solidity/tools/yulPhaser/Phaser.cpp

/*
	This file is part of solidity.

	solidity is free software: you can redistribute it and/or modify
	it under the terms of the GNU General Public License as published by
	the Free Software Foundation, either version 3 of the License, or
	(at your option) any later version.

	solidity is distributed in the hope that it will be useful,
	but WITHOUT ANY WARRANTY; without even the implied warranty of
	MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
	GNU General Public License for more details.

	You should have received a copy of the GNU General Public License
	along with solidity.  If not, see <http://www.gnu.org/licenses/>.
*/

#include <tools/yulPhaser/Phaser.h>

#include <tools/yulPhaser/AlgorithmRunner.h>
#include <tools/yulPhaser/Common.h>
#include <tools/yulPhaser/Exceptions.h>
#include <tools/yulPhaser/FitnessMetrics.h>
#include <tools/yulPhaser/GeneticAlgorithms.h>
#include <tools/yulPhaser/Program.h>
#include <tools/yulPhaser/SimulationRNG.h>

#include <liblangutil/CharStream.h>

#include <libsolutil/Assertions.h>
#include <libsolutil/CommonData.h>
#include <libsolutil/CommonIO.h>

#include <boost/filesystem.hpp>

#include <iostream>

using namespace std;
using namespace solidity;
using namespace solidity::langutil;
using namespace solidity::util;
using namespace solidity::yul;
using namespace solidity::phaser;

namespace po = boost::program_options;

namespace
{

map<PhaserMode, string> const PhaserModeToStringMap =
{
	{PhaserMode::RunAlgorithm, "run-algorithm"},
	{PhaserMode::PrintOptimisedPrograms, "print-optimised-programs"},
	{PhaserMode::PrintOptimisedASTs, "print-optimised-asts"},
};
map<string, PhaserMode> const StringToPhaserModeMap = invertMap(PhaserModeToStringMap);

map<Algorithm, string> const AlgorithmToStringMap =
{
	{Algorithm::Random, "random"},
	{Algorithm::GEWEP, "GEWEP"},
	{Algorithm::Classic, "classic"},
};
map<string, Algorithm> const StringToAlgorithmMap = invertMap(AlgorithmToStringMap);

map<MetricChoice, string> MetricChoiceToStringMap =
{
	{MetricChoice::CodeSize, "code-size"},
	{MetricChoice::RelativeCodeSize, "relative-code-size"},
};
map<string, MetricChoice> const StringToMetricChoiceMap = invertMap(MetricChoiceToStringMap);

map<MetricAggregatorChoice, string> const MetricAggregatorChoiceToStringMap =
{
	{MetricAggregatorChoice::Average, "average"},
	{MetricAggregatorChoice::Sum, "sum"},
	{MetricAggregatorChoice::Maximum, "maximum"},
	{MetricAggregatorChoice::Minimum, "minimum"},
};
map<string, MetricAggregatorChoice> const StringToMetricAggregatorChoiceMap = invertMap(MetricAggregatorChoiceToStringMap);

map<CrossoverChoice, string> const CrossoverChoiceToStringMap =
{
	{CrossoverChoice::SinglePoint, "single-point"},
	{CrossoverChoice::TwoPoint, "two-point"},
	{CrossoverChoice::Uniform, "uniform"},
};
map<string, CrossoverChoice> const StringToCrossoverChoiceMap = invertMap(CrossoverChoiceToStringMap);

}

istream& phaser::operator>>(istream& _inputStream, PhaserMode& _phaserMode) { return deserializeChoice(_inputStream, _phaserMode, StringToPhaserModeMap); }
ostream& phaser::operator<<(ostream& _outputStream, PhaserMode _phaserMode) { return serializeChoice(_outputStream, _phaserMode, PhaserModeToStringMap); }
istream& phaser::operator>>(istream& _inputStream, Algorithm& _algorithm) { return deserializeChoice(_inputStream, _algorithm, StringToAlgorithmMap); }
ostream& phaser::operator<<(ostream& _outputStream, Algorithm _algorithm) { return serializeChoice(_outputStream, _algorithm, AlgorithmToStringMap); }
istream& phaser::operator>>(istream& _inputStream, MetricChoice& _metric) { return deserializeChoice(_inputStream, _metric, StringToMetricChoiceMap); }
ostream& phaser::operator<<(ostream& _outputStream, MetricChoice _metric) { return serializeChoice(_outputStream, _metric, MetricChoiceToStringMap); }
istream& phaser::operator>>(istream& _inputStream, MetricAggregatorChoice& _aggregator) { return deserializeChoice(_inputStream, _aggregator, StringToMetricAggregatorChoiceMap); }
ostream& phaser::operator<<(ostream& _outputStream, MetricAggregatorChoice _aggregator) { return serializeChoice(_outputStream, _aggregator, MetricAggregatorChoiceToStringMap); }
istream& phaser::operator>>(istream& _inputStream, CrossoverChoice& _crossover) { return deserializeChoice(_inputStream, _crossover, StringToCrossoverChoiceMap); }
ostream& phaser::operator<<(ostream& _outputStream, CrossoverChoice _crossover) { return serializeChoice(_outputStream, _crossover, CrossoverChoiceToStringMap); }

