linear_solver.cc

Go to the documentation of this file.

// Copyright 2010-2021 Google LLC
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
//     http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.

//

#include "ortools/linear_solver/linear_solver.h"

#if !defined(_MSC_VER)
#include <unistd.h>
#endif

#include <atomic>
#include <cmath>
#include <cstddef>
#include <cstdint>
#include <utility>

#include "absl/status/status.h"
#include "absl/status/statusor.h"
#include "absl/strings/ascii.h"
#include "absl/strings/match.h"
#include "absl/strings/str_cat.h"
#include "absl/strings/str_format.h"
#include "absl/strings/str_replace.h"
#include "absl/synchronization/mutex.h"
#include "absl/synchronization/notification.h"
#include "absl/time/time.h"
#include "ortools/base/accurate_sum.h"
#include "ortools/base/commandlineflags.h"
#include "ortools/base/integral_types.h"
#include "ortools/base/logging.h"
#include "ortools/base/map_util.h"
#include "ortools/base/stl_util.h"
#include "ortools/base/threadpool.h"
#include "ortools/linear_solver/linear_solver.pb.h"
#include "ortools/linear_solver/model_exporter.h"
#include "ortools/linear_solver/model_validator.h"
#include "ortools/port/file.h"
#include "ortools/util/fp_utils.h"
#include "ortools/util/time_limit.h"

ABSL_FLAG(bool, verify_solution, false,
          "Systematically verify the solution when calling Solve()"
          ", and change the return value of Solve() to ABNORMAL if"
          " an error was detected.");
ABSL_FLAG(bool, log_verification_errors, true,
          "If --verify_solution is set: LOG(ERROR) all errors detected"
          " during the verification of the solution.");
ABSL_FLAG(bool, linear_solver_enable_verbose_output, false,
          "If set, enables verbose output for the solver. Setting this flag"
          " is the same as calling MPSolver::EnableOutput().");

ABSL_FLAG(bool, mpsolver_bypass_model_validation, false,
          "If set, the user-provided Model won't be verified before Solve()."
          " Invalid models will typically trigger various error responses"
          " from the underlying solvers; sometimes crashes.");

namespace operations_research {

bool SolverTypeIsMip(MPModelRequest::SolverType solver_type) {
  switch (solver_type) {
    case MPModelRequest::GLOP_LINEAR_PROGRAMMING:
    case MPModelRequest::CLP_LINEAR_PROGRAMMING:
    case MPModelRequest::GLPK_LINEAR_PROGRAMMING:
    case MPModelRequest::GUROBI_LINEAR_PROGRAMMING:
    case MPModelRequest::XPRESS_LINEAR_PROGRAMMING:
    case MPModelRequest::CPLEX_LINEAR_PROGRAMMING:
      return false;

    case MPModelRequest::SCIP_MIXED_INTEGER_PROGRAMMING:
    case MPModelRequest::GLPK_MIXED_INTEGER_PROGRAMMING:
    case MPModelRequest::CBC_MIXED_INTEGER_PROGRAMMING:
    case MPModelRequest::GUROBI_MIXED_INTEGER_PROGRAMMING:
    case MPModelRequest::KNAPSACK_MIXED_INTEGER_PROGRAMMING:
    case MPModelRequest::BOP_INTEGER_PROGRAMMING:
    case MPModelRequest::SAT_INTEGER_PROGRAMMING:
    case MPModelRequest::XPRESS_MIXED_INTEGER_PROGRAMMING:
    case MPModelRequest::CPLEX_MIXED_INTEGER_PROGRAMMING:
      return true;
  }
  LOG(DFATAL) << "Invalid SolverType: " << solver_type;
  return false;
}

double MPConstraint::GetCoefficient(const MPVariable* const var) const {
  DLOG_IF(DFATAL, !interface_->solver_->OwnsVariable(var)) << var;
  if (var == nullptr) return 0.0;
  return gtl::FindWithDefault(coefficients_, var, 0.0);
}

void MPConstraint::SetCoefficient(const MPVariable* const var, double coeff) {
  DLOG_IF(DFATAL, !interface_->solver_->OwnsVariable(var)) << var;
  if (var == nullptr) return;
  if (coeff == 0.0) {
    auto it = coefficients_.find(var);
    // If setting a coefficient to 0 when this coefficient did not
    // exist or was already 0, do nothing: skip
    // interface_->SetCoefficient() and do not store a coefficient in
    // the map.  Note that if the coefficient being set to 0 did exist
    // and was not 0, we do have to keep a 0 in the coefficients_ map,
    // because the extraction of the constraint might rely on it,
    // depending on the underlying solver.
    if (it != coefficients_.end() && it->second != 0.0) {
      const double old_value = it->second;
      it->second = 0.0;
      interface_->SetCoefficient(this, var, 0.0, old_value);
    }
    return;
  }
  auto insertion_result = coefficients_.insert(std::make_pair(var, coeff));
  const double old_value =
      insertion_result.second ? 0.0 : insertion_result.first->second;
  insertion_result.first->second = coeff;
  interface_->SetCoefficient(this, var, coeff, old_value);
}

void MPConstraint::Clear() {
  interface_->ClearConstraint(this);
  coefficients_.clear();
}

void MPConstraint::SetBounds(double lb, double ub) {
  const bool change = lb != lb_ || ub != ub_;
  lb_ = lb;
  ub_ = ub;
  if (change && interface_->constraint_is_extracted(index_)) {
    interface_->SetConstraintBounds(index_, lb_, ub_);
  }
}

double MPConstraint::dual_value() const {
  if (!interface_->IsContinuous()) {
    LOG(DFATAL) << "Dual value only available for continuous problems";
    return 0.0;
  }
  if (!interface_->CheckSolutionIsSynchronizedAndExists()) return 0.0;
  return dual_value_;
}

MPSolver::BasisStatus MPConstraint::basis_status() const {
  if (!interface_->IsContinuous()) {
    LOG(DFATAL) << "Basis status only available for continuous problems";
    return MPSolver::FREE;
  }
  if (!interface_->CheckSolutionIsSynchronizedAndExists()) {
    return MPSolver::FREE;
  }
  // This is done lazily as this method is expected to be rarely used.
  return interface_->row_status(index_);
}

bool MPConstraint::ContainsNewVariables() {
  const int last_variable_index = interface_->last_variable_index();
  for (const auto& entry : coefficients_) {
    const int variable_index = entry.first->index();
    if (variable_index >= last_variable_index ||
        !interface_->variable_is_extracted(variable_index)) {
      return true;
    }
  }
  return false;
}

