be consecutive) as in the unsorted file (where they were not necessarily consecutive).

Some Available Special Registers

These are automatically allocated by the compiler for every COBOL program.

We can provide COBOL with an estimate of the number of records to be sorted using the Sort-File-Size register. That is, before the Sort statement, we can move a number to Sort-File-Size. This has the same effect as using the FILSZ control statement with DFSORT.
The Sort and Merge operations give us a return code in the Sort-Return register: 0 for success, 16 for failure. Likewise, if the programmer moves a nonzero value to Sort-Return, the operation will terminate after the next Return or Release.
Normally DFSORT writes its messages to the SYSOUT data set. As this is also used by the COBOL Display verb, confusion could result. We can change the destination for the DFSORT messages produced by a COBOL program by using the Sort-Message special register. Before the Sort statement, move a DDname to Sort-Message, as in:
```
   Move 'SORTOPUT' to Sort-Message.
```
For this to work, we need to have an output DD statement named SORTOPUT in the execution step.
The COBOL Sort verb uses DFSORT, and it can make use of DFSORT control statements. It will look for these in data sets listed under the names IGZSRTCD and SORTCNTL (by default). If we want to change the DD name used here, we can use the Sort-Control special register. Before the Sort statement, move a DDname to Sort-Control, as in:
```
   Move 'MYSORTCS' to Sort-Control.
```
For this to work, we need to have an output DD statement named MYSORTCS in the execution step, containing DFSORT control statements. For instance:
```
          //MYSORTCS  DD *
            OPTION FILSZ=E3500,SKIPREC=3 
          /*
```

Example of using special registers

Suppose we have the following in the JCL in the execution step:

          //SORTCS DD *
            OPTION SKIPREC=1
          /*
          //SYSOUT   DD SYSOUT=*
          //SORTOPUT DD SYSOUT=*
          //PRINTER  DD SYSOUT=*

Suppose we have the following in the Procedure Division:

           Move 20 to Sort-File-Size.
   
           Move 'SORTOPUT' to Sort-Message.

           Move 'SORTCS' to Sort-Control.

           Sort S-W-File
               On Ascending Key ....
               Using ....
               Giving ....

Now:

Anything written by the Display verb will be written to the SYSOUT data set.
All routine messages written by DFSORT (which is called by the Sort verb) will be written to the SORTOPUT data set instead of SYSOUT.
Instead of looking for extra DFSORT control statements in the IGZSRTCD data set, COBOL will look for them in the SORTCS data set.
DFSORT is expecting approximately 20 input records.
DFSORT will skip the first record before sorting the rest.
Because the three output data sets are listed in the order given, messages from Display will all be printed before messages from Sort, and both will precede anything written to the PRINTER data set.

Notice the use of the word OPTION in the SORTCS data set.

Compiler messages involved in sorting

If we use the Using or Giving options, an informational message is generated urging the use of the FASTSRT option to speed up the file input and/or output involved. If we were using large files, this might be useful. The default is NOFASTSRT.

If we use the Sort-File-Size special register, a warning is generated saying that the value provided might be overridden by a value found in the DFSORT FILSZ control statement in the Sort-Control data set.

An alternate use of the Sort verb

The Sort verb can also be used to sort tables. This requires the use of both an input procedure (which Releases one table entry at a time to the workspace file) and an output procedure (which Returns one record at a time from the workspace file and copies it into the table). This option may or may not be more work, in a given case, than writing code for a sorting algorithm.

In any case, this is a ridiculously slow way to sort a table. The table is in memory, and we are sorting it essentially by writing to a file, sorting the file, and reading the file back into the table.