SAS: dynamically specify dataset name in hash object in proc fcmp function

I would like to specify the name of the dataset as an input parameter to the fcmp function that is declaring the hash object:

    function somefunction(dsn $, k1 $, k2 $, k3 $, k4 $);
      declare hash h(dataset: "work.someDatasetName");
	  *declare hash h(dataset: dsn);
      rc = h.defineKey('k1', 'k2', 'k3', 'k4');
      rc = h.defineData('d1', 'd2', 'd3', 'd4', 'd5', 'd6');
      rc = h.definedone();
	  rc = h.find();

The commented line declare hash h(dataset: dsn) will not work. The declare function requires a literal or character variable. How can transfer the value of dsn from the function arguments into the hash object declaration statement?

The error that I get is:

163 declare hash h(dataset: dsn);
___
22
202
ERROR 22-322: Expecting a quoted string.
ERROR 202-322: The option or parameter is not recognized and will be ignored.

SAS Documentation

argument_tag:value

specifies the information that is used to create an instance of the hash object.
There are five valid hash object argument and value tags:
dataset: 'dataset_name <(datasetoption)>'

Specifies the name of a SAS data set to load into the hash object.
The name of the SAS data set can be a literal or character variable. The data set name must be enclosed in single or double quotation marks. Macro variables must be enclosed in double quotation marks.

I think this limit is documented.

• PROC FCMP Hash and Hash Iterator Language Elements
  • DECLARE Statement: Hash Object and Hash Iterator Object

https://documentation.sas.com/doc/en/pgmsascdc/9.4_3.5/lecompobjref/n098ljby3kptf0n1caygy0dipu2r.htm#p17vi6r6c5zhmen1aqtyxoml73os

> The name of the SAS data set must be a literal. A literal data set
> name must be enclosed in single or double quotation marks.

As pointed out by Tom, it is not possible to pass a string variable argument to the declaration of a hash object within an fcmp function (in the same way that you can for a data step hash object).

That said, there may well be an alternative (and better performing) lookup solution.

You mention that this lookup function is to be 'called' millions of times:-

• Do you mean called from within a data step - ie where the data step is iterating through (millions of) observations from a 'master' dataset and each observation has variables for the dataset name and other input parameters listed in your example fcmp function?

• How many of these lookup tables are there?

• How large are these lookup tables? How many unique values per dataset? Can they all fit into memory?

• It would appear from your example fcmp function that these lookup tables have either the same data structure, or at least the same key and data variables required to satisfy the lookup?

Depending on your answers to the above, data step hash object(s), with the addition of some macro code, may well satisfy your requirement - either pre-loaded before iterating through your 'master' dataset or loaded on the fly.

Or there maybe an alternative lookup approach that is more appropriate for your particular need. Difficult to say without more information.

EDIT:

A small number of lookup tables could (depending on size) be preloaded in your data step - ie before you start iterating thru your master dataset with, for example, do until(eof_master) code.

I would then use if/then logic (or select/when) within the do until loop to lookup values (in the relevant hash object) based on the incoming dataset name (from your master).

The pre-loading of hash objects will likely more than compensate for the if/then or select/when.

NB There is no need to use macro code and remember to add a stop stmt after the do until loop.