// ----- MPObjective -----

double MPObjective::GetCoefficient(const MPVariable* const var) const {
  DLOG_IF(DFATAL, !interface_->solver_->OwnsVariable(var)) << var;
  if (var == nullptr) return 0.0;
  return gtl::FindWithDefault(coefficients_, var, 0.0);
}

void MPObjective::SetCoefficient(const MPVariable* const var, double coeff) {
  DLOG_IF(DFATAL, !interface_->solver_->OwnsVariable(var)) << var;
  if (var == nullptr) return;
  if (coeff == 0.0) {
    auto it = coefficients_.find(var);
    // See the discussion on MPConstraint::SetCoefficient() for 0 coefficients,
    // the same reasoning applies here.
    if (it == coefficients_.end() || it->second == 0.0) return;
    it->second = 0.0;
  } else {
    coefficients_[var] = coeff;
  }
  interface_->SetObjectiveCoefficient(var, coeff);
}

void MPObjective::SetOffset(double value) {
  offset_ = value;
  interface_->SetObjectiveOffset(offset_);
}

namespace {
void CheckLinearExpr(const MPSolver& solver, const LinearExpr& linear_expr) {
  for (auto var_value_pair : linear_expr.terms()) {
    CHECK(solver.OwnsVariable(var_value_pair.first))
        << "Bad MPVariable* in LinearExpr, did you try adding an integer to an "
           "MPVariable* directly?";
  }
}
}  // namespace

void MPObjective::OptimizeLinearExpr(const LinearExpr& linear_expr,
                                     bool is_maximization) {
  CheckLinearExpr(*interface_->solver_, linear_expr);
  interface_->ClearObjective();
  coefficients_.clear();
  SetOffset(linear_expr.offset());
  for (const auto& kv : linear_expr.terms()) {
    SetCoefficient(kv.first, kv.second);
  }
  SetOptimizationDirection(is_maximization);
}

void MPObjective::AddLinearExpr(const LinearExpr& linear_expr) {
  CheckLinearExpr(*interface_->solver_, linear_expr);
  SetOffset(offset_ + linear_expr.offset());
  for (const auto& kv : linear_expr.terms()) {
    SetCoefficient(kv.first, GetCoefficient(kv.first) + kv.second);
  }
}

void MPObjective::Clear() {
  interface_->ClearObjective();
  coefficients_.clear();
  offset_ = 0.0;
  SetMinimization();
}

void MPObjective::SetOptimizationDirection(bool maximize) {
  // Note(user): The maximize_ bool would more naturally belong to the
  // MPObjective, but it actually has to be a member of MPSolverInterface,
  // because some implementations (such as GLPK) need that bool for the
  // MPSolverInterface constructor, i.e. at a time when the MPObjective is not
  // constructed yet (MPSolverInterface is always built before MPObjective
  // when a new MPSolver is constructed).
  interface_->maximize_ = maximize;
  interface_->SetOptimizationDirection(maximize);
}

bool MPObjective::maximization() const { return interface_->maximize_; }

bool MPObjective::minimization() const { return !interface_->maximize_; }

double MPObjective::Value() const {
  // Note(user): implementation-wise, the objective value belongs more
  // naturally to the MPSolverInterface, since all of its implementations write
  // to it directly.
  return interface_->objective_value();
}

double MPObjective::BestBound() const {
  // Note(user): the best objective bound belongs to the interface for the
  // same reasons as the objective value does.
  return interface_->best_objective_bound();
}

// ----- MPVariable -----

double MPVariable::solution_value() const {
  if (!interface_->CheckSolutionIsSynchronizedAndExists()) return 0.0;
  // If the underlying solver supports integer variables, and this is an integer
  // variable, we round the solution value (i.e., clients usually expect precise
  // integer values for integer variables).
  return (integer_ && interface_->IsMIP()) ? round(solution_value_)
                                           : solution_value_;
}

double MPVariable::unrounded_solution_value() const {
  if (!interface_->CheckSolutionIsSynchronizedAndExists()) return 0.0;
  return solution_value_;
}

double MPVariable::reduced_cost() const {
  if (!interface_->IsContinuous()) {
    LOG(DFATAL) << "Reduced cost only available for continuous problems";
    return 0.0;
  }
  if (!interface_->CheckSolutionIsSynchronizedAndExists()) return 0.0;
  return reduced_cost_;
}

MPSolver::BasisStatus MPVariable::basis_status() const {
  if (!interface_->IsContinuous()) {
    LOG(DFATAL) << "Basis status only available for continuous problems";
    return MPSolver::FREE;
  }
  if (!interface_->CheckSolutionIsSynchronizedAndExists()) {
    return MPSolver::FREE;
  }
  // This is done lazily as this method is expected to be rarely used.
  return interface_->column_status(index_);
}

void MPVariable::SetBounds(double lb, double ub) {
  const bool change = lb != lb_ || ub != ub_;
  lb_ = lb;
  ub_ = ub;
  if (change && interface_->variable_is_extracted(index_)) {
    interface_->SetVariableBounds(index_, lb_, ub_);
  }
}

void MPVariable::SetInteger(bool integer) {
  if (integer_ != integer) {
    integer_ = integer;
    if (interface_->variable_is_extracted(index_)) {
      interface_->SetVariableInteger(index_, integer);
    }
  }
}

void MPVariable::SetBranchingPriority(int priority) {
  if (priority == branching_priority_) return;
  branching_priority_ = priority;
  interface_->BranchingPriorityChangedForVariable(index_);
}

// ----- Interface shortcuts -----

bool MPSolver::IsMIP() const { return interface_->IsMIP(); }

std::string MPSolver::SolverVersion() const {
  return interface_->SolverVersion();
}

void* MPSolver::underlying_solver() { return interface_->underlying_solver(); }

// ---- Solver-specific parameters ----

absl::Status MPSolver::SetNumThreads(int num_threads) {
  if (num_threads < 1) {
    return absl::InvalidArgumentError("num_threads must be a positive number.");
  }
  const absl::Status status = interface_->SetNumThreads(num_threads);
  if (status.ok()) {
    num_threads_ = num_threads;
  }
  return status;
}

bool MPSolver::SetSolverSpecificParametersAsString(
    const std::string& parameters) {
  solver_specific_parameter_string_ = parameters;
  return interface_->SetSolverSpecificParametersAsString(parameters);
}

// ----- Solver -----

#if defined(USE_CLP) || defined(USE_CBC)
extern MPSolverInterface* BuildCLPInterface(MPSolver* const solver);
#endif
#if defined(USE_CBC)
extern MPSolverInterface* BuildCBCInterface(MPSolver* const solver);
#endif
#if defined(USE_GLPK)
extern MPSolverInterface* BuildGLPKInterface(bool mip, MPSolver* const solver);
#endif
extern MPSolverInterface* BuildBopInterface(MPSolver* const solver);
extern MPSolverInterface* BuildGLOPInterface(MPSolver* const solver);
extern MPSolverInterface* BuildSatInterface(MPSolver* const solver);
#if defined(USE_SCIP)
extern MPSolverInterface* BuildSCIPInterface(MPSolver* const solver);
#endif
extern MPSolverInterface* BuildGurobiInterface(bool mip,
                                               MPSolver* const solver);
#if defined(USE_CPLEX)
extern MPSolverInterface* BuildCplexInterface(bool mip, MPSolver* const solver);

extern MPSolverInterface* BuildGLOPInterface(MPSolver* const solver);
#endif
#if defined(USE_XPRESS)
extern MPSolverInterface* BuildXpressInterface(bool mip,
                                               MPSolver* const solver);
#endif

