STL Algorithms In Action

STL Algorithms in Action
STL Algorithms and their
everyday application
Michael VanLoon
CPPcon 2015

What are algorithms?
al∙go∙rithm
noun: algorithm; plural noun: algorithms
A procedure for solving a … problem (as of
finding the greatest common divisor) in a finite
number of steps that frequently involves
repetition of an operation; broadly: a step-by-
step procedure for solving a problem or
accomplishing some end especially by a
computer.
Merriam-Webster dictionary
Michael VanLoon - http://codeache.net - CPPcon 2015 2

What are STL algorithms?
• A pre-built library of general-purpose algorithms
designed to solve specific problems
• Come “for free” with your C++ compiler
• Operate on sequences or sequenced containers
• Declarative in syntax – no explicit (“raw”) loops
• Iterate over some or all members of a sequence
performing an operation on each element in turn
• Designed by experts and proven bug-free by the
millions of lines of other peoples’ programs that
already use them!

What is a raw loop anyway?
• It’s a for, while, or do… while loop
• Explicitly coded
• Often contains many lines of code (should it?)
• May cause side-effects outside its scope
vector<int> out;
bool found = false;
for (const auto& i: v) {
if (i >= 42) {
out.emplace_back(i);
++global_count;
if (i == 42) {
found = true;
}
}
}

Why use algorithms?
• Often more efficient than hand-written loop
• Cleaner and more clearly abstracted than a
raw loop
• Contains side-effects inside a clear interface
• Prevents accidental leakage of side-effects
• Eases reasoning about functionality and
reasoning about post conditions
• Less likely to fail under non-obvious conditions
• Eases reasoning about the surrounding code

Classes of STL algorithms
• Non-modifying sequence operations (25.2)
• Mutating sequence operations (25.3)
• Sorting and related operations (25.4)
• General C algorithms (25.5)
• General numeric operations (26.7)
(section of the C++ standard INCITS/ISO/IEC 14882-2011[2012])

STL algorithms
non-modifying sequence operations
• Do not modify the input sequence.
• Do not emit a result sequence.
• Algorithm will not cause side-effects in
input sequence.
• Function object, if present, may cause side-
effects by modifying itself, the sequence (in
certain cases, e.g. for_each), or its
environment.

STL algorithms
• all_of
• any_of
• none_of
• for_each
• find
– find_if
– find_if_not
• find_end
• find_first_of
• adjacent_find
• count
– count_if
• mismatch
• equal
• is_permutation
• search
– search_n

STL algorithms
mutating sequence operations
• Do not modify the input sequence, except in situations
when output overlaps input, resulting in modification in-
place (e.g. transform).
• Emit an output sequence of results.
• Output sequence may potentially overlap input
sequence for certain algorithms (e.g. transform).
Others (e.g. copy) explicitly disallow overlap/in-place.
• Algorithm will explicitly cause side-effects in output
sequence.
• Function object, if present, may cause side-effects by
modifying itself or its environment. Function object
should not modify the input or output sequences.

STL algorithms
• copy
– copy_n
– copy_if
– copy_backward
• move
– move_backward
• swap_ranges
– iter_swap
• transform
• replace
– replace_if
– replace_copy
– replace_copy_if
• fill
– fill_n
• generate
– generate_n
• remove
– remove_if
– remove_copy
– remove_copy_if
• unique
– unique_copy
• reverse
– reverse_copy
• rotate
– rotate_copy
• shuffle
– random_shuffle
• partition
– is_partitioned
– stable_partition
– partition_copy
– partition_point

STL algorithms
sorting and related operations
• A mix of non-modifying and mutating
operations
• Mutating operations modify sequences in
place (e.g. sort, make_heap), or emit output
to an output sequence (e.g. merge,
partial_sort_copy)
• Default compare function is operator<
• Explicit compare function object, if supplied,
must not modify the sequence or iterators

STL algorithms
• sorting
– sort
– stable_sort
– partial_sort
– partial_sort_copy
• nth_element
• binary search
– lower_bound
– upper_bound
– equal_range
– binary_search
• merge
– merge
– inplace_merge
• set operations on sorted
structures
– includes
– set_union
– set_intersection
– set_difference
– set_symmetric_difference
• heap operations
– push_heap
– pop_heap
– make_heap
– sort_heap
• minimum and maximum
– min
– max
– minmax
– min_element
– max_element
– minmax_element
• lexicographical comparisons
– lexicographical_compare
• permutation generators
– next_permutation
– prev_permutation

STL algorithms
general numeric operations
• Library of algorithms for doing numeric
operations.
• Consist of components for complex number
types, random number generation, numeric
(n-at-a-time) arrays, generalized numeric
algorithms, and facilities included from the ISO
C library.1
1Description from the standard library that is surprisingly understandable
by humans.