GeneticAlgorithmFactory::Options GeneticAlgorithmFactory::Options::fromCommandLine(po::variables_map const& _arguments)
{
	return {
		_arguments["algorithm"].as<Algorithm>(),
		_arguments["min-chromosome-length"].as<size_t>(),
		_arguments["max-chromosome-length"].as<size_t>(),
		_arguments["crossover"].as<CrossoverChoice>(),
		_arguments["uniform-crossover-swap-chance"].as<double>(),
		_arguments.count("random-elite-pool-size") > 0 ?
			_arguments["random-elite-pool-size"].as<double>() :
			optional<double>{},
		_arguments["gewep-mutation-pool-size"].as<double>(),
		_arguments["gewep-crossover-pool-size"].as<double>(),
		_arguments["gewep-randomisation-chance"].as<double>(),
		_arguments["gewep-deletion-vs-addition-chance"].as<double>(),
		_arguments.count("gewep-genes-to-randomise") > 0 ?
			_arguments["gewep-genes-to-randomise"].as<double>() :
			optional<double>{},
		_arguments.count("gewep-genes-to-add-or-delete") > 0 ?
			_arguments["gewep-genes-to-add-or-delete"].as<double>() :
			optional<double>{},
		_arguments["classic-elite-pool-size"].as<double>(),
		_arguments["classic-crossover-chance"].as<double>(),
		_arguments["classic-mutation-chance"].as<double>(),
		_arguments["classic-deletion-chance"].as<double>(),
		_arguments["classic-addition-chance"].as<double>(),
	};
}

unique_ptr<GeneticAlgorithm> GeneticAlgorithmFactory::build(
	Options const& _options,
	size_t _populationSize
)
{
	assert(_populationSize > 0);

	switch (_options.algorithm)
	{
		case Algorithm::Random:
		{
			double elitePoolSize = 1.0 / _populationSize;

			if (_options.randomElitePoolSize.has_value())
				elitePoolSize = _options.randomElitePoolSize.value();

			return make_unique<RandomAlgorithm>(RandomAlgorithm::Options{
				/* elitePoolSize = */ elitePoolSize,
				/* minChromosomeLength = */ _options.minChromosomeLength,
				/* maxChromosomeLength = */ _options.maxChromosomeLength,
			});
		}
		case Algorithm::GEWEP:
		{
			double percentGenesToRandomise = 1.0 / _options.maxChromosomeLength;
			double percentGenesToAddOrDelete = percentGenesToRandomise;

			if (_options.gewepGenesToRandomise.has_value())
				percentGenesToRandomise = _options.gewepGenesToRandomise.value();
			if (_options.gewepGenesToAddOrDelete.has_value())
				percentGenesToAddOrDelete = _options.gewepGenesToAddOrDelete.value();

			return make_unique<GenerationalElitistWithExclusivePools>(GenerationalElitistWithExclusivePools::Options{
				/* mutationPoolSize = */ _options.gewepMutationPoolSize,
				/* crossoverPoolSize = */ _options.gewepCrossoverPoolSize,
				/* randomisationChance = */ _options.gewepRandomisationChance,
				/* deletionVsAdditionChance = */ _options.gewepDeletionVsAdditionChance,
				/* percentGenesToRandomise = */ percentGenesToRandomise,
				/* percentGenesToAddOrDelete = */ percentGenesToAddOrDelete,
				/* crossover = */ _options.crossover,
				/* uniformCrossoverSwapChance = */ _options.uniformCrossoverSwapChance,
			});
		}
		case Algorithm::Classic:
		{
			return make_unique<ClassicGeneticAlgorithm>(ClassicGeneticAlgorithm::Options{
				/* elitePoolSize = */ _options.classicElitePoolSize,
				/* crossoverChance = */ _options.classicCrossoverChance,
				/* mutationChance = */ _options.classicMutationChance,
				/* deletionChance = */ _options.classicDeletionChance,
				/* additionChance = */ _options.classicAdditionChance,
				/* crossover = */ _options.crossover,
				/* uniformCrossoverSwapChance = */ _options.uniformCrossoverSwapChance,
			});
		}
		default:
			assertThrow(false, solidity::util::Exception, "Invalid Algorithm value.");
	}
}