namespace {
MPSolverInterface* BuildSolverInterface(MPSolver* const solver) {
  DCHECK(solver != nullptr);
  switch (solver->ProblemType()) {
    case MPSolver::BOP_INTEGER_PROGRAMMING:
      return BuildBopInterface(solver);
    case MPSolver::SAT_INTEGER_PROGRAMMING:
      return BuildSatInterface(solver);
    case MPSolver::GLOP_LINEAR_PROGRAMMING:
      return BuildGLOPInterface(solver);
#if defined(USE_GLPK)
    case MPSolver::GLPK_LINEAR_PROGRAMMING:
      return BuildGLPKInterface(false, solver);
    case MPSolver::GLPK_MIXED_INTEGER_PROGRAMMING:
      return BuildGLPKInterface(true, solver);
#endif
#if defined(USE_CLP) || defined(USE_CBC)
    case MPSolver::CLP_LINEAR_PROGRAMMING:
      return BuildCLPInterface(solver);
#endif
#if defined(USE_CBC)
    case MPSolver::CBC_MIXED_INTEGER_PROGRAMMING:
      return BuildCBCInterface(solver);
#endif
#if defined(USE_SCIP)
    case MPSolver::SCIP_MIXED_INTEGER_PROGRAMMING:
      return BuildSCIPInterface(solver);
#endif
    case MPSolver::GUROBI_LINEAR_PROGRAMMING:
      return BuildGurobiInterface(false, solver);
    case MPSolver::GUROBI_MIXED_INTEGER_PROGRAMMING:
      return BuildGurobiInterface(true, solver);
#if defined(USE_CPLEX)
    case MPSolver::CPLEX_LINEAR_PROGRAMMING:
      return BuildCplexInterface(false, solver);
    case MPSolver::CPLEX_MIXED_INTEGER_PROGRAMMING:
      return BuildCplexInterface(true, solver);
#endif
#if defined(USE_XPRESS)
    case MPSolver::XPRESS_MIXED_INTEGER_PROGRAMMING:
      return BuildXpressInterface(true, solver);
    case MPSolver::XPRESS_LINEAR_PROGRAMMING:
      return BuildXpressInterface(false, solver);
#endif
    default:
      // TODO(user): Revert to the best *available* interface.
      LOG(FATAL) << "Linear solver not recognized.";
  }
  return nullptr;
}
}  // namespace

namespace {
int NumDigits(int n) {
// Number of digits needed to write a non-negative integer in base 10.
// Note(user): max(1, log(0) + 1) == max(1, -inf) == 1.
#if defined(_MSC_VER)
  return static_cast<int>(std::max(1.0L, log(1.0L * n) / log(10.0L) + 1.0));
#else
  return static_cast<int>(std::max(1.0, log10(static_cast<double>(n)) + 1.0));
#endif
}
}  // namespace

MPSolver::MPSolver(const std::string& name,
                   OptimizationProblemType problem_type)
    : name_(name),
      problem_type_(problem_type),
      construction_time_(absl::Now()) {
  interface_.reset(BuildSolverInterface(this));
  if (absl::GetFlag(FLAGS_linear_solver_enable_verbose_output)) {
    EnableOutput();
  }
  objective_.reset(new MPObjective(interface_.get()));
}

MPSolver::~MPSolver() { Clear(); }

extern bool GurobiIsCorrectlyInstalled();

// static
bool MPSolver::SupportsProblemType(OptimizationProblemType problem_type) {
#ifdef USE_CLP
  if (problem_type == CLP_LINEAR_PROGRAMMING) return true;
#endif
#ifdef USE_GLPK
  if (problem_type == GLPK_LINEAR_PROGRAMMING ||
      problem_type == GLPK_MIXED_INTEGER_PROGRAMMING) {
    return true;
  }
#endif
  if (problem_type == BOP_INTEGER_PROGRAMMING) return true;
  if (problem_type == SAT_INTEGER_PROGRAMMING) return true;
  if (problem_type == GLOP_LINEAR_PROGRAMMING) return true;
  if (problem_type == GUROBI_LINEAR_PROGRAMMING ||
      problem_type == GUROBI_MIXED_INTEGER_PROGRAMMING) {
    return GurobiIsCorrectlyInstalled();
  }
#ifdef USE_SCIP
  if (problem_type == SCIP_MIXED_INTEGER_PROGRAMMING) return true;
#endif
#ifdef USE_CBC
  if (problem_type == CBC_MIXED_INTEGER_PROGRAMMING) return true;
#endif
#ifdef USE_XPRESS
  if (problem_type == XPRESS_MIXED_INTEGER_PROGRAMMING ||
      problem_type == XPRESS_LINEAR_PROGRAMMING) {
    return true;
  }
#endif
#ifdef USE_CPLEX
  if (problem_type == CPLEX_LINEAR_PROGRAMMING ||
      problem_type == CPLEX_MIXED_INTEGER_PROGRAMMING) {
    return true;
  }
#endif
  return false;
}

// TODO(user): post c++ 14, instead use
//   std::pair<MPSolver::OptimizationProblemType, const absl::string_view>
// once pair gets a constexpr constructor.
namespace {
struct NamedOptimizationProblemType {
  MPSolver::OptimizationProblemType problem_type;
  absl::string_view name;
};
}  // namespace

#if defined(_MSC_VER)
const
#else
constexpr
#endif
    NamedOptimizationProblemType kOptimizationProblemTypeNames[] = {
        {MPSolver::GLOP_LINEAR_PROGRAMMING, "glop"},
        {MPSolver::CLP_LINEAR_PROGRAMMING, "clp"},
        {MPSolver::GUROBI_LINEAR_PROGRAMMING, "gurobi_lp"},
        {MPSolver::GLPK_LINEAR_PROGRAMMING, "glpk_lp"},
        {MPSolver::CPLEX_LINEAR_PROGRAMMING, "cplex_lp"},
        {MPSolver::XPRESS_LINEAR_PROGRAMMING, "xpress_lp"},
        {MPSolver::SCIP_MIXED_INTEGER_PROGRAMMING, "scip"},
        {MPSolver::CBC_MIXED_INTEGER_PROGRAMMING, "cbc"},
        {MPSolver::SAT_INTEGER_PROGRAMMING, "sat"},
        {MPSolver::BOP_INTEGER_PROGRAMMING, "bop"},
        {MPSolver::GUROBI_MIXED_INTEGER_PROGRAMMING, "gurobi"},
        {MPSolver::GLPK_MIXED_INTEGER_PROGRAMMING, "glpk"},
        {MPSolver::KNAPSACK_MIXED_INTEGER_PROGRAMMING, "knapsack"},
        {MPSolver::CPLEX_MIXED_INTEGER_PROGRAMMING, "cplex"},
        {MPSolver::XPRESS_MIXED_INTEGER_PROGRAMMING, "xpress"},

};
// static
bool MPSolver::ParseSolverType(absl::string_view solver_id,
                               MPSolver::OptimizationProblemType* type) {
  // Normalize the solver id.
  const std::string id =
      absl::StrReplaceAll(absl::AsciiStrToUpper(solver_id), {{"-", "_"}});

  // Support the full enum name
  MPModelRequest::SolverType solver_type;
  if (MPModelRequest::SolverType_Parse(id, &solver_type)) {
    *type = static_cast<MPSolver::OptimizationProblemType>(solver_type);
    return true;
  }

  // Names are stored in lower case.
  std::string lower_id = absl::AsciiStrToLower(id);

  // Remove any "_mip" suffix, since they are optional.
  if (absl::EndsWith(lower_id, "_mip")) {
    lower_id = lower_id.substr(0, lower_id.size() - 4);
  }

  // Rewrite CP-SAT into SAT.
  if (lower_id == "cp_sat") {
    lower_id = "sat";
  }

  // Reverse lookup in the kOptimizationProblemTypeNames[] array.
  for (auto& named_solver : kOptimizationProblemTypeNames) {
    if (named_solver.name == lower_id) {
      *type = named_solver.problem_type;
      return true;
    }
  }

  return false;
}

const absl::string_view ToString(
    MPSolver::OptimizationProblemType optimization_problem_type) {
  for (const auto& named_solver : kOptimizationProblemTypeNames) {
    if (named_solver.problem_type == optimization_problem_type) {
      return named_solver.name;
    }
  }
  LOG(FATAL) << "Unrecognized solver type: "
             << static_cast<int>(optimization_problem_type);
  return "";
}

bool AbslParseFlag(const absl::string_view text,
                   MPSolver::OptimizationProblemType* solver_type,
                   std::string* error) {
  DCHECK(solver_type != nullptr);
  DCHECK(error != nullptr);
  const bool result = MPSolver::ParseSolverType(text, solver_type);
  if (!result) {
    *error = absl::StrCat("Solver type: ", text, " does not exist.");
  }
  return result;
}

