Handling Lists, Arrays, and Other Expressions
The Wolfram Symbolic Transfer Protocol (WSTP) allows you to exchange data of any type with external programs. For more common types of data, you simply need to give appropriate :ArgumentTypes: or :ReturnType: specifications in your WSTP template file.
Wolfram Language specification | C specification | |
Integer | integer | int |
Real | floating‐point number | double |
IntegerList | list of integers | int*,long |
RealList | list of floating‐point numbers | double*,long |
String | character string | char* |
Symbol | symbol name | char* |
Manual | call WSTP routines directly | void |
:Begin:
:Function: h
:Pattern: h[a_List]
:Arguments: {a}
:ArgumentTypes: {IntegerList}
:ReturnType: Integer
:End:
Here is the C source code for the function. Note the extra argument alen which is used to pass the length of the list.
int h(int *a, long alen) {
int i, tot=0;
for(i=0; i<alen; i++)
tot += a[i];
return tot;
}
The pattern is matched, but the elements in the list are of the wrong type for the external code, so $Failed is returned:
You can mix basic types of arguments in any way you want. Whenever you use IntegerList or RealList, however, you have to include an extra argument in your C program to represent the length of the list.
:ArgumentTypes: {IntegerList, RealList, Integer}
void f(int *a, long alen, double *b, long blen, int c)
Note that when a list is passed to a C program by WSTP its first element is assumed to be at position 0, as is standard in C, rather than at position 1, as is standard in the Wolfram Language.
In addition, following C standards, character strings specified by String are passed as char* objects, terminated by ∖0 null bytes. "Portability of WSTP Programs" discusses how to handle special characters.
WSPutInteger32(stdlink,int i) | put a single integer |
WSPutReal64(stdlink,double x) | put a single floating‐point number |
WSPutInteger32List(stdlink,int*a,int n) | |
put a list of n integers starting from location a | |
WSPutReal64List(stdlink,double*a,int n) | |
put a list of n floating‐point numbers starting from location a | |
WSPutInteger32Array(stdlink,int*a,int*dims,NULL,int d) | |
put an array of integers to form a depth d list with dimensions dims | |
WSPutReal64Array(stdlink,double*a,int*dims,NULL,int d) | |
put an array of floating‐point numbers | |
WSPutString(stdlink,char*s) | put a character string |
WSPutSymbol(stdlink,char*s) | put a character string as a symbol name |
WSPutFunction(stdlink,char*s,int n) | |
begin putting a function with head s and n arguments |
When you use a WSTP template file, what mprep and mcc actually do is to create a C program that includes explicit calls to WSTP library functions. If you want to see an example of how to use the WSTP library functions directly, you can look at the source code of this program. Note when you use mcc, you typically need to give a -g option, otherwise the source code that is generated is automatically deleted.
If your external function just returns a single integer or floating‐point number, then you can specify this just by giving Integer or Real as the :ReturnType: in your WSTP template file. But because of the way memory allocation and deallocation work in C, you cannot directly give :ReturnType: specifications such as IntegerList or RealList. And instead, to return such structures, you must explicitly call WSTP library functions within your C program, and give Manual as the :ReturnType: specification.
Here is the WSTP template for a function that takes an integer as an argument, and returns a list of the digits in its binary representation using explicit WSTP functions.
:Begin:
:Function: bits
:Pattern: bits[i_Integer]
:Arguments: {i}
:ArgumentTypes: {Integer}
:ReturnType: Manual
:End:
void bits(int i) {
int a[32], k;
for(k=0; k<32; k++) {
a[k] = i%2;
i >>= 1;
if (i==0) break;
}
if (k<32) k++;
WSPutInteger32List(stdlink, a, k);
return ;
}
If you declare an array in C as int a[n1][n2][n3], then you can use WSPutInteger32Array() to send it to the Wolfram Language as a depth 3 list.
int a[8][16][100];
int dims[] = {8, 16, 100};
WSPutInteger32Array(stdlink, a, dims, NULL, 3);
You can use WSTP functions to create absolutely any Wolfram Language expression. The basic idea is to call a sequence of WSTP functions that correspond directly to the FullForm representation of the Wolfram Language expression.