CodeWeights CodeWeightFactory::buildFromCommandLine(po::variables_map const& _arguments)
{
	return {
		_arguments["expression-statement-cost"].as<size_t>(),
		_arguments["assignment-cost"].as<size_t>(),
		_arguments["variable-declaration-cost"].as<size_t>(),
		_arguments["function-definition-cost"].as<size_t>(),
		_arguments["if-cost"].as<size_t>(),
		_arguments["switch-cost"].as<size_t>(),
		_arguments["case-cost"].as<size_t>(),
		_arguments["for-loop-cost"].as<size_t>(),
		_arguments["break-cost"].as<size_t>(),
		_arguments["continue-cost"].as<size_t>(),
		_arguments["leave-cost"].as<size_t>(),
		_arguments["block-cost"].as<size_t>(),
		_arguments["function-call-cost"].as<size_t>(),
		_arguments["identifier-cost"].as<size_t>(),
		_arguments["literal-cost"].as<size_t>(),
	};
}

FitnessMetricFactory::Options FitnessMetricFactory::Options::fromCommandLine(po::variables_map const& _arguments)
{
	return {
		_arguments["metric"].as<MetricChoice>(),
		_arguments["metric-aggregator"].as<MetricAggregatorChoice>(),
		_arguments["relative-metric-scale"].as<size_t>(),
		_arguments["chromosome-repetitions"].as<size_t>(),
	};
}

unique_ptr<FitnessMetric> FitnessMetricFactory::build(
	Options const& _options,
	vector<Program> _programs,
	vector<shared_ptr<ProgramCache>> _programCaches,
	CodeWeights const& _weights
)
{
	assert(_programCaches.size() == _programs.size());
	assert(_programs.size() > 0 && "Validations should prevent this from being executed with zero files.");

	vector<shared_ptr<FitnessMetric>> metrics;
	switch (_options.metric)
	{
		case MetricChoice::CodeSize:
		{
			for (size_t i = 0; i < _programs.size(); ++i)
				metrics.push_back(make_unique<ProgramSize>(
					_programCaches[i] != nullptr ? optional<Program>{} : move(_programs[i]),
					move(_programCaches[i]),
					_weights,
					_options.chromosomeRepetitions
				));

			break;
		}
		case MetricChoice::RelativeCodeSize:
		{
			for (size_t i = 0; i < _programs.size(); ++i)
				metrics.push_back(make_unique<RelativeProgramSize>(
					_programCaches[i] != nullptr ? optional<Program>{} : move(_programs[i]),
					move(_programCaches[i]),
					_options.relativeMetricScale,
					_weights,
					_options.chromosomeRepetitions
				));
			break;
		}
		default:
			assertThrow(false, solidity::util::Exception, "Invalid MetricChoice value.");
	}

	switch (_options.metricAggregator)
	{
		case MetricAggregatorChoice::Average:
			return make_unique<FitnessMetricAverage>(move(metrics));
		case MetricAggregatorChoice::Sum:
			return make_unique<FitnessMetricSum>(move(metrics));
		case MetricAggregatorChoice::Maximum:
			return make_unique<FitnessMetricMaximum>(move(metrics));
		case MetricAggregatorChoice::Minimum:
			return make_unique<FitnessMetricMinimum>(move(metrics));
		default:
			assertThrow(false, solidity::util::Exception, "Invalid MetricAggregatorChoice value.");
	}
}

PopulationFactory::Options PopulationFactory::Options::fromCommandLine(po::variables_map const& _arguments)
{
	return {
		_arguments["min-chromosome-length"].as<size_t>(),
		_arguments["max-chromosome-length"].as<size_t>(),
		_arguments.count("population") > 0 ?
			_arguments["population"].as<vector<string>>() :
			vector<string>{},
		_arguments.count("random-population") > 0 ?
			_arguments["random-population"].as<vector<size_t>>() :
			vector<size_t>{},
		_arguments.count("population-from-file") > 0 ?
			_arguments["population-from-file"].as<vector<string>>() :
			vector<string>{},
	};
}