/* static */
MPSolver::OptimizationProblemType MPSolver::ParseSolverTypeOrDie(
    const std::string& solver_id) {
  MPSolver::OptimizationProblemType problem_type;
  CHECK(MPSolver::ParseSolverType(solver_id, &problem_type)) << solver_id;
  return problem_type;
}

/* static */
MPSolver* MPSolver::CreateSolver(const std::string& solver_id) {
  MPSolver::OptimizationProblemType problem_type;
  if (!MPSolver::ParseSolverType(solver_id, &problem_type)) {
    LOG(WARNING) << "Unrecognized solver type: " << solver_id;
    return nullptr;
  }
  if (!MPSolver::SupportsProblemType(problem_type)) {
    LOG(WARNING) << "Support for " << solver_id
                 << " not linked in, or the license was not found.";
    return nullptr;
  }
  MPSolver* solver = new MPSolver("", problem_type);
  return solver;
}

MPVariable* MPSolver::LookupVariableOrNull(const std::string& var_name) const {
  if (!variable_name_to_index_) GenerateVariableNameIndex();

  absl::flat_hash_map<std::string, int>::const_iterator it =
      variable_name_to_index_->find(var_name);
  if (it == variable_name_to_index_->end()) return nullptr;
  return variables_[it->second];
}

MPConstraint* MPSolver::LookupConstraintOrNull(
    const std::string& constraint_name) const {
  if (!constraint_name_to_index_) GenerateConstraintNameIndex();

  const auto it = constraint_name_to_index_->find(constraint_name);
  if (it == constraint_name_to_index_->end()) return nullptr;
  return constraints_[it->second];
}

// ----- Methods using protocol buffers -----

MPSolverResponseStatus MPSolver::LoadModelFromProto(
    const MPModelProto& input_model, std::string* error_message) {
  Clear();

  // The variable and constraint names are dropped, because we allow
  // duplicate names in the proto (they're not considered as 'ids'),
  // unlike the MPSolver C++ API which crashes if there are duplicate names.
  // Clearing the names makes the MPSolver generate unique names.
  return LoadModelFromProtoInternal(input_model, /*clear_names=*/true,
                                    /*check_model_validity=*/true,
                                    error_message);
}

MPSolverResponseStatus MPSolver::LoadModelFromProtoWithUniqueNamesOrDie(
    const MPModelProto& input_model, std::string* error_message) {
  Clear();

  // Force variable and constraint name indexing (which CHECKs name uniqueness).
  GenerateVariableNameIndex();
  GenerateConstraintNameIndex();

  return LoadModelFromProtoInternal(input_model, /*clear_names=*/false,
                                    /*check_model_validity=*/true,
                                    error_message);
}

MPSolverResponseStatus MPSolver::LoadModelFromProtoInternal(
    const MPModelProto& input_model, bool clear_names,
    bool check_model_validity, std::string* error_message) {
  CHECK(error_message != nullptr);
  if (check_model_validity) {
    const std::string error = FindErrorInMPModelProto(input_model);
    if (!error.empty()) {
      *error_message = error;
      LOG_IF(INFO, OutputIsEnabled())
          << "Invalid model given to LoadModelFromProto(): " << error;
      if (absl::GetFlag(FLAGS_mpsolver_bypass_model_validation)) {
        LOG_IF(INFO, OutputIsEnabled())
            << "Ignoring the model error(s) because of"
            << " --mpsolver_bypass_model_validation.";
      } else {
        return absl::StrContains(error, "Infeasible") ? MPSOLVER_INFEASIBLE
                                                      : MPSOLVER_MODEL_INVALID;
      }
    }
  }

  if (input_model.has_quadratic_objective()) {
    *error_message =
        "Optimizing a quadratic objective is only supported through direct "
        "proto solves. Please use MPSolver::SolveWithProto, or the solver's "
        "direct proto solve function.";
    return MPSOLVER_MODEL_INVALID;
  }

  MPObjective* const objective = MutableObjective();
  // Passing empty names makes the MPSolver generate unique names.
  const std::string empty;
  for (int i = 0; i < input_model.variable_size(); ++i) {
    const MPVariableProto& var_proto = input_model.variable(i);
    MPVariable* variable =
        MakeNumVar(var_proto.lower_bound(), var_proto.upper_bound(),
                   clear_names ? empty : var_proto.name());
    variable->SetInteger(var_proto.is_integer());
    if (var_proto.branching_priority() != 0) {
      variable->SetBranchingPriority(var_proto.branching_priority());
    }
    objective->SetCoefficient(variable, var_proto.objective_coefficient());
  }

  for (const MPConstraintProto& ct_proto : input_model.constraint()) {
    if (ct_proto.lower_bound() == -infinity() &&
        ct_proto.upper_bound() == infinity()) {
      continue;
    }

    MPConstraint* const ct =
        MakeRowConstraint(ct_proto.lower_bound(), ct_proto.upper_bound(),
                          clear_names ? empty : ct_proto.name());
    ct->set_is_lazy(ct_proto.is_lazy());
    for (int j = 0; j < ct_proto.var_index_size(); ++j) {
      ct->SetCoefficient(variables_[ct_proto.var_index(j)],
                         ct_proto.coefficient(j));
    }
  }

  for (const MPGeneralConstraintProto& general_constraint :
       input_model.general_constraint()) {
    switch (general_constraint.general_constraint_case()) {
      case MPGeneralConstraintProto::kIndicatorConstraint: {
        const auto& proto =
            general_constraint.indicator_constraint().constraint();
        if (proto.lower_bound() == -infinity() &&
            proto.upper_bound() == infinity()) {
          continue;
        }

        const int constraint_index = NumConstraints();
        MPConstraint* const constraint = new MPConstraint(
            constraint_index, proto.lower_bound(), proto.upper_bound(),
            clear_names ? "" : proto.name(), interface_.get());
        if (constraint_name_to_index_) {
          gtl::InsertOrDie(&*constraint_name_to_index_, constraint->name(),
                           constraint_index);
        }
        constraints_.push_back(constraint);
        constraint_is_extracted_.push_back(false);

        constraint->set_is_lazy(proto.is_lazy());
        for (int j = 0; j < proto.var_index_size(); ++j) {
          constraint->SetCoefficient(variables_[proto.var_index(j)],
                                     proto.coefficient(j));
        }

        MPVariable* const variable =
            variables_[general_constraint.indicator_constraint().var_index()];
        constraint->indicator_variable_ = variable;
        constraint->indicator_value_ =
            general_constraint.indicator_constraint().var_value();

        if (!interface_->AddIndicatorConstraint(constraint)) {
          *error_message = "Solver doesn't support indicator constraints";
          return MPSOLVER_MODEL_INVALID;
        }
        break;
      }
      default:
        *error_message = absl::StrFormat(
            "Optimizing general constraints of type %i is only supported "
            "through direct proto solves. Please use MPSolver::SolveWithProto, "
            "or the solver's direct proto solve function.",
            general_constraint.general_constraint_case());
        return MPSOLVER_MODEL_INVALID;
    }
  }

  objective->SetOptimizationDirection(input_model.maximize());
  if (input_model.has_objective_offset()) {
    objective->SetOffset(input_model.objective_offset());
  }