This sets up the Wolfram Language function Plus with 2 arguments.
WSPutFunction(stdlink, "Plus", 2);
WSPutInteger32(stdlink, 77);
WSPutSymbol(stdlink, "x");
In general, you first call WSPutFunction(), giving the head of the Wolfram Language function you want to create, and the number of arguments it has. Then you call other WSTP functions to fill in each of these arguments in turn. "Expressions" discusses the general structure of Wolfram Language expressions and the notion of heads.
WSPutFunction(stdlink, "List", 2);
WSPutInteger32List(stdlink, r, 10);
WSPutFunction(stdlink, "List", 2);
WSPutReal64(stdlink, 4.5);
WSPutInteger32(stdlink, 11);
WSPutInteger32Array() and WSPutReal64Array() allow you to send arrays which are laid out in memory in the one‐dimensional way that C pre‐allocates them. But if you create arrays during the execution of a C program, it is more common to set them up as nested collections of pointers. You can send such arrays to the Wolfram Language by using a sequence of WSPutFunction() calls, ending with an WSPutInteger32List() call.
int ***a;
WSPutFunction(stdlink, "List", n1);
for (i=0; i<n1; i++) {
WSPutFunction(stdlink, "List", n2);
for (j=0; j<n2; j++) {
WSPutInteger32List(stdlink, a[i][j], n3);
}
}
It is important to realize that any expression you create using WSTP functions will be evaluated as soon as it is sent to the Wolfram Language. This means, for example, that if you wanted to transpose an array that you were sending back to the Wolfram Language, all you would need to do is to wrap a Transpose around the expression representing the array. You can then do this simply by calling WSPutFunction(stdlink,"Transpose",1); just before you start creating the expression that represents the array.
The idea of postprocessing data that you send back to the Wolfram Language has many uses. One example is as a way of sending lists whose length you do not know in advance.
Flatten flattens out the list:
Sequence automatically flattens itself:
In order to call WSPutInteger32List(), you need to know the length of the list you want to send. But by creating a sequence of nested Sequence objects, you can avoid having to know the length of your whole list in advance.
This sets up the List around your result.
WSPutFunction(stdlink, "List", 1);
while( condition ) {
/* generate an element */
Create the next level Sequence object.
WSPutFunction(stdlink, "Sequence", 2);
WSPutInteger32(stdlink, i );
}
This closes off your last Sequence object.
WSPutFunction(stdlink, "Sequence", 0);
WSGetInteger32(stdlink,int*i) |
get an integer, storing it at address
i
|
WSGetReal64(stdlink,double*x) | get a floating‐point number, storing it at address x |
WSTP provides functions like WSPutInteger32() to send data from an external program into the Wolfram Language. WSTP also provides functions like WSGetInteger32() that allow you to get data from the Wolfram Language into an external program.
The list that you give for :ArgumentTypes: in a WSTP template can end with Manual, indicating that after other arguments have been received, you will call WSTP functions to get additional expressions.
:Begin:
:Function: f
:Pattern: f[i_Integer, x_Real, y_Real]
:Arguments: {i, x, y}
:ArgumentTypes: {Integer, Manual}
:ReturnType: Real
:End:
double f(int i) {
double x, y;
WSGetReal64() explicitly gets data from the link.
WSGetReal64(stdlink, &x);
WSGetReal64(stdlink, &y);
return i+x+y;
}
WSTP functions such as WSGetInteger32(link,pi) work much like standard C library functions such as fscanf(fp,"%d",pi). The first argument specifies the link from which to get data. The last argument gives the address at which the data that is obtained should be stored.
WSCheckFunction(stdlink,"name",int*n) | |
check the head of a function and store how many arguments it has |
:Begin:
:Function: f
:Pattern: f[a:{___Integer}]
:Arguments: {a}
:ArgumentTypes: {Manual}
:ReturnType: Integer
:End:
int f(void) {
int n, i;
int a[MAX];
WSCheckFunction(stdlink, "List", &n);
for (i=0; i<n; i++)
WSGetInteger32(stdlink, a+i);
In simple cases, it is usually possible to ensure on the Wolfram Language side that the data you send to an external program has the structure that is expected. But in general the return value from WSCheckFunction() will be MLSUCCESS only if the data consists of a function with the name you specify.