Population PopulationFactory::build(
	Options const& _options,
	shared_ptr<FitnessMetric> _fitnessMetric
)
{
	Population population = buildFromStrings(_options.population, _fitnessMetric);

	size_t combinedSize = 0;
	for (size_t populationSize: _options.randomPopulation)
		combinedSize += populationSize;

	population = move(population) + buildRandom(
		combinedSize,
		_options.minChromosomeLength,
		_options.maxChromosomeLength,
		_fitnessMetric
	);

	for (string const& populationFilePath: _options.populationFromFile)
		population = move(population) + buildFromFile(populationFilePath, _fitnessMetric);

	return population;
}

Population PopulationFactory::buildFromStrings(
	vector<string> const& _geneSequences,
	shared_ptr<FitnessMetric> _fitnessMetric
)
{
	vector<Chromosome> chromosomes;
	for (string const& geneSequence: _geneSequences)
		chromosomes.emplace_back(geneSequence);

	return Population(move(_fitnessMetric), move(chromosomes));
}

Population PopulationFactory::buildRandom(
	size_t _populationSize,
	size_t _minChromosomeLength,
	size_t _maxChromosomeLength,
	shared_ptr<FitnessMetric> _fitnessMetric
)
{
	return Population::makeRandom(
		move(_fitnessMetric),
		_populationSize,
		_minChromosomeLength,
		_maxChromosomeLength
	);
}

Population PopulationFactory::buildFromFile(
	string const& _filePath,
	shared_ptr<FitnessMetric> _fitnessMetric
)
{
	return buildFromStrings(readLinesFromFile(_filePath), move(_fitnessMetric));
}

ProgramCacheFactory::Options ProgramCacheFactory::Options::fromCommandLine(po::variables_map const& _arguments)
{
	return {
		_arguments["program-cache"].as<bool>(),
	};
}

vector<shared_ptr<ProgramCache>> ProgramCacheFactory::build(
	Options const& _options,
	vector<Program> _programs
)
{
	vector<shared_ptr<ProgramCache>> programCaches;
	for (Program& program: _programs)
		programCaches.push_back(_options.programCacheEnabled ? make_shared<ProgramCache>(move(program)) : nullptr);

	return programCaches;
}

ProgramFactory::Options ProgramFactory::Options::fromCommandLine(po::variables_map const& _arguments)
{
	return {
		_arguments["input-files"].as<vector<string>>(),
		_arguments["prefix"].as<string>(),
	};
}

vector<Program> ProgramFactory::build(Options const& _options)
{
	vector<Program> inputPrograms;
	for (auto& path: _options.inputFiles)
	{
		CharStream sourceCode = loadSource(path);
		variant<Program, ErrorList> programOrErrors = Program::load(sourceCode);
		if (holds_alternative<ErrorList>(programOrErrors))
		{
			cerr << get<ErrorList>(programOrErrors) << endl;
			assertThrow(false, InvalidProgram, "Failed to load program " + path);
		}

		get<Program>(programOrErrors).optimise(Chromosome(_options.prefix).optimisationSteps());
		inputPrograms.push_back(move(get<Program>(programOrErrors)));
	}

	return inputPrograms;
}

CharStream ProgramFactory::loadSource(string const& _sourcePath)
{
	assertThrow(boost::filesystem::exists(_sourcePath), MissingFile, "Source file does not exist: " + _sourcePath);

	string sourceCode = readFileAsString(_sourcePath);
	return CharStream(sourceCode, _sourcePath);
}

void Phaser::main(int _argc, char** _argv)
{
	optional<po::variables_map> arguments = parseCommandLine(_argc, _argv);
	if (!arguments.has_value())
		return;

	initialiseRNG(arguments.value());

	runPhaser(arguments.value());
}

Phaser::CommandLineDescription Phaser::buildCommandLineDescription()
{
	size_t const lineLength = po::options_description::m_default_line_length;
	size_t const minDescriptionLength = lineLength - 23;

	po::options_description keywordDescription(
		"yul-phaser, a tool for finding the best sequence of Yul optimisation phases.\n"
		"\n"
		"Usage: yul-phaser [options] <file>\n"
		"Reads <file> as Yul code and tries to find the best order in which to run optimisation"
		" phases using a genetic algorithm.\n"
		"Example:\n"
		"yul-phaser program.yul\n"
		"\n"
		"Allowed options",
		lineLength,
		minDescriptionLength
	);