STL algorithms
• accumulate
• inner_product
• partial_sum
• adjacent_difference
• iota

STL algorithms
C library algorithms
• These are shown for completeness.
• You may need to know about these for
legacy reasons.
• In general, there is nothing these can do
that you can’t do better with the modern
algorithms previously mentioned.

STL algorithms
C library algorithms
• bsearch
• qsort

for_each and transform
• Your go-to generic algorithms for doing
general things to sequences
• Applies an operation to each element in a
sequence, in order
• Very similar except completely different

for_each
• Applies an operation to each element in a sequence (like many
algorithms)
• Is a non-modifying sequence operation
• Algorithm produces no side-effect
• Function object may produce a side-effect by modifying the
input sequence
• Function object may produce a side-effect by modifying itself
• Returns a moved copy of the function object
• for_each is considered non-modifying because it produces
no output range; it relies on the function object for mutation, if
any

transform
• Applies an operation to each element in a sequence (like many
algorithms)
• Is a mutating sequence operation
• If the input range(s) and result range are the same, or overlap,
mutates objects in-place
• Algorithm explicitly produces a side-effect
• Function object may not produce a side-effect
• transform is considered mutating because it explicitly
produces an output range modified by applying the function
object to elements, and forbids the function object from
modifying any of the range elements or iterators
• Returns iterator pointing one past last element in result range

for_each example
Generate a single hash for all strings in a vector
struct HashString {
void operator()(const string& s) {
hash = accumulate(s.begin(), s.end(), hash, hash_char);
}
uint32_t hash = 0;
};

accumulate
• Is a non-modifying numerics operation
• Algorithm produces no side-effect
• Function object may not modify the sequence or the iterator
• Function object may produce a side-effect by returning a return
code different from input parameter
• accumulate differs from for_each in that the algorithm
carries a value rather than a function object from visit to visit,
applying the operation to each element and the current value
• accumulate differs from for_each in that it has a default
operation: operator+

for_each example
struct HashString {
}
uint32_t hash = 0;
};

for_each example
struct HashString {
}
uint32_t hash = 0;
};
template<typename Cont>
uint32_t hash_all_strings(const Cont& v) {
const auto hasher = for_each(v.begin(), v.end(), HashString());
return hasher.hash;
}

for_each example
struct HashString {
}
uint32_t hash = 0;
};
uint32_t hash_all_strings(const Cont& v) {
const auto hasher = for_each(v.begin(), v.end(), HashString());
return hasher.hash;
}
void test_for_each_hash() {
vector<string> v{ "one", "two", "three", "four", "five" };
uint32_t hash = hash_all_strings(v);
cout << "Hash: " << hash << dec << endl;
}

for_each example (cont)
… the rest of the code…
uint32_t rotl(uint32_t value, unsigned int count) {
const uint32_t mask =
(CHAR_BIT * sizeof(value) - 1);
count &= mask;
return (value << count) |
(value >> ((-count)&mask));
}
uint32_t hash_char(uint32_t hash, char c)
{
hash = rotl(hash, c); // circular rotate left
hash ^= c;
return hash;
}

transform example
Generate hash for each string in a vector
uint32_t hash_string(const string& s) {
return accumulate(s.begin(), s.end(), 0, hash_char);
};

transform example
};
vector<uint32_t> hash_each_string(const Cont& v) {
vector<uint32_t> res;
transform(v.begin(), v.end(), back_inserter(res), hash_string);
return res;
}

transform example
};
vector<uint32_t> hash_each_string(const Cont& v) {
vector<uint32_t> res;
transform(v.begin(), v.end(), back_inserter(res), hash_string);
return res;
}
void test_transform_hash() {
vector<string> v{ "one", "two", "three", "four", "five" };
auto res = hash_each_string(v);
cout << "Hashes: ";
for_each(res.begin(), res.end(),
[](uint32_t rh){ cout << rh << " "; });
cout << endl;
}

