9.3 Indirect Function Calls

This section describes an advanced, gawk-specific extension.

Often, you may wish to defer the choice of function to call until runtime. For example, you may have different kinds of records, each of which should be processed differently.

Normally, you would have to use a series of if-else statements to decide which function to call. By using indirect function calls, you can specify the name of the function to call as a string variable, and then call the function. Let’s look at an example.

Suppose you have a file with your test scores for the classes you are taking, and you wish to get the sum and the average of your test scores. The first field is the class name. The following fields are the functions to call to process the data, up to a “marker” field ‘data:’. Following the marker, to the end of the record, are the various numeric test scores.

Here is the initial file:

Biology_101 sum average data: 87.0 92.4 78.5 94.9
Chemistry_305 sum average data: 75.2 98.3 94.7 88.2
English_401 sum average data: 100.0 95.6 87.1 93.4

To process the data, you might write initially:

{
    class = $1
    for (i = 2; $i != "data:"; i++) {
        if ($i == "sum")
            sum()   # processes the whole record
        else if ($i == "average")
            average()
        …           # and so on
    }
}

This style of programming works, but can be awkward. With indirect function calls, you tell gawk to use the value of a variable as the name of the function to call.

The syntax is similar to that of a regular function call: an identifier immediately followed by an opening parenthesis, any arguments, and then a closing parenthesis, with the addition of a leading ‘@’ character:

the_func = "sum"
result = @the_func()   # calls the sum() function

Here is a full program that processes the previously shown data, using indirect function calls:

# indirectcall.awk --- Demonstrate indirect function calls

# average --- return the average of the values in fields $first - $last

function average(first, last,   sum, i)
{
    sum = 0;
    for (i = first; i <= last; i++)
        sum += $i

    return sum / (last - first + 1)
}

# sum --- return the sum of the values in fields $first - $last

function sum(first, last,   ret, i)
{
    ret = 0;
    for (i = first; i <= last; i++)
        ret += $i

    return ret
}

These two functions expect to work on fields; thus, the parameters first and last indicate where in the fields to start and end. Otherwise, they perform the expected computations and are not unusual:

# For each record, print the class name and the requested statistics
{
    class_name = $1
    gsub(/_/, " ", class_name)  # Replace _ with spaces

    # find start
    for (i = 1; i <= NF; i++) {
        if ($i == "data:") {
            start = i + 1
            break
        }
    }

    printf("%s:\n", class_name)
    for (i = 2; $i != "data:"; i++) {
        the_function = $i
        printf("\t%s: <%s>\n", $i, @the_function(start, NF) "")
    }
    print ""
}

This is the main processing for each record. It prints the class name (with underscores replaced with spaces). It then finds the start of the actual data, saving it in start. The last part of the code loops through each function name (from $2 up to the marker, ‘data:’), calling the function named by the field. The indirect function call itself occurs as a parameter in the call to printf. (The printf format string uses ‘%s’ as the format specifier so that we can use functions that return strings, as well as numbers. Note that the result from the indirect call is concatenated with the empty string, in order to force it to be a string value.)

Here is the result of running the program:

$ gawk -f indirectcall.awk class_data1
-| Biology 101:
-|     sum: <352.8>
-|     average: <88.2>
-|
-| Chemistry 305:
-|     sum: <356.4>
-|     average: <89.1>
-|
-| English 401:
-|     sum: <376.1>
-|     average: <94.025>

The ability to use indirect function calls is more powerful than you may think at first. The C and C++ languages provide “function pointers,” which are a mechanism for calling a function chosen at runtime. One of the most well-known uses of this ability is the C qsort() function, which sorts an array using the famous “quicksort” algorithm (see the Wikipedia article for more information). To use this function, you supply a pointer to a comparison function. This mechanism allows you to sort arbitrary data in an arbitrary fashion.