	po::options_description generalDescription("GENERAL", lineLength, minDescriptionLength);
	generalDescription.add_options()
		("help", "Show help message and exit.")
		("input-files", po::value<vector<string>>()->required()->value_name("<PATH>"), "Input files.")
		(
			"prefix",
			po::value<string>()->value_name("<CHROMOSOME>")->default_value(""),
			"Initial optimisation steps automatically applied to every input program.\n"
			"The result is treated as if it was the actual input, i.e. the steps are not considered "
			"a part of the chromosomes and cannot be mutated. The values of relative metric values "
			"are also relative to the fitness of a program with these steps applied rather than the "
			"fitness of the original program.\n"
			"Note that phaser always adds a 'hgo' prefix to ensure that chromosomes can "
			"contain arbitrary optimisation steps. This implicit prefix cannot be changed or "
			"or removed using this option. The value given here is applied after it."
		)
		("seed", po::value<uint32_t>()->value_name("<NUM>"), "Seed for the random number generator.")
		(
			"rounds",
			po::value<size_t>()->value_name("<NUM>"),
			"The number of rounds after which the algorithm should stop. (default=no limit)."
		)
		(
			"mode",
			po::value<PhaserMode>()->value_name("<NAME>")->default_value(PhaserMode::RunAlgorithm),
			(
				"Mode of operation. The default is to run the algorithm but you can also tell phaser "
				"to do something else with its parameters, e.g. just print the optimised programs and exit.\n"
				"\n"
				"AVAILABLE MODES:\n"
				"* " + toString(PhaserMode::RunAlgorithm) + "\n" +
				"* " + toString(PhaserMode::PrintOptimisedPrograms) + "\n" +
				"* " + toString(PhaserMode::PrintOptimisedASTs)
			).c_str()
		)
	;
	keywordDescription.add(generalDescription);

	po::options_description algorithmDescription("ALGORITHM", lineLength, minDescriptionLength);
	algorithmDescription.add_options()
		(
			"algorithm",
			po::value<Algorithm>()->value_name("<NAME>")->default_value(Algorithm::GEWEP),
			(
				"Algorithm\n"
				"\n"
				"AVAILABLE ALGORITHMS:\n"
				"* " + toString(Algorithm::GEWEP) + "\n" +
				"* " + toString(Algorithm::Classic) + "\n" +
				"* " + toString(Algorithm::Random)
			).c_str()
		)
		(
			"no-randomise-duplicates",
			po::bool_switch(),
			"By default, after each round of the algorithm duplicate chromosomes are removed from"
			"the population and replaced with randomly generated ones. "
			"This option disables this postprocessing."
		)
		(
			"min-chromosome-length",
			po::value<size_t>()->value_name("<NUM>")->default_value(100),
			"Minimum length of randomly generated chromosomes."
		)
		(
			"max-chromosome-length",
			po::value<size_t>()->value_name("<NUM>")->default_value(100),
			"Maximum length of randomly generated chromosomes."
		)
		(
			"crossover",
			po::value<CrossoverChoice>()->value_name("<NAME>")->default_value(CrossoverChoice::Uniform),
			(
				"Type of the crossover operator to use.\n"
				"\n"
				"AVAILABLE CROSSOVER OPERATORS:\n"
				"* " + toString(CrossoverChoice::SinglePoint) + "\n" +
				"* " + toString(CrossoverChoice::TwoPoint) + "\n" +
				"* " + toString(CrossoverChoice::Uniform)
			).c_str()
		)
		(
			"uniform-crossover-swap-chance",
			po::value<double>()->value_name("<PROBABILITY>")->default_value(0.5),
			"Chance of two genes being swapped between chromosomes in uniform crossover."
		)
	;
	keywordDescription.add(algorithmDescription);

	po::options_description gewepAlgorithmDescription("GEWEP ALGORITHM", lineLength, minDescriptionLength);
	gewepAlgorithmDescription.add_options()
		(
			"gewep-mutation-pool-size",
			po::value<double>()->value_name("<FRACTION>")->default_value(0.25),
			"Percentage of population to regenerate using mutations in each round."
		)
		(
			"gewep-crossover-pool-size",
			po::value<double>()->value_name("<FRACTION>")->default_value(0.25),
			"Percentage of population to regenerate using crossover in each round."
		)
		(
			"gewep-randomisation-chance",
			po::value<double>()->value_name("<PROBABILITY>")->default_value(0.9),
			"The chance of choosing gene randomisation as the mutation to perform."
		)
		(
			"gewep-deletion-vs-addition-chance",
			po::value<double>()->value_name("<PROBABILITY>")->default_value(0.5),
			"The chance of choosing gene deletion as the mutation if randomisation was not chosen."
		)
		(
			"gewep-genes-to-randomise",
			po::value<double>()->value_name("<PROBABILITY>"),
			"The chance of any given gene being mutated in gene randomisation. "
			"(default=1/max-chromosome-length)"
		)
		(
			"gewep-genes-to-add-or-delete",
			po::value<double>()->value_name("<PROBABILITY>"),
			"The chance of a gene being added (or deleted) in gene addition (or deletion). "
			"(default=1/max-chromosome-length)"
		)
	;
	keywordDescription.add(gewepAlgorithmDescription);