any_of, all_of, and none_of
• Apply a function object to a sequence
• Determines whether any, all, or none of the
elements in the sequence are true as
determined by the function object
• May return before evaluating all elements
in sequence if outcome is determined early

all_of example
validate http headers
static const regex
reHeader("([A-Za-z0-9!#$%&'*+.^_`|~-]+): *(.+) *");
inline bool
headers_valid(const vector<string>& headers) {
return all_of(headers.begin(), headers.end(),
[](const auto& header) -> bool {
smatch matches;
return regex_match(header, matches, reHeader);
}
);
}

all_of example
validate http headers test and output
void all_of_headers() {
vector<string> h1 = { "Foo: bar", "Content-type: application/json",
"Accept: text/html,text/json,application/json" };
cout << "headers valid: " << boolalpha << headers_valid(h1) << endl;
vector<string> h2 = { "Foo : bar", "Content-type: application/json",
":Accept: text/html,text/json,application/json" };
vector<string> h4 = { "Foo: bar", " Content-type: application/json"
}
output:
headers valid: true
headers valid: false

any_of example
http header search
inline bool
header_search(const vector<string>& headers,
const string& find_header, const string& find_value)
{
return any_of(headers.begin(), headers.end(),
[&find_header, &find_value](const auto& header) -> bool {
const regex reHeader(
"(" + find_header + "): *(" + find_value + ") *",
regex::icase);
smatch matches;
return regex_match(header, matches, reHeader);
}
);
}

any_of example
http header search test and output
void any_of_headers_simple() {
"X-SuperPower: toestrength",
cout << "headers valid: " << boolalpha
<< header_search(h1, "X-SuperPower", "toestrength") << endl;
"X-SuperPower: supersmell",
<< header_search(h2, "X-SuperPower", "toestrength") << endl;
vector<string> h3 = { "Foo : bar", "Content-type: application/json",
"X-SuperPower: toestrength",
<< header_search(h3, "X-Superpower", "toeStrength") << endl;
}
output:
headers valid: true
headers valid: true

another for_each example
simultaneously validate and search http headers
struct HeaderData {
int good_headers = 0;
int bad_headers = 0;
multimap<string, string> found_headers;
string find_header;
string find_value;
operator bool() const { return !bad_headers && good_headers > 0; }
void operator() (const string& header) {
static const regex reValid("([A-Za-z0-9!#$%&'*+.^_`|~-]+): *(.+) *");
smatch matches;
bool match = regex_match(header, matches, reValid);
if (match) {
++good_headers;
const regex reHeader("(" + find_header + "): *(" + find_value + ") *", regex::icase);
if (regex_match(header, matches, reHeader)) {
found_headers.emplace(matches[1], matches[2]);
}
} else {
++bad_headers;
}
}
};

simultaneously validate and search http headers
struct HeaderData {
int good_headers = 0;
int bad_headers = 0;
multimap<string, string> found_headers;
string find_header;
string find_value;
operator bool() const;
void operator() (const string& header);
};
const HeaderData header_parse(const vector<string>& headers, const string&
find_header, const string& find_value) {
HeaderData hd;
hd.find_header = find_header;
hd.find_value = find_value;
return for_each(headers.begin(), headers.end(), hd);
}

simultaneous validate/search test
void any_of_headers_full() {
{
vector<string> h1 = { "Foo: bar", "Content-type: application/json", "X-SuperPower: toestrength",
const HeaderData& hd = header_parse(h1, "X-SuperPower", "toestrength");
cout << "headers parse: " << hd << ", good " << hd.good_headers << ", bad " << hd.bad_headers;
for_each(hd.found_headers.begin(), hd.found_headers.end(), [](const auto& val) {
cout << "nt'" << val.first << "', '" << val.second << "'";
});
cout << endl;
}
{
vector<string> h2 = { "Foo: bar", "Content-type: application/json", "X-SuperPower: supersmell", "Accept: text/ht
ml,text/json,application/json" };
});
cout << endl;
}
{
vector<string> h3 = { "Foo : bar", "Content-type: application/json", "X-Superpower: toestrength", "Accept: text/
html,text/json,application/json" };
});
cout << endl;
}
}

simultaneous validate/search output
output:
headers parse: true, good 4, bad 0
'X-SuperPower', 'toestrength'
headers parse: true, good 4, bad 0
headers parse: false, good 3, bad 1
'X-Superpower', 'toestrength'

adjacent_find
• adjacent_find searches for adjacent items
(pairs of elements next to each other) in a
sequence that meet a certain condition.
• Returns an iterator to the first of the pair of
elements meeting the condition.
• The default condition is equality (i.e. find two
adjacent items that are equal).
• A custom comparator may be provided to look
for other adjacent conditions.