We can do something similar using gawk, like this:

# quicksort.awk --- Quicksort algorithm, with user-supplied
#                   comparison function

# quicksort --- C.A.R. Hoare's quicksort algorithm. See Wikipedia
#               or almost any algorithms or computer science text.

function quicksort(data, left, right, less_than,    i, last)
{
    if (left >= right)  # do nothing if array contains fewer
        return          # than two elements

    quicksort_swap(data, left, int((left + right) / 2))
    last = left
    for (i = left + 1; i <= right; i++)
        if (@less_than(data[i], data[left]))
            quicksort_swap(data, ++last, i)
    quicksort_swap(data, left, last)
    quicksort(data, left, last - 1, less_than)
    quicksort(data, last + 1, right, less_than)
}

# quicksort_swap --- helper function for quicksort, should really be inline

function quicksort_swap(data, i, j,      temp)
{
    temp = data[i]
    data[i] = data[j]
    data[j] = temp
}

The quicksort() function receives the data array, the starting and ending indices to sort (left and right), and the name of a function that performs a “less than” comparison. It then implements the quicksort algorithm.

To make use of the sorting function, we return to our previous example. The first thing to do is write some comparison functions:

# num_lt --- do a numeric less than comparison

function num_lt(left, right)
{
    return ((left + 0) < (right + 0))
}

# num_ge --- do a numeric greater than or equal to comparison

function num_ge(left, right)
{
    return ((left + 0) >= (right + 0))
}

The num_ge() function is needed to perform a descending sort; when used to perform a “less than” test, it actually does the opposite (greater than or equal to), which yields data sorted in descending order.

Next comes a sorting function. It is parameterized with the starting and ending field numbers and the comparison function. It builds an array with the data and calls quicksort() appropriately, and then formats the results as a single string:

# do_sort --- sort the data according to `compare'
#             and return it as a string

function do_sort(first, last, compare,      data, i, retval)
{
    delete data
    for (i = 1; first <= last; first++) {
        data[i] = $first
        i++
    }

    quicksort(data, 1, i-1, compare)

    retval = data[1]
    for (i = 2; i in data; i++)
        retval = retval " " data[i]

    return retval
}

Finally, the two sorting functions call do_sort(), passing in the names of the two comparison functions:

# sort --- sort the data in ascending order and return it as a string

function sort(first, last)
{
    return do_sort(first, last, "num_lt")
}

# rsort --- sort the data in descending order and return it as a string

function rsort(first, last)
{
    return do_sort(first, last, "num_ge")
}

Here is an extended version of the data file:

Biology_101 sum average sort rsort data: 87.0 92.4 78.5 94.9
Chemistry_305 sum average sort rsort data: 75.2 98.3 94.7 88.2
English_401 sum average sort rsort data: 100.0 95.6 87.1 93.4

Finally, here are the results when the enhanced program is run:

$ gawk -f quicksort.awk -f indirectcall.awk class_data2
-| Biology 101:
-|     sum: <352.8>
-|     average: <88.2>
-|     sort: <78.5 87.0 92.4 94.9>
-|     rsort: <94.9 92.4 87.0 78.5>
-|
-| Chemistry 305:
-|     sum: <356.4>
-|     average: <89.1>
-|     sort: <75.2 88.2 94.7 98.3>
-|     rsort: <98.3 94.7 88.2 75.2>
-|
-| English 401:
-|     sum: <376.1>
-|     average: <94.025>
-|     sort: <87.1 93.4 95.6 100.0>
-|     rsort: <100.0 95.6 93.4 87.1>

Another example where indirect functions calls are useful can be found in processing arrays. This is described in Traversing Arrays of Arrays.

Remember that you must supply a leading ‘@’ in front of an indirect function call.

Starting with version 4.1.2 of gawk, indirect function calls may also be used with built-in functions and with extension functions (see section Writing Extensions for gawk). There are some limitations when calling built-in functions indirectly, as follows.

• You cannot pass a regular expression constant to a built-in function through an indirect function call.⁶³ This applies to the sub(), gsub(), gensub(), match(), split() and patsplit() functions.
• If calling sub() or gsub(), you may only pass two arguments, since those functions are unusual in that they update their third argument. This means that $0 will be updated.

gawk does its best to make indirect function calls efficient. For example, in the following case:

for (i = 1; i <= n; i++)
    @the_func()

gawk looks up the actual function to call only once.