	po::options_description classicGeneticAlgorithmDescription("CLASSIC GENETIC ALGORITHM", lineLength, minDescriptionLength);
	classicGeneticAlgorithmDescription.add_options()
		(
			"classic-elite-pool-size",
			po::value<double>()->value_name("<FRACTION>")->default_value(0.25),
			"Percentage of population to regenerate using mutations in each round."
		)
		(
			"classic-crossover-chance",
			po::value<double>()->value_name("<FRACTION>")->default_value(0.75),
			"Chance of a chromosome being selected for crossover."
		)
		(
			"classic-mutation-chance",
			po::value<double>()->value_name("<FRACTION>")->default_value(0.01),
			"Chance of a gene being mutated."
		)
		(
			"classic-deletion-chance",
			po::value<double>()->value_name("<PROBABILITY>")->default_value(0.01),
			"Chance of a gene being deleted."
		)
		(
			"classic-addition-chance",
			po::value<double>()->value_name("<PROBABILITY>")->default_value(0.01),
			"Chance of a random gene being added."
		)
	;
	keywordDescription.add(classicGeneticAlgorithmDescription);

	po::options_description randomAlgorithmDescription("RANDOM ALGORITHM", lineLength, minDescriptionLength);
	randomAlgorithmDescription.add_options()
		(
			"random-elite-pool-size",
			po::value<double>()->value_name("<FRACTION>"),
			"Percentage of the population preserved in each round. "
			"(default=one individual, regardless of population size)"
		)
	;
	keywordDescription.add(randomAlgorithmDescription);

	po::options_description populationDescription("POPULATION", lineLength, minDescriptionLength);
	populationDescription.add_options()
		(
			"population",
			po::value<vector<string>>()->multitoken()->value_name("<CHROMOSOMES>"),
			"List of chromosomes to be included in the initial population. "
			"You can specify multiple values separated with spaces or invoke the option multiple times "
			"and all the values will be included."
		)
		(
			"random-population",
			po::value<vector<size_t>>()->value_name("<SIZE>"),
			"The number of randomly generated chromosomes to be included in the initial population."
		)
		(
			"population-from-file",
			po::value<vector<string>>()->value_name("<FILE>"),
			"A text file with a list of chromosomes (one per line) to be included in the initial population."
		)
		(
			"population-autosave",
			po::value<string>()->value_name("<FILE>"),
			"If specified, the population is saved in the specified file after each round. (default=autosave disabled)"
		)
	;
	keywordDescription.add(populationDescription);

	po::options_description metricsDescription("METRICS", lineLength, minDescriptionLength);
	metricsDescription.add_options()
		(
			"metric",
			po::value<MetricChoice>()->value_name("<NAME>")->default_value(MetricChoice::RelativeCodeSize),
			(
				"Metric used to evaluate the fitness of a chromosome.\n"
				"\n"
				"AVAILABLE METRICS:\n"
				"* " + toString(MetricChoice::CodeSize) + "\n" +
				"* " + toString(MetricChoice::RelativeCodeSize)
			).c_str()
		)
		(
			"metric-aggregator",
			po::value<MetricAggregatorChoice>()->value_name("<NAME>")->default_value(MetricAggregatorChoice::Average),
			(
				"Operator used to combine multiple fitness metric values obtained by evaluating a "
				"chromosome separately for each input program.\n"
				"\n"
				"AVAILABLE METRIC AGGREGATORS:\n"
				"* " + toString(MetricAggregatorChoice::Average) + "\n" +
				"* " + toString(MetricAggregatorChoice::Sum) + "\n" +
				"* " + toString(MetricAggregatorChoice::Maximum) + "\n" +
				"* " + toString(MetricAggregatorChoice::Minimum)
			).c_str()
		)
		(
			"relative-metric-scale",
			po::value<size_t>()->value_name("<EXPONENT>")->default_value(3),
			"Scaling factor for values produced by relative fitness metrics. \n"
			"Since all metrics must produce integer values, the fractional part of the result is discarded. "
			"To keep the numbers meaningful, a relative metric multiples its values by a scaling factor "
			"and this option specifies the exponent of this factor. "
			"For example with value of 3 the factor is 10^3 = 1000 and the metric will return "
			"500 to represent 0.5, 1000 for 1.0, 2000 for 2.0 and so on. "
			"Using a bigger factor allows discerning smaller relative differences between chromosomes "
			"but makes the numbers less readable and may also lose precision if the numbers are very large."
		)
		(
			"chromosome-repetitions",
			po::value<size_t>()->value_name("<COUNT>")->default_value(1),
			"Number of times to repeat the sequence optimisation steps represented by a chromosome."
		)
	;
	keywordDescription.add(metricsDescription);

