SAS9.2新增了这个语句，文档里的信息很详细，但要追根溯源的话，SAS Global Forum 2007中一文的应该算上：《User-Written DATA Step Functions》，作者Jason Secosky。从这里可以看到该语句的雏形，也有助于理解现在更完整的FCMP语句。文章主要内容：
PROC FCMP
PROC FCMP is invoked to create functions and CALL routines. The syntax for the procedure is:
PROC FCMP OUTLIB=libname.dataset.package INLIB=in-libraries ENCRYPT;
routine-declarations;
The OUTLIB= option is required and specifies the package where routines declared in the routine-declarations
section are stored. The routines declared in routine-declarations may call FCMP routines that exist in other
packages. In order to find these routines to check the validity of the call, the data sets specified on the INLIB= option
are searched, where INLIB=in-libraries can be:
inlib=library.dataset
inlib=(library1.dataset1 library2.dataset2 ... libraryN.datasetN)
inlib=library.datasetM – library.datasetN
If the routines being declared do not call FCMP routines in other packages, you do not need to specify the INLIB=
option. The ENCRYPT option specifies that the routines are encrypted when stored.
DECLARING FUNCTIONS
Routine-declarations are where one or more functions or CALL routines are declared. A routine consists of four
parts: a name, parameters, a body of code, and a RETURN statement. These four parts are specified between the
FUNCTION or SUBROUTINE keyword and an ENDSUB keyword. For functions, the syntax is:
FUNCTION name (parameter-1, …, parameter-N);
program-statements;
RETURN (expression);
ENDSUB;
After the FUNCTION keyword, the name of the function and its parameters are specified. Parameters in the function
declaration are termed formal parameters and can be used within the body of the function. To specify a string
parameter, a dollar sign ($) is placed after the parameter name. For functions, all parameters are passed by value.
This means the value of the actual parameter, the variable or value passed to the function from the calling
environment, is copied before being used by the function. This ensures any modification of the formal parameter by
the function doesn't change the original value.
DATA step _temporary_ arrays can also be passed to FCMP routines. Arrays are passed by value. That is, the array
is copied before calling the routine and the copy is used by the routine. Copying large arrays can be costly. The next
section, "Declaring CALL Routines', describes how to pass arrays without copying by using the OUTARGS
statement. The syntax for specifying a formal array parameter is:
FUNCTION name (numeric-array-parameter[*], character-array-parameter[*] $);
Program-statements are a series of DATA step statements that describe the work to be done by the function. In
general, most DATA step statements and functions are accessible from FCMP routines. The DATA step file and data
set I/O statements, like INPUT, INFILE, SET, and MERGE, are not available from FCMP routines. However, the PUT
statement is supported to a certain extent. For a more complete list of differences, see the section titled "FCMP and
DATA Step Differences" later in this paper.
All functions must return a value. This is done with the RETURN statement. The RETURN statement accepts a
parenthesized expression, which is the value to be returned to the calling environment. The function declaration
ends with an ENDSUB statement.
DECLARING CALL ROUTINES
CALL routines are also declared within routine-declarations by using the SUBROUTINE keyword instead of the
FUNCTION keyword. Functions and CALL routines have the same form, except CALL routines do not return a value
and can modify their parameters. The parameters to be modified are specified in an OUTARGS statement. The
syntax of a CALL routine declaration is:
SUBROUTINE name (parameter-1, …, parameter-N);
OUTARGS out-parameter-1, …, out-parameter-N;
program-statements;
RETURN;
ENDSUB;
A RETURN statement is optional within the definition of a CALL routine. When a RETURN statement executes,
execution is immediately returned to the caller. A RETURN statement within a CALL routine does not take a value to
be returned as CALL routines do not return values like functions do.
VARIABLE SCOPE
A critical part of routines and programs being independent of each other is variable scope. A variable's scope is the
section of code where a variable's value can be used. In the case of FCMP routines, variables declared outside of a
routine are not accessible inside a routine. Also, variables declared inside a routine are not accessible outside of the
routine. Variables declared within a routine are called local variables because their scope is "local" to the routine.
Local variables store intermediate results of a computation and cannot be accessed after a routine returns. When a
routine is called, memory for local variables is pushed on the call stack (Wikipedia 2007). When the routine returns,
the memory used by local variables is popped off the call stack.
