A collection of useful tips, tricks and traps. Feel free to contribute by sending an email or making an Ed post. In addition to the below, please refer to this CPMSoc blog post.

A useful Makefile

Create a file called Makefile in your directory (or just home directory for exams) with these contents (copied verbatim)

CXX = g++
CPPFLAGS = -Wall -Wextra -Wshadow -Wformat=2 -Wfloat-equal -Wconversion -Wlogical-op \
           -D_GLIBCXX_DEBUG -D_GLIBCXX_DEBUG_PEDANTIC -D_FORTIFY_SOURCE=2 \
           -fsanitize=address -fsanitize=undefined -fno-sanitize-recover -fstack-protector \
           -O2 --std=c++17 -g

These extra flags do a mix of static analysis and extra run time checks that catch some common mistakes in your program. Many instances of undefined behavior now at least trigger a warning when you run the program, instead of (worst case) silently working.

This is not a panacea though. It won’t catch anywhere near all bugs, but it catches some particularly simple ones.

One extra warning: these flags slow down execution speed dramatically. So switch to something more light weight if you need to run a program for a while or you want to get a gauge of the running time of a program.

A brute force comparison script

For many problems, a useful way of debugging is to write a brute force and a random data generator and repeatedly generate data and compare the brute force’s output to your program’s output until there’s a mismatch. This is particularly useful for problems with subtasks where you can use the subtask solution as your brute force.

For this, I will assume you have:

  • Your solution is an executable named soln.
  • You have a brute force executable named brute_force.
  • A random data generator titled gen. I usually write these in Python.

Then to generate random data until there’s a mismatch, just run

./gen > random.in; while diff <(./soln < random.in) <(./brute_force < random.in); do echo "all good"; ./gen > random.in; done

If this ever stops, a mismatch has been found and the input responsible is stored in random.in.

Speeding up IO

Iostream

Iostream and cin/cout by default are quite a bit slower than scanf/printf. This occasionally matters for problems with a lot of input. To speed up input, add the following lines directly below main:

int main() {
    cin.tie(nullptr);
    cin.sync_with_stdio(false);

The former stops cout from flushing every time cin tries to read (having cout flush is desirable for interactive programs but it introduces unnecessary flushing in our case).

The latter unsyncs iostream from cstdio. Calling this speeds up input even if you aren’t using cstdio but you should not then use C style input in your program.

Endl

There are 2 common ways to print a new line when using cin, either << '\n'; or << endl;

The difference is the latter one flushes output. Again this is desirable for interactivity but frequent flushing is extremely slow. For competitions, you should generally use '\n' not endl.

Common Traps

Overflows

  • Whenever you find yourself needing to multiply ints, check if you should use long longs instead.
  • If you are working mod a large prime (usually 1000000007), check you mod at each step. This is easiest if you define a helper function.

Long data type

  • In C++, the long data type is only guaranteed to be at least 32 bits.
  • Never use long. Always use either int for a 32 bit integer or long long for a 64 bit integer.

Mod negative numbers

  • C++ % is a bit weird, -5 % 3 == -2, not 1 as you might expect. If you need to do arithmetic modulo m, you might want to use helper functions along the lines of

long long madd(long long a, long long b) { return ((a + b) % m + m) % m; }

long long msub(long long a, long long b) { return ((a - b) % m + m) % m; }

Nested loops

Iterating through a 2D array row-by-row

for (int i = 0; i < m; i++)
  for (int j = 0; j < n; j++)
    total += a[i][j];

is significantly faster than column-by-column

for (int j = 0; j < n; j++)
  for (int i = 0; i < m; i++)
    total += a[i][j];

as the entries of a row are stored contiguously in memory. This can be the difference between a correct solution and one that times out!

Printing floating-point numbers

The simplest way to print floating-point numbers to a desired number of decimal places is to use C-style output: from the <cstdio> header, simply printf("%.9lf\n", x);.

There are a couple of common pitfalls using C++-style output. If x is a double, cout << x << '\n' prints it with six digits of precision by default. This is often insufficient, and in particular if x exceeds 1,000,000, then it will be printed in scientific notation (yes, really).

  • To insist on fixed-point notation (i.e. not scientific notation), use the manipulator std::fixed from the <ios> header (note: this is included within <iostream>).
  • To change the number of decimal places, use the manipulator std::setprecision() from the <iomanip> header.

So to print x to 9 decimal places using C++-style output, we can write

#include <iostream>
#include <iomanip>
using namespace std;

int main (void) {
    /*
        ...
    */
    cout << fixed << setprecision(9) << x << '\n';
}

Binary search on red-black trees

Recall the lower_bound() function from <algorithm>.

If a is a std::vector whose elements are in increasing order, then

lower_bound(a.begin(), a.end(), x)

finds an iterator to the first element which compares >= x in logarithmic time.

However, the same function runs in linear time on a std::set! This is because set iterators do not support random access, which is necessary for the binary search.

Instead, std::set includes a member function lower_bound() which navigates the underlying binary search tree in logarithmic time. Therefore, for a set b, we should instead call b.lower_bound(x).

The other functions in <algorithm> making use of binary search translate similarly.

Vector Set
binary_search(a.begin(), a.end(), x) b.find(x) != b.end()
lower_bound(a.begin(), a.end(), x) b.lower_bound(x)
upper_bound(a.begin(), a.end(), x) b.upper_bound(x)
equal_range(a.begin(), a.end(), x) b.equal_range(x)

Similar functions are available for std::multiset, std::map and std::multimap.

Unordered data structures have member functions find() and equal_range() which run in expected constant but worst case linear time.

Oddities with DOMJudge

First, note that you only get one verdict. If you fail different test cases for different reasons, there’s no guarantee which of the applicable verdicts you will get. So just because you have a TIME-LIMIT` verdict, it doesn’t mean your program doesn’t also have a runtime error or wrong answer.

RUN-ERROR

This is the most mysterious verdict. It essentially includes all possible runtime errors for your program. These include, but are not limited to:

  • going out of bounds.
  • exceeding memory limits (by default you are given 512 MB).

and strangely enough:

  • your program returning a non-zero exit code.

COMPILE-ERROR

Your program will be compiled with gcc 10.2.1, using the following options: g++ -x c++ -Wall -O2 -std=c++17 -static -pipe.

A program that compiles on your machine may still get this verdict for a couple of reasons.

  • Your compiler may include some headers, such as algorithm, without the header being explicitly specified in your program. The compiler error message will usually refer to some unknown function or type, which you can easily fix.
  • Some more cryptic errors refer to /usr/bin/ld, which is the linker, not the compiler. Your program may well be completely valid C++! Such errors are typically raised when your program tries to allocate memory on the heap in excess of the memory limit. For example, this will occur if you declare a 100,000 by 100,000 global array of ints (say, as the adjacency matrix of a large graph).