  // Stores any hints about where to start the solve.
  solution_hint_.clear();
  for (int i = 0; i < input_model.solution_hint().var_index_size(); ++i) {
    solution_hint_.push_back(
        std::make_pair(variables_[input_model.solution_hint().var_index(i)],
                       input_model.solution_hint().var_value(i)));
  }
  return MPSOLVER_MODEL_IS_VALID;
}

namespace {
MPSolverResponseStatus ResultStatusToMPSolverResponseStatus(
    MPSolver::ResultStatus status) {
  switch (status) {
    case MPSolver::OPTIMAL:
      return MPSOLVER_OPTIMAL;
    case MPSolver::FEASIBLE:
      return MPSOLVER_FEASIBLE;
    case MPSolver::INFEASIBLE:
      return MPSOLVER_INFEASIBLE;
    case MPSolver::UNBOUNDED:
      return MPSOLVER_UNBOUNDED;
    case MPSolver::ABNORMAL:
      return MPSOLVER_ABNORMAL;
    case MPSolver::MODEL_INVALID:
      return MPSOLVER_MODEL_INVALID;
    case MPSolver::NOT_SOLVED:
      return MPSOLVER_NOT_SOLVED;
  }
  return MPSOLVER_UNKNOWN_STATUS;
}
}  // namespace

void MPSolver::FillSolutionResponseProto(MPSolutionResponse* response) const {
  CHECK(response != nullptr);
  response->Clear();
  response->set_status(
      ResultStatusToMPSolverResponseStatus(interface_->result_status_));
  if (interface_->result_status_ == MPSolver::OPTIMAL ||
      interface_->result_status_ == MPSolver::FEASIBLE) {
    response->set_objective_value(Objective().Value());
    for (int i = 0; i < variables_.size(); ++i) {
      response->add_variable_value(variables_[i]->solution_value());
    }

    if (interface_->IsMIP()) {
      response->set_best_objective_bound(interface_->best_objective_bound());
    } else {
      // Dual values have no meaning in MIP.
      for (int j = 0; j < constraints_.size(); ++j) {
        response->add_dual_value(constraints_[j]->dual_value());
      }
      // Reduced cost have no meaning in MIP.
      for (int i = 0; i < variables_.size(); ++i) {
        response->add_reduced_cost(variables_[i]->reduced_cost());
      }
    }
  }
}

namespace {
bool InCategory(int status, int category) {
  if (category == MPSOLVER_OPTIMAL) return status == MPSOLVER_OPTIMAL;
  while (status > category) status >>= 4;
  return status == category;
}

void AppendStatusStr(const std::string& msg, MPSolutionResponse* response) {
  response->set_status_str(
      absl::StrCat(response->status_str(),
                   (response->status_str().empty() ? "" : "\n"), msg));
}
}  // namespace

// static
void MPSolver::SolveWithProto(const MPModelRequest& model_request,
                              MPSolutionResponse* response,
                              std::atomic<bool>* interrupt) {
  CHECK(response != nullptr);

  if (interrupt != nullptr &&
      !SolverTypeSupportsInterruption(model_request.solver_type())) {
    response->set_status(MPSOLVER_INCOMPATIBLE_OPTIONS);
    response->set_status_str(
        "Called MPSolver::SolveWithProto with an underlying solver that "
        "doesn't support interruption.");
    return;
  }

  MPSolver solver(model_request.model().name(),
                  static_cast<MPSolver::OptimizationProblemType>(
                      model_request.solver_type()));
  if (model_request.enable_internal_solver_output()) {
    solver.EnableOutput();
  }

  // If interruption support is not required, we don't need access to the
  // underlying solver and can solve it directly if the interface supports it.
  auto optional_response =
      solver.interface_->DirectlySolveProto(model_request, interrupt);
  if (optional_response) {
    *response = std::move(optional_response).value();
    return;
  }

  const absl::optional<LazyMutableCopy<MPModelProto>> optional_model =
      ExtractValidMPModelOrPopulateResponseStatus(model_request, response);
  if (!optional_model) {
    LOG_IF(WARNING, model_request.enable_internal_solver_output())
        << "Failed to extract a valid model from protocol buffer. Status: "
        << ProtoEnumToString<MPSolverResponseStatus>(response->status()) << " ("
        << response->status() << "): " << response->status_str();
    return;
  }
  std::string error_message;
  response->set_status(solver.LoadModelFromProtoInternal(
      optional_model->get(), /*clear_names=*/true,
      /*check_model_validity=*/false, &error_message));
  // Even though we don't re-check model validity here, there can be some
  // problems found by LoadModelFromProto, eg. unsupported features.
  if (response->status() != MPSOLVER_MODEL_IS_VALID) {
    response->set_status_str(error_message);
    LOG_IF(WARNING, model_request.enable_internal_solver_output())
        << "LoadModelFromProtoInternal() failed even though the model was "
        << "valid! Status: "
        << ProtoEnumToString<MPSolverResponseStatus>(response->status()) << " ("
        << response->status() << "); Error: " << error_message;
    return;
  }
  if (model_request.has_solver_time_limit_seconds()) {
    solver.SetTimeLimit(
        absl::Seconds(model_request.solver_time_limit_seconds()));
  }
  std::string warning_message;
  if (model_request.has_solver_specific_parameters()) {
    if (!solver.SetSolverSpecificParametersAsString(
            model_request.solver_specific_parameters())) {
      if (model_request.ignore_solver_specific_parameters_failure()) {
        // We'll add a warning message in status_str after the solve.
        warning_message =
            "Warning: the solver specific parameters were not successfully "
            "applied";
      } else {
        response->set_status(MPSOLVER_MODEL_INVALID_SOLVER_PARAMETERS);
        return;
      }
    }
  }

  if (interrupt == nullptr) {
    // If we don't need interruption support, we can save some overhead by
    // running the solve in the current thread.
    solver.Solve();
    solver.FillSolutionResponseProto(response);
  } else {
    const absl::Time start_time = absl::Now();
    absl::Time interrupt_time;
    bool interrupted_by_user = false;
    {
      absl::Notification solve_finished;
      auto polling_func = [&interrupt, &solve_finished, &solver,
                           &interrupted_by_user, &interrupt_time,
                           &model_request]() {
        constexpr absl::Duration kPollDelay = absl::Microseconds(100);
        constexpr absl::Duration kMaxInterruptionDelay = absl::Seconds(10);

        while (!interrupt->load()) {
          if (solve_finished.HasBeenNotified()) return;
          absl::SleepFor(kPollDelay);
        }

        // If we get here, we received an interruption notification before the
        // solve finished "naturally".
        solver.InterruptSolve();
        interrupt_time = absl::Now();
        interrupted_by_user = true;

        // SUBTLE: our call to InterruptSolve() can be ignored by the
        // underlying solver for several reasons:
        // 1) The solver thread doesn't poll its 'interrupted' bit often
        //    enough and takes too long to realize that it should return, or
        //    its mere return + FillSolutionResponse() takes too long.
        // 2) The user interrupted the solve so early that Solve() hadn't
        //    really started yet when we called InterruptSolve().
        // In case 1), we should just wait a little longer. In case 2), we
        // should call InterruptSolve() again, maybe several times. To both
        // accommodate cases where the solver takes really a long time to
        // react to the interruption, while returning as quickly as possible,
        // we poll the solve_finished notification with increasing durations
        // and call InterruptSolve again, each time.
        for (absl::Duration poll_delay = kPollDelay;
             absl::Now() <= interrupt_time + kMaxInterruptionDelay;
             poll_delay *= 2) {
          if (solve_finished.WaitForNotificationWithTimeout(poll_delay)) {
            return;
          } else {
            solver.InterruptSolve();
          }
        }