Note that if you want to get a nested collection of lists or other objects, you can do this by making an appropriate sequence of calls to WSCheckFunction().
WSGetInteger32List(stdlink,int**a,int*n) | |
get a list of integers, allocating the memory needed to store it | |
WSGetReal64List(stdlink,double**a,int*n) | |
get a list of floating‐point numbers | |
WSReleaseInteger32List(stdlink,int*a,int n) | |
release the memory associated with a list of integers | |
WSReleaseReal64List(stdlink,double*a,int n) | |
release the memory associated with a list of floating‐point numbers |
When an external program gets data from the Wolfram Language, it must set up a place to store the data. If the data consists of a single integer, as in WSGetInteger32(stdlink,&n), then it suffices just to have declared this integer using intn.
But when the data consists of a list of integers of potentially any length, memory must be allocated to store this list at the time when the external program is actually called.
WSGetInteger32List(stdlink,&a,&n) will automatically do this allocation, setting a to be a pointer to the result. Note that memory allocated by functions like WSGetInteger32List() is always in a special reserved area, so you cannot modify or free it directly.
int f(void) {
int n;
int *a;
WSGetInteger32List(stdlink, &a, &n);
WSReleaseInteger32List(stdlink, a, n);
...
}
If you use IntegerList as an :ArgumentTypes: specification, then WSTP will automatically release the memory used for the list after your external function exits. But if you get a list of integers explicitly using WSGetInteger32List(), then you must not forget to release the memory used to store the list after you have finished with it.
WSGetInteger32Array(stdlink,int**a,int**dims,char***heads,int*d) | |
get an array of integers of any depth | |
WSGetReal64Array(stdlink,double**a,int**dims,char***heads,int*d) | |
get an array of floating‐point numbers of any depth | |
WSReleaseInteger32Array(stdlink,int*a,int*dims,char**heads,int d) | |
release memory associated with an integer array | |
WSReleaseReal64Array(stdlink,double*a,int*dims,char**heads,int d) | |
release memory associated with a floating‐point array |
WSGetInteger32List() extracts a one‐dimensional array of integers from a single Wolfram Language list. WSGetInteger32Array() extracts an array of integers from a collection of lists or other Wolfram Language functions nested to any depth.
The name of the Wolfram Language function at level i in the structure is stored as a string in heads[i]. The size of the structure at level i is stored in dims[i], while the total depth is stored in d.
If you pass a list of complex numbers to your external program, then WSGetReal64Array() will create a two‐dimensional array containing a sequence of pairs of real and imaginary parts. In this case, heads[0] will be "List" while heads[1] will be "Complex".
Note that you can conveniently exchange arbitrary‐precision numbers with external programs by converting them to lists of digits in the Wolfram Language using IntegerDigits and RealDigits.
WSGetString(stdlink,char**s) | get a character string |
WSGetSymbol(stdlink,char**s) | get a symbol name |
WSReleaseString(stdlink,char*s) | release memory associated with a character string |
WSReleaseSymbol(stdlink,char*s) | release memory associated with a symbol name |
If you use String as an :ArgumentTypes: specification, then WSTP will automatically release the memory that is used to store the string after your function exits. This means that if you want to continue to refer to the string, you must allocate memory for it, and explicitly copy each character in it.
If you get a string using WSGetString(), however, then WSTP will not automatically release the memory used for the string when your function exits. As a result, you can continue referring to the string. Be careful not to modify the contents of the string by writing to the memory that is returned by WSGetString(). When you no longer need the string, you must nevertheless explicitly call WSReleaseString() in order to release the memory associated with it.
WSGetFunction(stdlink,char**s,int*n) | |
begin getting a function, storing the name of the head in s and the number of arguments in n | |
WSReleaseSymbol(stdlink,char*s) | release memory associated with a function name |
If you know what function to expect in your external program, then it is usually simpler to call WSCheckFunction(). But if you do not know what function to expect, you have no choice but to call WSGetFunction(). If you do this, you need to be sure to call WSReleaseSymbol() to release the memory associated with the name of the function that is found by WSGetFunction().