adjacent_find example
simple is_sorted implementation
vecInt_t v{ 1, 2, 3, 4, 5, 5, 6, 7, 8 };
// Greater works because it's asking if the first value is
// greater than the second value. If so, then the test
// fails (not sorted). If the first value is less than or
// equal to the second value, no match and success.
vecInt_t::iterator it =
adjacent_find(v.begin(), v.end(), greater<int>());
if (it == v.end())
cout << "Vector is sorted" << endl;
else
cout << "Vector not sorted, value " << *(it + 1)
<< ", at position " << it - v.begin() + 1 << endl;
output:
Vector is sorted
Vector not sorted, value 3, at position 9

test for sequence deviation
typename Cont::const_iterator checkDeviation(const Cont& cont,
double allowed_dev)
{
return adjacent_find(cont.begin(), cont.end(),
[allowed_dev](const typename
Cont::value_type& v1,
const typename
Cont::value_type& v2)
{
auto limit = v1 * allowed_dev;
return (v2 > v1 + limit) ||
(v2 < v1 - limit);
});
}

test for sequence deviation test and output
vecDbl_t v{ 1.0, 1.05, 1.06, 1.04, 1.09, 1.15, 1.2 };
vecDbl_t::const_iterator it = checkDeviation(v, 0.1);
if (it == v.end())
cout << "Vector is within deviation limits" << endl;
else
cout << "Vector outside deviation limits, values " << *it << " and ”
<< *(it + 1) << ", at position " << it - v.begin() + 1 << endl;
v.push_back(2.0);
it = checkDeviation(v, 0.1);
if (it == v.end())
cout << "Vector is within deviation limits" << endl;
else
cout << "Vector outside deviation limits, values " << *it << " and ”
<< *(it + 1) << ", at position " << it - v.begin() + 1 << endl;
output:
Vector is within deviation limits
Vector outside deviation limits, values 1.2 and 2, at position 7

remove_if (with erase)
Scott Meyers, “Effective STL,” items 9 and 32
• Scenario: you want to erase several items
from a container that meet a condition
• You could write a loop with some checks,
some explicit erases, and potential iterator
invalidation
• Or…

remove_if (with erase)
Scott Meyers, “Effective STL,” items 9 and 32
struct Foo {
bool expired;
… other members…
};
vector<Foo> v;
v.erase(
remove_if(v.begin(), v.end(),
[](const Foo& foo){ return foo.expired; }),
v.end());
Output:
before: [val: 1, expired: false] [val: 2, expired: true] [val: 3,
expired: false] [val: 4, expired: false] [val: 5, expired: true]
after: [val: 1, expired: false] [val: 3, expired: false] [val: 4,
expired: false]

Know your sorts and sorta-sorts
Scott Meyers’ “Effective STL” Item 31
• Sorting algorithms:
– sort1
– stable_sort1
– partial_sort, partial_sort_copy1
• Sorta-sorts:
– nth_element1
– partition, partition_copy2
– stable_partition2
1 Requires random access iterators
2 Requires bidirectional iterators

Know your sorts
• sort
– Most general-purpose sort
– Order of equivalent items
implementation-defined
– In some cases, may be more
efficient than stable_sort
since equivalent items can be
rearranged at sort’s discretion
– Sorts in place
• stable_sort
– Order of equivalent items
preserved
– Sorts in place
• partial_sort
– Sort a subset of a sequence,
drawing from a subset that is
equal to or larger than the sorted
sequence
– There is no stable version of
partial_sort
• partial_sort_copy
– Like partial_sort, but…
– Sorts specified subset of an
input sequence, emitting to an
output sequence

sort and stable_sort
sort objects by last name, first, middle
• Scenario: assume an object with strings
containing fist name, middle name, and last
name, among other things
struct Person {
string first;
string middle;
string last;
… other Person stuff…
};
• We want to sort said objects by all three fields,
with precedence last > first > middle