        LOG(DFATAL)
            << "MPSolver::InterruptSolve() seems to be ignored by the "
               "underlying solver, despite repeated calls over at least "
            << absl::FormatDuration(kMaxInterruptionDelay)
            << ". Solver type used: "
            << MPModelRequest_SolverType_Name(model_request.solver_type());

        // Note that in opt builds, the polling thread terminates here with an
        // error message, but we let Solve() finish, ignoring the user
        // interruption request.
      };

      // The choice to do polling rather than solving in the second thread is
      // not arbitrary, as we want to maintain any custom thread options set by
      // the user. They shouldn't matter for polling, but for solving we might
      // e.g. use a larger stack.
      ThreadPool thread_pool("SolverThread", /*num_threads=*/1);
      thread_pool.StartWorkers();
      thread_pool.Schedule(polling_func);

      // Make sure the interruption notification didn't arrive while waiting to
      // be scheduled.
      if (!interrupt->load()) {
        solver.Solve();
        solver.FillSolutionResponseProto(response);
      } else {  // *interrupt == true
        response->set_status(MPSOLVER_CANCELLED_BY_USER);
        response->set_status_str(
            "Solve not started, because the user set the atomic<bool> in "
            "MPSolver::SolveWithProto() to true before solving could "
            "start.");
      }
      solve_finished.Notify();

      // We block until the thread finishes when thread_pool goes out of scope.
    }

    if (interrupted_by_user) {
      // Despite the interruption, the solver might still have found a useful
      // result. If so, don't overwrite the status.
      if (InCategory(response->status(), MPSOLVER_NOT_SOLVED)) {
        response->set_status(MPSOLVER_CANCELLED_BY_USER);
      }
      AppendStatusStr(
          absl::StrFormat(
              "User interrupted MPSolver::SolveWithProto() by setting the "
              "atomic<bool> to true at %s (%s after solving started.)",
              absl::FormatTime(interrupt_time),
              absl::FormatDuration(interrupt_time - start_time)),
          response);
    }
  }

  if (!warning_message.empty()) {
    AppendStatusStr(warning_message, response);
  }
}

void MPSolver::ExportModelToProto(MPModelProto* output_model) const {
  DCHECK(output_model != nullptr);
  output_model->Clear();
  // Name
  output_model->set_name(Name());
  // Variables
  for (int j = 0; j < variables_.size(); ++j) {
    const MPVariable* const var = variables_[j];
    MPVariableProto* const variable_proto = output_model->add_variable();
    // TODO(user): Add option to avoid filling the var name to avoid overly
    // large protocol buffers.
    variable_proto->set_name(var->name());
    variable_proto->set_lower_bound(var->lb());
    variable_proto->set_upper_bound(var->ub());
    variable_proto->set_is_integer(var->integer());
    if (objective_->GetCoefficient(var) != 0.0) {
      variable_proto->set_objective_coefficient(
          objective_->GetCoefficient(var));
    }
    if (var->branching_priority() != 0) {
      variable_proto->set_branching_priority(var->branching_priority());
    }
  }

  // Map the variables to their indices. This is needed to output the
  // variables in the order they were created, which in turn is needed to have
  // repeatable results with ExportModelAsLpFormat and ExportModelAsMpsFormat.
  // This step is needed as long as the variable indices are given by the
  // underlying solver at the time of model extraction.
  // TODO(user): remove this step.
  absl::flat_hash_map<const MPVariable*, int> var_to_index;
  for (int j = 0; j < variables_.size(); ++j) {
    var_to_index[variables_[j]] = j;
  }

  // Constraints
  for (int i = 0; i < constraints_.size(); ++i) {
    MPConstraint* const constraint = constraints_[i];
    MPConstraintProto* constraint_proto;
    if (constraint->indicator_variable() != nullptr) {
      MPGeneralConstraintProto* const general_constraint_proto =
          output_model->add_general_constraint();
      general_constraint_proto->set_name(constraint->name());
      MPIndicatorConstraint* const indicator_constraint_proto =
          general_constraint_proto->mutable_indicator_constraint();
      indicator_constraint_proto->set_var_index(
          constraint->indicator_variable()->index());
      indicator_constraint_proto->set_var_value(constraint->indicator_value());
      constraint_proto = indicator_constraint_proto->mutable_constraint();
    } else {
      constraint_proto = output_model->add_constraint();
    }
    constraint_proto->set_name(constraint->name());
    constraint_proto->set_lower_bound(constraint->lb());
    constraint_proto->set_upper_bound(constraint->ub());
    constraint_proto->set_is_lazy(constraint->is_lazy());
    // Vector linear_term will contain pairs (variable index, coeff), that will
    // be sorted by variable index.
    std::vector<std::pair<int, double>> linear_term;
    for (const auto& entry : constraint->coefficients_) {
      const MPVariable* const var = entry.first;
      const int var_index = gtl::FindWithDefault(var_to_index, var, -1);
      DCHECK_NE(-1, var_index);
      const double coeff = entry.second;
      linear_term.push_back(std::pair<int, double>(var_index, coeff));
    }
    // The cost of sort is expected to be low as constraints usually have very
    // few terms.
    std::sort(linear_term.begin(), linear_term.end());
    // Now use linear term.
    for (const std::pair<int, double>& var_and_coeff : linear_term) {
      constraint_proto->add_var_index(var_and_coeff.first);
      constraint_proto->add_coefficient(var_and_coeff.second);
    }
  }

  output_model->set_maximize(Objective().maximization());
  output_model->set_objective_offset(Objective().offset());

  if (!solution_hint_.empty()) {
    PartialVariableAssignment* const hint =
        output_model->mutable_solution_hint();
    for (const auto& var_value_pair : solution_hint_) {
      hint->add_var_index(var_value_pair.first->index());
      hint->add_var_value(var_value_pair.second);
    }
  }
}

absl::Status MPSolver::LoadSolutionFromProto(const MPSolutionResponse& response,
                                             double tolerance) {
  interface_->result_status_ = static_cast<ResultStatus>(response.status());
  if (response.status() != MPSOLVER_OPTIMAL &&
      response.status() != MPSOLVER_FEASIBLE) {
    return absl::InvalidArgumentError(absl::StrCat(
        "Cannot load a solution unless its status is OPTIMAL or FEASIBLE"
        " (status was: ",
        ProtoEnumToString<MPSolverResponseStatus>(response.status()), ")"));
  }
  // Before touching the variables, verify that the solution looks legit:
  // each variable of the MPSolver must have its value listed exactly once, and
  // each listed solution should correspond to a known variable.
  if (response.variable_value_size() != variables_.size()) {
    return absl::InvalidArgumentError(absl::StrCat(
        "Trying to load a solution whose number of variables (",
        response.variable_value_size(),
        ") does not correspond to the Solver's (", variables_.size(), ")"));
  }
  interface_->ExtractModel();

