Generators are helper programs that output testcases. Most programming task usually require has a large input (for example, an array of up to $$$2 * 10^5$$$ elements, a tree of up to $$$10^5$$$ vertices). In these tasks, adding testcase manually is obviously not sufficient, and generators come to the rescue.

If you are writing a generator in C++, it is recommended to use the testlib.h library.

Types of generators

There are two types of generators:

1. Single-test generator: output exactly one testcase in a single run. Usually, in order to generate several testcases, one must run the generator several times with different command line parameters. Such generators output the testcase to the standard output stream (to the screen).
2. Multiple-test generator: output many testcases in a single run. Such generators output testcases to files (one file for each testcase).

An example single-test generator with testlib.h

The following generator output a pair of integer with each element from $$$1$$$ to $$$n$$$ — where $$$n$$$ is a command line parameter passed to the generator.

#include "testlib.h"
#include <iostream>

using namespace std;

int main(int argc, char* argv[])
{
    registerGen(argc, argv, 1);
    int n = atoi(argv[1]);
    cout << rnd.next(1, n) << " ";
    cout << rnd.next(1, n) << endl;
}

Why testlib.h?

On the surface, it seems that testlib.h is not necessary to write a generator. This is actually not true. Almost all generators need to generate random values, and it is tempted to use rand(). However, this is a bad practice. A basic principle of writing generators is that: a generator must output the same testcase when compiled by any compiler on any platform, if it is run in the same way (using the same command line parameter). When using rand() or C++11 classes like mt19937/uniform_int_distribution, your program will output different testcases when compiled with different compilers.

The random value generator in testlib.h ensures that the same value is generated regardless of the (test) generator and platform. In addition, testlib.h has various conveniences for generating tests, for example, rnd.next("[a-z]{1,10}") will return a random word of length $$$1$$$ to $$$10$$$ from letters a to z.

Translator's note: There are more issues with using rand() aside from the above one. Refer to this blog for detailed explanation about these issues.

Available method

To initialize a testlib generator, the first line of your generator must be of the form registerGen(argc, argv, 1); (where 1 is the version of the random number generator used). After that, it will be possible to use the rnd object, which will be initialized with a hash from all the command line arguments. Thus, the output of g 100 and g.exe "100" will be the same, while g 100 0 will be different.

rnd is of type random_t. That is, you can create your own generator, but usually this is not necessary.

rnd has many useful member functions. Here are some examples:

Also, please do not use std::random_shuffle, use the shuffle from testlib.h instead. It also takes two iterators, but works using rnd.

Translator's note: If my understanding is correct, rnd.wnext is defined as follow:

Example: generating an undirected tree

Below is the code of an undirected tree generator that takes two parameters — the number of vertices and the 'elongation' of the tree. For example:

• For $$$n = 10$$$, $$$t = 1000$$$, a path graph (degree of all vertices are at most $$$2$$$) is likely to be generated
• For $$$n = 10$$$, $$$t = -1000$$$, a star graph (there's only one non-leaf vertex in the tree) is likely to be generated.

registerGen(argc, argv, 1);

int n = atoi(argv[1]);
int t = atoi(argv[2]);

vector<int> p(n);

/* setup parents for vertices 1..n-1 */
forn(i, n)
    if (i > 0)
        p[i] = rnd.wnext(i, t);

printf("%d\n", n);

/* shuffle vertices 1..n-1 */
vector<int> perm(n);
forn(i, n)
    perm[i] = i;
shuffle(perm.begin() + 1, perm.end());

/* put edges considering shuffled vertices */
vector<pair<int,int> > edges;
for (int i = 1; i < n; i++)
    if (rnd.next(2))
        edges.push_back(make_pair(perm[i], perm[p[i]]));
    else
        edges.push_back(make_pair(perm[p[i]], perm[i]));

/* shuffle edges */
shuffle(edges.begin(), edges.end());

for (int i = 0; i + 1 < n; i++)
    printf("%d %d\n", edges[i].first + 1, edges[i].second + 1);

How to write a multiple-test generator?

A multiple-test generator in one execution can output more than one test. Tests by such a generator are output to files. In the generator on testlib.h it is enough to write startTest(test_index) before the test output. This will re-open (freopen) the standard output stream to a file named test_index. Please note that in the Polygon system in this case, you need to write something like multigen a b c > {4-10} in the script (if it is assumed that starting the multi-generator will return tests 4, 5, 6, 7, 8, 9, and 10).

Other notes about generators

• Strictly follow the format of the test — spaces, line breaks should be placed correctly. The test should end with a line feed. For example, if the test consists of a single number, then output it as cout << rnd.next (1, n) << endl; — with a line feed at the end.
• If the test size is large, it is prefered to use printf instead of cout — this will improve the performance of the generator.
• It is better to use cout to output long long, but if you want printf, then use the I64 constant (for example, printf(I64, x);).
• You must not forget about various cases of undefined C++ language behavior. For example, in the generator example above, you cannot combine two cout commands into one, because then the order of the rnd.next function calls is not defined.

Translator's note: about the third point, using lld constant with printf to output long long used to be problematic in the past, but is no longer an issue now.

Further examples

Further examples of generators can be found in the release notes or directly at the GitHub repository.