	po::options_description metricWeightDescription("METRIC WEIGHTS", lineLength, minDescriptionLength);
	metricWeightDescription.add_options()
		// TODO: We need to figure out the best set of weights for the phaser.
		// This one is just a stopgap to make sure no statement or expression has zero cost.
		("expression-statement-cost", po::value<size_t>()->value_name("<COST>")->default_value(1))
		("assignment-cost",           po::value<size_t>()->value_name("<COST>")->default_value(1))
		("variable-declaration-cost", po::value<size_t>()->value_name("<COST>")->default_value(1))
		("function-definition-cost",  po::value<size_t>()->value_name("<COST>")->default_value(1))
		("if-cost",                   po::value<size_t>()->value_name("<COST>")->default_value(2))
		("switch-cost",               po::value<size_t>()->value_name("<COST>")->default_value(1))
		("case-cost",                 po::value<size_t>()->value_name("<COST>")->default_value(2))
		("for-loop-cost",             po::value<size_t>()->value_name("<COST>")->default_value(3))
		("break-cost",                po::value<size_t>()->value_name("<COST>")->default_value(2))
		("continue-cost",             po::value<size_t>()->value_name("<COST>")->default_value(2))
		("leave-cost",                po::value<size_t>()->value_name("<COST>")->default_value(2))
		("block-cost",                po::value<size_t>()->value_name("<COST>")->default_value(1))
		("function-call-cost",        po::value<size_t>()->value_name("<COST>")->default_value(1))
		("identifier-cost",           po::value<size_t>()->value_name("<COST>")->default_value(1))
		("literal-cost",              po::value<size_t>()->value_name("<COST>")->default_value(1))
	;
	keywordDescription.add(metricWeightDescription);

	po::options_description cacheDescription("CACHE", lineLength, minDescriptionLength);
	cacheDescription.add_options()
		(
			"program-cache",
			po::bool_switch(),
			"Enables caching of intermediate programs corresponding to chromosome prefixes.\n"
			"This speeds up fitness evaluation by a lot but eats tons of memory if the chromosomes are long. "
			"Disabled by default since there's currently no way to set an upper limit on memory usage but "
			"highly recommended if your computer has enough RAM."
		)
	;
	keywordDescription.add(cacheDescription);

	po::options_description outputDescription("OUTPUT", lineLength, minDescriptionLength);
	outputDescription.add_options()
		(
			"show-initial-population",
			po::bool_switch(),
			"Print the state of the population before the algorithm starts."
		)
		(
			"show-only-top-chromosome",
			po::bool_switch(),
			"Print only the best chromosome found in each round rather than the whole population."
		)
		(
			"hide-round",
			po::bool_switch(),
			"Hide information about the current round (round number and elapsed time)."
		)
		(
			"show-cache-stats",
			po::bool_switch(),
			"Print information about cache size and effectiveness after each round."
		)
		(
			"show-seed",
			po::bool_switch(),
			"Print the selected random seed."
		)
	;
	keywordDescription.add(outputDescription);

	po::positional_options_description positionalDescription;
	positionalDescription.add("input-files", -1);

	return {keywordDescription, positionalDescription};
}

optional<po::variables_map> Phaser::parseCommandLine(int _argc, char** _argv)
{
	auto [keywordDescription, positionalDescription] = buildCommandLineDescription();

	po::variables_map arguments;
	po::notify(arguments);

	po::command_line_parser parser(_argc, _argv);
	parser.options(keywordDescription).positional(positionalDescription);
	po::store(parser.run(), arguments);

	if (arguments.count("help") > 0)
	{
		cout << keywordDescription << endl;
		return nullopt;
	}

	if (arguments.count("input-files") == 0)
		assertThrow(false, NoInputFiles, "Missing argument: input-files.");

	return arguments;
}

void Phaser::initialiseRNG(po::variables_map const& _arguments)
{
	uint32_t seed;
	if (_arguments.count("seed") > 0)
		seed = _arguments["seed"].as<uint32_t>();
	else
		seed = SimulationRNG::generateSeed();