Variable scope can get confusing when local variables in different routines have the same name. When this occurs,
each local variable is distinct. In the following program, the DATA step and CALL routines SUBA and SUBB have a
local variable named x. Each x is distinct from the other x variables. When this program executes, the DATA step
calls SUBA and SUBA calls SUBB. Each environment writes the value of x to the log. The log output shows how
each x is distinct from the others.
proc fcmp outlib=sasuser.funcs.math;
subroutine subA();
x = 5;
call subB();
put 'In subA:' x=;
endsub;
subroutine subB();
x = 'subB';
put 'In subB:' x=;
endsub;
run;
options cmplib=sasuser.funcs;
data _null_;
x = 99;
call subA();
put 'In DATA step: ' x=;
run;
Output:
In subB: x=subB
In subA: x=5
In DATA step: x=99
RECURSION
Recursion is a problem-solving technique that reduces a problem to a smaller one that is simpler to solve and then
combines the results of the simpler solution(s) to form a complete solution. A recursive function is a function that
calls itself, either directly or indirectly. FCMP routines can be recursive.
Each time a routine is called, whether it is recursive or not, memory for local variables is pushed on the call stack.
The memory on the call stack ensures independence of local variables for each call. This can be confusing when a
routine calls itself. When a routine calls itself, both the caller and callee must have their own set of local variables for
intermediate results. If the callee were able to modify the caller’s local variables, it would not be easy to program a
recursive solution. A call stack ensures the independence of local variables for each call.
In the next example, the FCMP routine ALLPERMK takes two parameters, n and k and prints all P(n, k) = n! / (n-k)!
permutations that contain exactly k out of the n elements. The elements are represented as binary (0/1) values. The
function ALLPERMK calls the recursive function PERMK to traverse the entire solution space and only output items
that match a particular filter.
proc fcmp outlib=sasuser.funcs.math;
subroutine allpermk(n, k);
array scratch[1]/nosymbols;
call dynamic_array(scratch, n);
call permk(n, k, scratch, 1, 0);
endsub;
subroutine permk(n, k, scratch[*], m, i);
outargs scratch;
if m-1 = n then do;
if i = k then
put scratch[*];
end;
else do;
scratch[m] = 1;
call permk(n, k, scratch, m+1, i+1);
scratch[m] = 0;
call permk(n, k, scratch, m+1, i);
end;
endsub;
run; quit;
options cmplib=sasuser.funcs;
data _null_;
call allpermk(5,3);
run;
This program uses the /NOSYMBOLS option on the ARRAY statement to create an array without a variable for each
array element. A /NOSYMBOLS array can only be accessed with an array reference, scratch[m], and is equivalent to
a DATA step _temporary_ array. A /NOSYMBOLS array takes less memory than a regular array because no space
is allocated for variables. ALLPERMK also uses FCMP dynamic arrays, which are discussed in the next section.
DYNAMIC ARRAYS
In FCMP routines, arrays can be resized. This is done by calling the built-in CALL routine DYNAMIC_ARRAY. The
syntax is:
CALL DYNAMIC_ARRAY (array, new-dim1-size, …, new-dimN-size);
When using dynamic arrays, keep in mind that support is limited to FCMP routines. When an array is resized, the
resized array is only available within the routine that resized it. It is not possible to resize a DATA step array or to
return an FCMP dynamic array to a DATA step.
DIRECTORY TRAVERSAL EXAMPLE
Several users have requested functions that traverse a directory hierarchy. Implementing this functionality with DATA
step and macro code is difficult because recursion or pseudo-recursion is not easy to code. In this section, we
develop a routine, named DIR_ENTRIES, that fills an array with the full pathname of all the files in a directory
hierarchy. This exercise shows the similarity between FCMP and DATA step syntax and underscores how FCMP
routines simplify a program and produce independent, reusable code.
DIR_ENTRIES takes as input a starting directory, a result array to fill with pathnames, an output parameter that is
the number of pathnames placed in the result array, and another output parameter that indicates if the complete
result set was truncated because the result array was not big enough. The flow of control for DIR_ENTRIES is:
1. Open the starting directory.
2. For each entry in the directory,
• If the entry is a directory, call DIR_ENTRIES to fill the result array with the subdirectory's pathnames.
• Otherwise, the entry is a file, so add the file's path to the result array.