vector<Person> v{{
{ "Joe", "P", "Smith" },
{ "Jane", "Q", "Jones" },
{ "Frank", "P", "Johnson" },
{ "Sarah", "B", "Smith" },
{ "Joe", "X", "Jones" },
{ "Joe", "A", "Smith" } }};
// Sort by least influential data first
sort(v.begin(), v.end(),
[](const Person& a, const Person& b){ return a.middle < b.middle; });

vector<Person> v{{
{ "Joe", "P", "Smith" },
{ "Joe", "X", "Jones" },
{ "Joe", "A", "Smith" } }};
stable_sort(v.begin(), v.end(),
[](const Person& a, const Person& b){ return a.first < b.first; });

vector<Person> v{{
{ "Joe", "P", "Smith" },
{ "Joe", "X", "Jones" },
{ "Joe", "A", "Smith" } }};
[](const Person& a, const Person& b){ return a.first < b.first; });
[](const Person& a, const Person& b){ return a.last < b.last; });
// Sort by most influential data last

visual: sort objects by last name, first, middle

partial_sort
• Takes an input sequence of sort candidates
• Sorts the top n elements into a potentially smaller
output sequence
• Order of items from input sequence that are
unsorted in output sequence are in implementation
defined order
• partial_sort is an in-place operation
• partial_sort_copy copies sorted output to a
separate sequence
• partial_sort is obviously more efficient than a
full sort

partial_sort example
vector<int> v1{ 42, 17, 89, 22,
34, 78, 63, 12,
57, 99 };
partial_sort(v1.begin(),
v1.begin() + 5,
v1.begin() + 8,
greater<int>());

vector<int> v1{ 42, 17, 89, 22,
34, 78, 63, 12,
57, 99 };
v1.begin() + 5,
v1.begin() + 8,
greater<int>());

visual
42
17
89
22
34
78
63
12
57
99
89
78
63
42
34
17
22
12
57
99
ﬁrst
last
ﬁrst
middle
sort input range
sort output range

Know your sorta-sorts
• partition
– Reorder sequence so all
items before partition point
are less, and all items after
are not less
– Order of items in lower and
upper subsets is
implementation defined
– Order of equivalent items is
– Operates in place
• stable_partition
– Like partition, but…
preserved
• partition_copy
– Like partition, but…
– Copies items from input
sequence to one of two
output sequences depending
on whether supplied function
object returns false or true for
each item
– There is no
stable_partition_copy

Know your sorta-sorts, cont.
• nth_element
– Reorders sequence such that all items before “nth”
are less, and all items after “nth” are not less
– Order of items in lower and upper subsets is
implementation defined
– Order of equivalent items is implementation-
defined
– “nth” element is exactly the value that would exist
in a fully sorted sequence (but without fully sorting)

Know your sorta-sorts, cont.
Comparison between partition and nth_element
partition
• partitions a sequence, based
on condition
• partition point is not
guaranteed to be value that
would be at that position in
fully sorted sequence
• input:
– sequence begin, end
– comparison function
• output:
– reordered sequence
– iterator to partition point
nth_element
• partitions a sequence, based
on position
• nth element is element that
would exist in that position in
fully sorted sequence
• input:
– sequence begin, end
– iterator to “nth” position
– optional comparison function
• output:
– reordered sequence, partitioned
around nth element

partition example
partition elements around 50
vector<int> v{ 12, 89, 31, 18,
7, 72, 69, 50,
49, 50, 51, 49 };
vector<int>::iterator part_it =
partition(v.begin(),
v.end(),
[](const int i) {
return i < 50;
});

partition example
vector<int> v{ 12, 89, 31, 18,
7, 72, 69, 50,
49, 50, 51, 49 };
vector<int>::iterator part_it =
partition(v.begin(),
v.end(),
[](const int i) {
return i < 50;
});

partition example
visual: partition elements around 50
12
89
31
18
7
72
69
50
12
49
31
18
7
49
69
50
partition
point
49
50
51
49
72
50
51
89

partition_copy example
vector<int> v{ 12, 89, 31, 18,
7, 72, 69, 50,
49, 50, 51, 49 };
partition_copy(v.begin(),
v.end(),
back_inserter(smaller),
back_inserter(larger),
[](const int i) {
return i < 50;
});

partition_copy example
visual: partition elements around 50

Mixing up partitions!
partition_copy, partition, and stable_partition example
• Extend our prior Person struct to be an Employee
struct
• Partition Management and Individuals into two
separate containers
• Partition Management between Executives and
Managers
• Partition Architects before other Individuals
• Stable partition the Senior and Junior employees
among already partitioned individuals, leaving
already partitioned elements in same relative
position