	SimulationRNG::reset(seed);
	if (_arguments["show-seed"].as<bool>())
		cout << "Random seed: " << seed << endl;
}

AlgorithmRunner::Options Phaser::buildAlgorithmRunnerOptions(po::variables_map const& _arguments)
{
	return {
		_arguments.count("rounds") > 0 ? static_cast<optional<size_t>>(_arguments["rounds"].as<size_t>()) : nullopt,
		_arguments.count("population-autosave") > 0 ? static_cast<optional<string>>(_arguments["population-autosave"].as<string>()) : nullopt,
		!_arguments["no-randomise-duplicates"].as<bool>(),
		_arguments["min-chromosome-length"].as<size_t>(),
		_arguments["max-chromosome-length"].as<size_t>(),
		_arguments["show-initial-population"].as<bool>(),
		_arguments["show-only-top-chromosome"].as<bool>(),
		!_arguments["hide-round"].as<bool>(),
		_arguments["show-cache-stats"].as<bool>(),
	};
}

void Phaser::runPhaser(po::variables_map const& _arguments)
{
	auto programOptions = ProgramFactory::Options::fromCommandLine(_arguments);
	auto cacheOptions = ProgramCacheFactory::Options::fromCommandLine(_arguments);
	auto metricOptions = FitnessMetricFactory::Options::fromCommandLine(_arguments);
	auto populationOptions = PopulationFactory::Options::fromCommandLine(_arguments);

	vector<Program> programs = ProgramFactory::build(programOptions);
	vector<shared_ptr<ProgramCache>> programCaches = ProgramCacheFactory::build(cacheOptions, programs);
	CodeWeights codeWeights = CodeWeightFactory::buildFromCommandLine(_arguments);
	unique_ptr<FitnessMetric> fitnessMetric = FitnessMetricFactory::build(
		metricOptions,
		programs,
		programCaches,
		codeWeights
	);
	Population population = PopulationFactory::build(populationOptions, move(fitnessMetric));

	if (_arguments["mode"].as<PhaserMode>() == PhaserMode::RunAlgorithm)
		runAlgorithm(_arguments, move(population), move(programCaches));
	else
		printOptimisedProgramsOrASTs(_arguments, population, move(programs), _arguments["mode"].as<PhaserMode>());
}

void Phaser::runAlgorithm(
	po::variables_map const& _arguments,
	Population _population,
	vector<shared_ptr<ProgramCache>> _programCaches
)
{
	auto algorithmOptions = GeneticAlgorithmFactory::Options::fromCommandLine(_arguments);

	unique_ptr<GeneticAlgorithm> geneticAlgorithm = GeneticAlgorithmFactory::build(
		algorithmOptions,
		_population.individuals().size()
	);

	AlgorithmRunner algorithmRunner(move(_population), move(_programCaches), buildAlgorithmRunnerOptions(_arguments), cout);
	algorithmRunner.run(*geneticAlgorithm);
}

void Phaser::printOptimisedProgramsOrASTs(
	po::variables_map const& _arguments,
	Population const& _population,
	vector<Program> _programs,
	PhaserMode phaserMode
)
{
	assert(phaserMode == PhaserMode::PrintOptimisedPrograms || phaserMode == PhaserMode::PrintOptimisedASTs);
	assert(_programs.size() == _arguments["input-files"].as<vector<string>>().size());

	if (_population.individuals().size() == 0)
	{
		cout << "<EMPTY POPULATION>" << endl;
		return;
	}

	vector<string> const& paths = _arguments["input-files"].as<vector<string>>();
	for (auto& individual: _population.individuals())
	{
		cout << "Chromosome: " << individual.chromosome << endl;

		for (size_t i = 0; i < _programs.size(); ++i)
		{
			for (size_t j = 0; j < _arguments["chromosome-repetitions"].as<size_t>(); ++j)
				_programs[i].optimise(individual.chromosome.optimisationSteps());

			cout << "Program: " << paths[i] << endl;
			if (phaserMode == PhaserMode::PrintOptimisedPrograms)
				cout << _programs[i] << endl;
			else
				cout << _programs[i].toJson() << endl;
		}
	}
}