3. Close the starting directory.
OPENING AND CLOSING A DIRECTORY
We begin by abstracting two common operations into their own routines. Opening and closing a directory are
handled by the CALL routines DIROPEN and DIRCLOSE. DIROPEN takes a directory path and has the following
flow of control:
1. Create a fileref for the path using the filename function.
2. If the filename function fails, output an error message to the log and return.
3. Otherwise, use the DOPEN function to open the directory and get a directory ID.
4. Clear the directory fileref.
5. Return the directory ID.
The DIRCLOSE CALL routine is passed a directory ID, which is passed to DCLOSE. DIRCLOSE sets the passed
directory ID to missing so that an error occurs if a program tries to use the directory ID after the directory has been
closed. The following code implements the DIROPEN and DIRCLOSE call routines:
proc fcmp outlib=sasuser.funcs.dir;
function diropen(dir $);
length dir $ 256 fref $ 8;
rc = filename(fref, dir);
if rc = 0 then do;
did = dopen(fref);
rc = filename(fref);
end;
else do;
msg = sysmsg();
put msg '(DIROPEN(' dir= ')';
did = .;
end;
return(did);
endsub;
subroutine dirclose(did);
outargs did;
rc = dclose(did);
did = .;
endsub;
GATHERING FILENAMES
File paths are collected by the DIR_ENTRIES CALL routine. DIR_ENTRIES takes:
• a starting directory
• a result array to fill
• an output parameter to fill with the number of entries in the result array
• another output parameter to set to 0 if all pathnames fit in the result array or 1 if some of the pathnames do
not fit into the array.
You do not have to specify the result array in the OUTARGS statement because all arrays are output parameters.
The body of DIR_ENTRIES is almost identical to the code to implement this in a DATA step, yet DIR_ENTRIES is a
real CALL routine that is easily reused in several different programs.
DIR_ENTRIES calls DIROPEN to open a directory and get a directory ID. The routine then calls DNUM to get the
number of entries in the directory. For each entry in the directory, DREAD is called to get the name of the entry. Now
that we've got the entry name, the routine calls MOPEN to determine if the entry is a file or a directory.
If the entry is a file, MOPEN returns a positive value. In this case, the full path to the file is added to the result array.
If the result array is full, the truncation output argument is set to 1.
If the entry is a directory, MOPEN returns a value less than or equal to 0. In this case, DIR_ENTRIES needs to
gather the pathnames for the entries in this subdirectory. This is done by recursively calling DIR_ENTRIES and
passing the subdirectory's path as the starting path. When DIR_ENTRIES returns, the result array contains the paths
of the subdirectory's entries.
subroutine dir_entries(dir $, files[*] $, n, trunc);
outargs files, n, trunc;
length dir entry $ 256;
if trunc then return;
did = diropen(dir);
if did <= 0 then return;
dnum = dnum(did);
do i = 1 to dnum;
entry = dread(did, i);
/* If this entry is a file, then add to array */
/* Else entry is a directory, recurse. */
fid = mopen(did, entry);
entry = trim(dir) || '/' || entry;
if fid > 0 then do;
rc = fclose(fid);
if n < dim(files) then do;
trunc = 0;
n = n + 1;
files[n] = entry;
end;
else do;
trunc = 1;
call dirclose(did);
return;
end;
end;
else
call dir_entries(entry, files, n, trunc);
end;
call dirclose(did);
return;
endsub;
CALLING DIR_ENTRIES FROM A DATA STEP
Invoke DIR_ENTRIES like any other DATA step CALL routine. An array is declared with enough entries to hold all
the files that may be found. Then, the routine is called. Upon return from the routine, the result array is looped over
and each entry in the array is output to the log.
options cmplib=sasuser.funcs;
data _null_;
array files[1000] $ 256 _temporary_;
dnum = 0;
trunc = 0;
call dir_entries("c:/logs", files, dnum, trunc);
if trunc then put 'ERROR: Not enough result array entries. Increase array size.';
do i = 1 to dnum;
put files[i];
end;
run;
This example shows the similarity between FCMP syntax and the DATA step. For instance, numeric expressions and
flow of control statements are identical. The abstraction of DIROPEN into a FCMP function simplifies DIR_ENTRIES.
All of the FCMP routines created can be reused by other DATA steps without any need to modify the routines to work
in a new context.
FCMP AND DATA STEP DIFFERENCES
PROC FCMP was originally developed as a programming language for several SAS/STAT, SAS/ETS, and SAS/OR
procedures. Because the implementation isn't identical to the DATA step, there are differences between the two
languages. In this section, we highlight some of the differences between the two languages. For a complete list of
differences, please refer to The FCMP Procedure (SAS Institute Inc. 2003).
PUT STATEMENT
The syntax for the PUT statement is similar in FCMP and the DATA step, yet their operation can be different. The
FCMP PUT statement is typically used as a debugging tool, not as a report or file creation tool as in the DATA step.
Therefore, FCMP supports some, though not all, features of the DATA step PUT statement and adds features that
are useful for debugging.
The PUT statement in FCMP follows the output of each item with a space. This is similar to list mode output in the
DATA step. Detailed control over column and line position is supported to a lesser extent than in the DATA step.