some data…
struct Employee: public Person {
enum class EmployeeType {
Executive, Manager, Architect, Senior, Junior,
Management = Manager,
Individual = Architect
};
EmployeeType type;
… additional members and methods…
};
vector<Employee> v{
{ "Joe", "P", "Smith", Employee::EmployeeType::Manager },
{ "Jane", "Q", "Jones", Employee::EmployeeType::Junior },
{ "Frank", "P", "Johnson", Employee::EmployeeType::Architect },
{ "Sarah", "B", "Smith", Employee::EmployeeType::Executive },
{ "Joe", "X", "Jones", Employee::EmployeeType::Senior },
{ "Joe", "A", "Smith", Employee::EmployeeType::Junior },
{ "Chris", "M", "Williams", Employee::EmployeeType::Manager }
};

separate employees into management and individuals
vector<Employee> management, individuals;
partition_copy(v.begin(), v.end(),
back_inserter(management), back_inserter(individuals),
[](const Employee& e) {
return e.type <= Employee::EmployeeType::Manager;
});
vector<Employee>::iterator car_it =
partition(management.begin(), management.end(),
return e.type == Employee::EmployeeType::Executive;
});
vector<Employee>::iterator jet_it = management.begin();
cout << "Management partitioned:n";
for (auto it = management.begin(); it != management.end(); ++it) {
cout << (it < car_it ? "tjet" : "tcar") << *it;
}
executives get the company jet; managers get the company car

architects get the company segway; everyone else gets the bike
vector<Employee>::iterator bike_it =
partition(individuals.begin(), individuals.end(),
return e.type == Employee::EmployeeType::Architect;
});
vector<Employee>::iterator segway_it = individuals.begin();
cout << "Architects partitioned:n";
for (auto it = individuals.begin(); it != individuals.end(); ++it) {
cout << (it < bike_it ? "tsegway" : "tbike") << *it;
}
vector<Employee>::iterator old_bike_it =
stable_partition(individuals.begin(), individuals.end(),
return e.type <= Employee::EmployeeType::Senior;
});
cout << "Individuals partitioned:n";
for (auto it = individuals.begin(); it != individuals.end(); ++it) {
cout << (it < bike_it ? "tsegway" :
(it < old_bike_it ? "tnewbike" : "toldbike")) << *it;
}
partition the non-architects from higher to lower seniority

example output
Management partitioned:
jet Sarah B Smith: Executive
car Joe P Smith: Manager
car Chris M Williams: Manager
Architects partitioned (junior/senior unpartitioned):
segway Frank P Johnson: Architect
bike Jane Q Jones: Junior
bike Joe X Jones: Senior
bike Joe A Smith: Junior
Individuals fully partitioned:
segway Frank P Johnson: Architect
newbike Joe X Jones: Senior
oldbike Jane Q Jones: Junior
oldbike Joe A Smith: Junior

nth_element (not a super-hero, but still pretty super)
find median value in sequence
vector<int> v{ 12, 2, 89, 78, 18, 7, 72,
69, 81, 50, 49, 50, 51, 49 };
const size_t nth = v.size() / 2;
nth_element(v.begin(), v.begin() + nth, v.end());
output:
49 2 12 49 18 7 50 >50< 51 78 72 69 81 89
find percentile value in sequence
vector<int> v{ 12, 2, 89, 78, 18, 7, 72,
69, 81, 50, 49, 50, 51, 49 };
const size_t percentile = 75;
const size_t nth = v.size() * percentile / 100;
nth_element(v.begin(), v.begin() + nth, v.end());
output:
49 2 12 49 18 7 50 50 51 69 >72< 78 81 89

partial_sort vs. nth_element + sort
• Scenario: find top (say 20) web pages by
number of hits, among a much larger set
• We have a large sequence where we are only
concerned with the top 20 elements
• We want the top 20 elements of this sequence
in order; we don’t care about the rest
• Requirement: top 20 must be partitioned to
front of sequence, before the don’t-cares
• Requirement: top 20 must be sorted

partial_sort(vec.begin(),
vec.begin() + 20,
vec.end(),
greater<int64_t>());
nth_element(vec.begin(),
vec.begin() + 20,
vec.end(),
sort(vec.begin(), vec.begin() + 20,