  if (tolerance != infinity()) {
    // Look further: verify that the variable values are within the bounds.
    double largest_error = 0;
    int num_vars_out_of_bounds = 0;
    int last_offending_var = -1;
    for (int i = 0; i < response.variable_value_size(); ++i) {
      const double var_value = response.variable_value(i);
      MPVariable* var = variables_[i];
      // TODO(user): Use parameter when they become available in this class.
      const double lb_error = var->lb() - var_value;
      const double ub_error = var_value - var->ub();
      if (lb_error > tolerance || ub_error > tolerance) {
        ++num_vars_out_of_bounds;
        largest_error = std::max(largest_error, std::max(lb_error, ub_error));
        last_offending_var = i;
      }
    }
    if (num_vars_out_of_bounds > 0) {
      return absl::InvalidArgumentError(absl::StrCat(
          "Loaded a solution whose variables matched the solver's, but ",
          num_vars_out_of_bounds, " of ", variables_.size(),
          " variables were out of their bounds, by more than the primal"
          " tolerance which is: ",
          tolerance, ". Max error: ", largest_error, ", last offender var is #",
          last_offending_var, ": '", variables_[last_offending_var]->name(),
          "'"));
    }
  }
  for (int i = 0; i < response.variable_value_size(); ++i) {
    variables_[i]->set_solution_value(response.variable_value(i));
  }
  if (response.dual_value_size() > 0) {
    if (response.dual_value_size() != constraints_.size()) {
      return absl::InvalidArgumentError(absl::StrCat(
          "Trying to load a dual solution whose number of entries (",
          response.dual_value_size(), ") does not correspond to the Solver's (",
          constraints_.size(), ")"));
    }
    for (int i = 0; i < response.dual_value_size(); ++i) {
      constraints_[i]->set_dual_value(response.dual_value(i));
    }
  }
  if (response.reduced_cost_size() > 0) {
    if (response.reduced_cost_size() != variables_.size()) {
      return absl::InvalidArgumentError(absl::StrCat(
          "Trying to load a reduced cost solution whose number of entries (",
          response.reduced_cost_size(),
          ") does not correspond to the Solver's (", variables_.size(), ")"));
    }
    for (int i = 0; i < response.reduced_cost_size(); ++i) {
      variables_[i]->set_reduced_cost(response.reduced_cost(i));
    }
  }
  // Set the objective value, if is known.
  // NOTE(user): We do not verify the objective, even though we could!
  if (response.has_objective_value()) {
    interface_->objective_value_ = response.objective_value();
  }
  if (response.has_best_objective_bound()) {
    interface_->best_objective_bound_ = response.best_objective_bound();
  }
  // Mark the status as SOLUTION_SYNCHRONIZED, so that users may inspect the
  // solution normally.
  interface_->sync_status_ = MPSolverInterface::SOLUTION_SYNCHRONIZED;
  return absl::OkStatus();
}

void MPSolver::Clear() {
  {
    absl::MutexLock lock(&global_count_mutex_);
    global_num_variables_ += variables_.size();
    global_num_constraints_ += constraints_.size();
  }
  MutableObjective()->Clear();
  gtl::STLDeleteElements(&variables_);
  gtl::STLDeleteElements(&constraints_);
  if (variable_name_to_index_) {
    variable_name_to_index_->clear();
  }
  variable_is_extracted_.clear();
  if (constraint_name_to_index_) {
    constraint_name_to_index_->clear();
  }
  constraint_is_extracted_.clear();
  interface_->Reset();
  solution_hint_.clear();
}

void MPSolver::Reset() { interface_->Reset(); }

bool MPSolver::InterruptSolve() { return interface_->InterruptSolve(); }

void MPSolver::SetStartingLpBasis(
    const std::vector<BasisStatus>& variable_statuses,
    const std::vector<BasisStatus>& constraint_statuses) {
  interface_->SetStartingLpBasis(variable_statuses, constraint_statuses);
}

MPVariable* MPSolver::MakeVar(double lb, double ub, bool integer,
                              const std::string& name) {
  const int var_index = NumVariables();
  MPVariable* v =
      new MPVariable(var_index, lb, ub, integer, name, interface_.get());
  if (variable_name_to_index_) {
    gtl::InsertOrDie(&*variable_name_to_index_, v->name(), var_index);
  }
  variables_.push_back(v);
  variable_is_extracted_.push_back(false);
  interface_->AddVariable(v);
  return v;
}

MPVariable* MPSolver::MakeNumVar(double lb, double ub,
                                 const std::string& name) {
  return MakeVar(lb, ub, false, name);
}

MPVariable* MPSolver::MakeIntVar(double lb, double ub,
                                 const std::string& name) {
  return MakeVar(lb, ub, true, name);
}

MPVariable* MPSolver::MakeBoolVar(const std::string& name) {
  return MakeVar(0.0, 1.0, true, name);
}

void MPSolver::MakeVarArray(int nb, double lb, double ub, bool integer,
                            const std::string& name,
                            std::vector<MPVariable*>* vars) {
  DCHECK_GE(nb, 0);
  if (nb <= 0) return;
  const int num_digits = NumDigits(nb);
  for (int i = 0; i < nb; ++i) {
    if (name.empty()) {
      vars->push_back(MakeVar(lb, ub, integer, name));
    } else {
      std::string vname =
          absl::StrFormat("%s%0*d", name.c_str(), num_digits, i);
      vars->push_back(MakeVar(lb, ub, integer, vname));
    }
  }
}

void MPSolver::MakeNumVarArray(int nb, double lb, double ub,
                               const std::string& name,
                               std::vector<MPVariable*>* vars) {
  MakeVarArray(nb, lb, ub, false, name, vars);
}

void MPSolver::MakeIntVarArray(int nb, double lb, double ub,
                               const std::string& name,
                               std::vector<MPVariable*>* vars) {
  MakeVarArray(nb, lb, ub, true, name, vars);
}

void MPSolver::MakeBoolVarArray(int nb, const std::string& name,
                                std::vector<MPVariable*>* vars) {
  MakeVarArray(nb, 0.0, 1.0, true, name, vars);
}

MPConstraint* MPSolver::MakeRowConstraint(double lb, double ub) {
  return MakeRowConstraint(lb, ub, "");
}

MPConstraint* MPSolver::MakeRowConstraint() {
  return MakeRowConstraint(-infinity(), infinity(), "");
}

MPConstraint* MPSolver::MakeRowConstraint(double lb, double ub,
                                          const std::string& name) {
  const int constraint_index = NumConstraints();
  MPConstraint* const constraint =
      new MPConstraint(constraint_index, lb, ub, name, interface_.get());
  if (constraint_name_to_index_) {
    gtl::InsertOrDie(&*constraint_name_to_index_, constraint->name(),
                     constraint_index);
  }
  constraints_.push_back(constraint);
  constraint_is_extracted_.push_back(false);
  interface_->AddRowConstraint(constraint);
  return constraint;
}

MPConstraint* MPSolver::MakeRowConstraint(const std::string& name) {
  return MakeRowConstraint(-infinity(), infinity(), name);
}

MPConstraint* MPSolver::MakeRowConstraint(const LinearRange& range) {
  return MakeRowConstraint(range, "");
}