FCMP's PUT statement evaluates an expression and outputs the result by placing the expression in parentheses.
The DATA step does not have the ability to evaluate expressions in the PUT statement. In the following example, the
expressions x/100 and sqrt(y)/2 are evaluated and output.
put (x/100) (sqrt(y)/2);
IF EXPRESSIONS
An IF expression allows IF…THEN … ELSE conditions to be evaluated within an expression. IF expressions are
supported by FCMP and not by the DATA step. Some expressions can be simplified with IF expressions by not
splitting the expression between IF...THEN ... ELSE statements. For instance, the following two samples of code are
equivalent, yet the IF expression is more succinct:
x = if y < 100 then 1 else 0;
if y < 100 then
x = 1;
else
x = 0;
DATA SET INPUT AND OUTPUT
FCMP does not support the DATA and OUTPUT statements for creating and writing to an output data set. Nor does
it support the SET, MERGE, UPDATE, or MODIFY statements for data set input. Data are typically transferred into
and out of FCMP routines with parameters. If a large amount of data needs to be transferred, one suggestion is to
pass arrays to an FCMP routine.
FILE INPUT AND OUTPUT
FCMP does have PUT and FILE statements, yet the FILE statement is limited to the LOG and PRINT destinations.
There are no INFILE or INPUT statements in FCMP.
ARRAYS
FCMP uses parentheses after a name to represent a function call. When referencing an array, square braces [ ] or
curly braces { } must be used. For an array named ARR, this would look like ARR[i] or ARR{i}. FCMP limits the
number of dimensions for an array to 6.
DO STATEMENT
The DO statement in FCMP doesn't support character loop control variables. While the following code will work in a
DATA step, it will not work in FCMP:
do i = 'a', 'b', 'c';
DATA STEP DEBUGGER
When using the DATA step debugger, FCMP routines are like any other routine, it is not possible to debug the code
in the routine. One way to debug FCMP routines is to use the PUT statement within the routine.
ADDITIONAL FEATURES
There are several noteworthy features not mentioned in this paper that are summarized here. Details of these
features can be found in The FCMP Procedure (SAS Institute Inc. 2003).
• PROC REPORT uses the DATA step to evaluate compute blocks. With the DATA step being able to call
FCMP routines, they are also callable from PROC REPORT compute blocks.
• The FCMP Package Viewer is an application for traversing packages of functions that is located on the
Solutions menu of an interactive SAS session.
• FCMP has the ability to call C and C++ functions that have been registered with PROC PROTO. In addition,
there is support for C structures within FCMP.
• FCMP has a SOLVE function for computing implicit values of a function.
• Many Microsoft Excel functions, not typically available in SAS, have been implemented in FCMP. These can
be found in the SAShelp.Slkwxl data set. These functions were written for PROC SYLK, a Base SAS
procedure for importing and executing SYLK spreadsheets.
The consequent and alternative of an IF expression are expressions. This means that parentheses are used to
group operations instead of DO ... END blocks.Because parentheses are used for expression evaluation, they cannot be used for variable or format lists as in the
DATA step. Other DATA step features not available in FCMP are line pointers, format modifiers, column output, and
features provided by FILE statement options, like DLM= and DSD.Output:
c:/logs/2004/qtr1.log
c:/logs/2004/qtr2.log
c:/logs/2004/qtr3.log
c:/logs/2004/qtr4.log
c:/logs/2005/qtr1.log
c:/logs/2005/qtr2.log
c:/logs/2005/qtr3.log
c:/logs/2005/qtr4.log
c:/logs/2006/qtr1.log
c:/logs/2006/qtr2.logThe DYNAMIC_ARRAY CALL routine is passed the array to be resized, and a new size for each dimension of the
array. In ALLPERMK, a scratch array that is the size of the number of elements being permuted is needed. When
the function is created, this value is not known since it is passed in as parameter N. A dynamic array enables the
routine to allocate the amount of memory that is needed, instead of having to create an array that is large enough to
handle all possible cases.Output:
1 1 1 0 0
1 1 0 1 0
1 1 0 0 1
1 0 1 1 0
1 0 1 0 1
1 0 0 1 1
0 1 1 1 0
0 1 1 0 1
0 1 0 1 1
0 0 1 1 1The formal parameters listed in the OUTARGS statement are passed by reference instead of by value. This means
that any modification of the formal parameter by the CALL routine will modify the original variable that was passed. It
also means the value is not copied when the CALL routine is invoked. Reducing the number of copies can improve
performance when passing large amounts of data or arrays between a CALL routine and the calling environment.