• Are they equivalent?
vector<int64_t> v1 = init_vec();
vector<int64_t> v2 = v1; // make a copy
partial_sort(v1.begin(), v1.begin() + 20, v1.end(), greater<int64_t>());
nth_element(v2.begin(), v2.begin() + 20, v2.end(), greater<int64_t>());
sort(v2.begin(), v2.begin() + 20, greater<int64_t>());
cout << "sorted portions of vectors are equal: " << boolalpha
<< equal(v1.begin(), v1.begin() + 20, v2.begin()) << endl;
cout << "unsorted portions of vectors are equal: " << boolalpha
<< equal(v1.begin() + 20, v1.end(), v2.begin() + 20) << endl;
output:
sorted portions of vectors are equal: true
unsorted portions of vectors are equal: false

rotate
• Moves elements specified in a source range to
destination position, while moving displaced items to
source position
• Simple way to move one or more elements from one
position in a container to another without using erase and
insert
• Technically, this is a left rotate
• rotate moves element in the sequence in-place
• rotate_copy does the same, except that it copies the
elements to a separate sequence
• Requires:
– [first, middle) and [middle, last) are valid ranges
– i.e. first <= middle <= last

rotate
A naïve implementation – move one item in sequence
vector<int> v = { 0, 1, 2, 3, 4, 5, 6, 7, 8, 9 };
vector<int>::iterator it = v.begin() + 2;
int i = *it;
v.erase(it);
vector<int>::iterator it2 = v.begin() + 6 - 1;
v.insert(it2, i);
Output:
before: 0 1 2 3 4 5 6 7 8 9
After: 0 1 3 4 5 2 6 7 8 9

rotate
A naïve implementation – move one item in sequence
vector<int> v = { 0, 1, 2, 3, 4, 5, 6, 7, 8, 9 };
int i = *it;
v.erase(it);
vector<int>::iterator it2 = v.begin() + 6 - 1;
v.insert(it2, i);
Output:
before: 0 1 2 3 4 5 6 7 8 9
After: 0 1 3 4 5 2 6 7 8 9

rotate
A better implementation – move one item in sequence
vector<int> v = { 0, 1, 2, 3, 4, 5, 6, 7, 8, 9 };
vector<int>::iterator it2 = v.begin() + 6;
rotate(it, it + 1, it2);
Output:
before: 0 1 2 3 4 5 6 7 8 9
After: 0 1 3 4 5 2 6 7 8 9

rotate
Rotate a range of items
vector<int> v = { 0, 1, 2, 3, 4, 5, 6, 7, 8, 9 };
vector<int>::iterator it2 = v.begin() + 7;
rotate(it, it + 3, it2);
Output:
before: 0 1 2 3 4 5 6 7 8 9
after: 0 1 5 6 2 3 4 7 8 9

rotate
GCC’s definition
• Rotates the elements of the range
[first, last) by (middle - first)
positions so that the element at middle is
moved to first, the element at middle
+ 1 is moved to first + 1 and so on
for each element in the range.
• This effectively swaps the ranges [first,
middle) and [middle, last).

slide (move range of elements to new position)
Sean Parent, “C++ Seasoning”
• What if you wanted to implement an
algorithm that would move a range
to a point in either direction without
worrying about iterator ordering?
• For example, move a range of
selected GUI items to a new position
in a list.
• When using rotate, it is important
to get the order of iterators correct:
first <= middle <= last.
• Bonus: return a pair indicating
destination range of relocated
elements.

in a list.
elements.

template <typename I> // I models RandomAccessIterator
auto slide(I f, I l, I p) -> pair<I, I> {
if (p < f)
return { p, rotate(p, f, l) };
if (l < p)
return { rotate(f, l, p), p };
return { f, l };
}

gather (gather multiple elements around a point)
algorithm that would allow a scattered
selection of elements to be gathered to a
specific location?
• For example, move a range of scattered
multi-selected GUI items to a new
position in a list, gathered together.
• How can you gather diverse elements to
a single location without loops and
special-casing code?
• Bonus: return a pair indicating destination
range of relocated elements.

algorithm that would allow a scattered
selection of elements to be gathered to a
specific location?
• For example, move a range of scattered
multi-selected GUI items to a new
position in a list, gathered together.
• How can you gather diverse elements to
a single location without loops and
special-casing code?
• Bonus: return a pair indicating destination
range of relocated elements.

• Break the problem space into
two pieces:
– The sequence before the
destination position
– The sequence after the
• What algorithm can gather the
select items while maintaining
their relative position?

• Break the problem space
into two pieces:
– The sequence before the
– The sequence after the
• That looks a lot like
partitioning…

template <typename I, // models BidirectionalIterator
typename S> // models UnaryPredicate
auto gather(I f, I l, I p, S s) -> pair<I, I>
{
return { stable_partition(f, p, not1(s)),
stable_partition(p, l, s) };
}

set_difference
process configuration updates
• Scenario: you store some configuration
information in an ordered container
• You receive updated configuration on a
regular basis
• You want to quickly determine what has
changed: i.e. what new items have been
added, and what items have been removed

set_difference
process configuration updates
vector<string> new_items, removed_items;
set_difference(current.begin(), current.end(),
update.begin(), update.end(),
back_inserter(removed_items));
remove_config(removed_items);
set_difference(update.begin(), update.end(),
current.begin(), current.end(),
back_inserter(new_items));
add_config(new_items);

set_difference
process configuration updates test example
set<string> current { "one", "two", "three", "four", "five" };
set<string> update { "one", "three", "four", "six", "seven" };
vector<string> new_items, removed_items;
set_difference(current.begin(), current.end(),
update.begin(), update.end(),
back_inserter(removed_items));
remove_config(removed_items);
set_difference(update.begin(), update.end(),
current.begin(), current.end(),
back_inserter(new_items));
add_config(new_items);
output:
Removed: five two
Added: seven six

Coming soon to a compiler near you!
Parallel algorithms
• Currently experimental, proposed for a future
release.
• In the std::experimental::parallel
namespace.
• Have the same interfaces as standard
algorithms.
• Controlled by use of parallelism
execution_policy.
• See the Parallelism Technical Specification for
more info.

A quick review…
• Be familiar with all the varied algorithms
provided for free with your compiler
• Write some code that exercises each of
them so you are familiar with their usage
and individual personalities
• Write your own adaptations of existing
algorithms
• Implement your own algorithms

STL algorithms
• all_of
• any_of
• none_of
• for_each
• find
– find_if
– find_if_not
• find_end
• find_first_of
• adjacent_find
• count
– count_if
• mismatch
• equal
• is_permutation
• search
– search_n

STL algorithms
• copy
– copy_n
– copy_if
– copy_backward
• move
– move_backward
• swap_ranges
– iter_swap
• transform
• replace
– replace_if
– replace_copy
– replace_copy_if
• fill
– fill_n
• generate
– generate_n
• remove
– remove_if
– remove_copy
– remove_copy_if
• unique
– unique_copy
• reverse
– reverse_copy
• rotate
– rotate_copy
• shuffle
– random_shuffle
• partition
– is_partitioned
– stable_partition
– partition_copy
– partition_point

STL algorithms
• sorting
– sort
– stable_sort
– partial_sort
– partial_sort_copy
• nth_element
• binary search
– lower_bound
– upper_bound
– equal_range
– binary_search
• merge
– merge
– inplace_merge
• set operations on sorted
structures
– includes
– set_union
– set_intersection
– set_difference
– set_symmetric_difference
• heap operations
– push_heap
– pop_heap
– make_heap
– sort_heap
• minimum and maximum
– min
– max
– minmax
– min_element
– max_element
– minmax_element
• lexicographical comparisons
– lexicographical_compare
• permutation generators
– next_permutation
– prev_permutation

STL algorithms
• accumulate
• inner_product
• partial_sum
• adjacent_difference
• iota

Shout out to my “sponsor”
• F5 Networks is a highly technical company
with a belief in well engineered software
• F5 Networks has graciously sent me here,
and they tolerate encourage me working on
this stuff in addition to my “real” work
• If you’re looking for something cool and
challenging to do with C++, check out F5!
• https://f5.com/about-us/careers

Shameless advertising…
• I’m working on a book on this very topic!
• It isn’t done yet…
• If you have some creative or clever uses of
STL algorithms you’d like me consider, please
drop me a line with some example code!
• michaelv@codeache.net
• Visit my blog for more adventures:
http://codeache.net

STL Algorithms In Action

Recommended

More Related Content

What's hot (20)

Viewers also liked (20)

Similar to STL Algorithms In Action (20)

Recently uploaded (20)

STL Algorithms In Action