MPConstraint* MPSolver::MakeRowConstraint(const LinearRange& range,
                                          const std::string& name) {
  CheckLinearExpr(*this, range.linear_expr());
  MPConstraint* constraint =
      MakeRowConstraint(range.lower_bound(), range.upper_bound(), name);
  for (const auto& kv : range.linear_expr().terms()) {
    constraint->SetCoefficient(kv.first, kv.second);
  }
  return constraint;
}

int MPSolver::ComputeMaxConstraintSize(int min_constraint_index,
                                       int max_constraint_index) const {
  int max_constraint_size = 0;
  DCHECK_GE(min_constraint_index, 0);
  DCHECK_LE(max_constraint_index, constraints_.size());
  for (int i = min_constraint_index; i < max_constraint_index; ++i) {
    MPConstraint* const ct = constraints_[i];
    if (ct->coefficients_.size() > max_constraint_size) {
      max_constraint_size = ct->coefficients_.size();
    }
  }
  return max_constraint_size;
}

bool MPSolver::HasInfeasibleConstraints() const {
  bool hasInfeasibleConstraints = false;
  for (int i = 0; i < constraints_.size(); ++i) {
    if (constraints_[i]->lb() > constraints_[i]->ub()) {
      LOG(WARNING) << "Constraint " << constraints_[i]->name() << " (" << i
                   << ") has contradictory bounds:"
                   << " lower bound = " << constraints_[i]->lb()
                   << " upper bound = " << constraints_[i]->ub();
      hasInfeasibleConstraints = true;
    }
  }
  return hasInfeasibleConstraints;
}

bool MPSolver::HasIntegerVariables() const {
  for (const MPVariable* const variable : variables_) {
    if (variable->integer()) return true;
  }
  return false;
}

MPSolver::ResultStatus MPSolver::Solve() {
  MPSolverParameters default_param;
  return Solve(default_param);
}

MPSolver::ResultStatus MPSolver::Solve(const MPSolverParameters& param) {
  // Special case for infeasible constraints so that all solvers have
  // the same behavior.
  // TODO(user): replace this by model extraction to proto + proto validation
  // (the proto has very low overhead compared to the wrapper, both in
  // performance and memory, so it's ok).
  if (HasInfeasibleConstraints()) {
    interface_->result_status_ = MPSolver::INFEASIBLE;
    return interface_->result_status_;
  }

  MPSolver::ResultStatus status = interface_->Solve(param);
  if (absl::GetFlag(FLAGS_verify_solution)) {
    if (status != MPSolver::OPTIMAL && status != MPSolver::FEASIBLE) {
      VLOG(1) << "--verify_solution enabled, but the solver did not find a"
              << " solution: skipping the verification.";
    } else if (!VerifySolution(
                   param.GetDoubleParam(MPSolverParameters::PRIMAL_TOLERANCE),
                   absl::GetFlag(FLAGS_log_verification_errors))) {
      status = MPSolver::ABNORMAL;
      interface_->result_status_ = status;
    }
  }
  DCHECK_EQ(interface_->result_status_, status);
  return status;
}

void MPSolver::Write(const std::string& file_name) {
  interface_->Write(file_name);
}

namespace {
std::string PrettyPrintVar(const MPVariable& var) {
  const std::string prefix = "Variable '" + var.name() + "': domain = ";
  if (var.lb() >= MPSolver::infinity() || var.ub() <= -MPSolver::infinity() ||
      var.lb() > var.ub()) {
    return prefix + "∅";  // Empty set.
  }
  // Special case: integer variable with at most two possible values
  // (and potentially none).
  if (var.integer() && var.ub() - var.lb() <= 1) {
    const int64_t lb = static_cast<int64_t>(ceil(var.lb()));
    const int64_t ub = static_cast<int64_t>(floor(var.ub()));
    if (lb > ub) {
      return prefix + "∅";
    } else if (lb == ub) {
      return absl::StrFormat("%s{ %d }", prefix.c_str(), lb);
    } else {
      return absl::StrFormat("%s{ %d, %d }", prefix.c_str(), lb, ub);
    }
  }
  // Special case: single (non-infinite) real value.
  if (var.lb() == var.ub()) {
    return absl::StrFormat("%s{ %f }", prefix.c_str(), var.lb());
  }
  return prefix + (var.integer() ? "Integer" : "Real") + " in " +
         (var.lb() <= -MPSolver::infinity()
              ? std::string("]-∞")
              : absl::StrFormat("[%f", var.lb())) +
         ", " +
         (var.ub() >= MPSolver::infinity() ? std::string("+∞[")
                                           : absl::StrFormat("%f]", var.ub()));
}

std::string PrettyPrintConstraint(const MPConstraint& constraint) {
  std::string prefix = "Constraint '" + constraint.name() + "': ";
  if (constraint.lb() >= MPSolver::infinity() ||
      constraint.ub() <= -MPSolver::infinity() ||
      constraint.lb() > constraint.ub()) {
    return prefix + "ALWAYS FALSE";
  }
  if (constraint.lb() <= -MPSolver::infinity() &&
      constraint.ub() >= MPSolver::infinity()) {
    return prefix + "ALWAYS TRUE";
  }
  prefix += "<linear expr>";
  // Equality.
  if (constraint.lb() == constraint.ub()) {
    return absl::StrFormat("%s = %f", prefix.c_str(), constraint.lb());
  }
  // Inequalities.
  if (constraint.lb() <= -MPSolver::infinity()) {
    return absl::StrFormat("%s ≤ %f", prefix.c_str(), constraint.ub());
  }
  if (constraint.ub() >= MPSolver::infinity()) {
    return absl::StrFormat("%s ≥ %f", prefix.c_str(), constraint.lb());
  }
  return absl::StrFormat("%s ∈ [%f, %f]", prefix.c_str(), constraint.lb(),
                         constraint.ub());
}
}  // namespace

absl::Status MPSolver::ClampSolutionWithinBounds() {
  interface_->ExtractModel();
  for (MPVariable* const variable : variables_) {
    const double value = variable->solution_value();
    if (std::isnan(value)) {
      return absl::InvalidArgumentError(
          absl::StrCat("NaN value for ", PrettyPrintVar(*variable)));
    }
    if (value < variable->lb()) {
      variable->set_solution_value(variable->lb());
    } else if (value > variable->ub()) {
      variable->set_solution_value(variable->ub());
    }
  }
  interface_->sync_status_ = MPSolverInterface::SOLUTION_SYNCHRONIZED;
  return absl::OkStatus();
}

std::vector<double> MPSolver::ComputeConstraintActivities() const {
  // TODO(user): test this failure case.
  if (!interface_->CheckSolutionIsSynchronizedAndExists()) return {};
  std::vector<double> activities(constraints_.size(), 0.0);
  for (int i = 0; i < constraints_.size(); ++i) {
    const MPConstraint& constraint = *constraints_[i];
    AccurateSum<double> sum;
    for (const auto& entry : constraint.coefficients_) {
      sum.Add(entry.first->solution_value() * entry.second);
    }
    activities[i] = sum.Value();
  }
  return activities;
}

// TODO(user): split.
bool MPSolver::VerifySolution(double tolerance, bool log_errors) const {
  double max_observed_error = 0;
  if (tolerance < 0) tolerance = infinity();
  int num_errors = 0;

  // Verify variables.
  for (int i = 0; i < variables_.size(); ++i) {
    const MPVariable& var = *variables_[i];
    const double value = var.solution_value();
    // Check for NaN.
    if (std::isnan(value)) {
      ++num_errors;
      max_observed_error = infinity();
      LOG_IF(ERROR, log_errors) << "NaN value for " << PrettyPrintVar(var);
      continue;
    }
    // Check lower bound.
    if (var.lb() != -infinity()) {
      if (value < var.lb() - tolerance) {
        ++num_errors;
        max_observed_error = std::max(max_observed_error, var.lb() - value);
        LOG_IF(ERROR, log_errors)
            << "Value " << value << " too low for " << PrettyPrintVar(var);
      }
    }
    // Check upper bound.
    if (var.ub() != infinity()) {
      if (value > var.ub() + tolerance) {
        ++num_errors;
        max_observed_error = std::max(max_observed_error, value - var.ub());
        LOG_IF(ERROR, log_errors)
            << "Value " << value << " too high for " << PrettyPrintVar(var);
      }
    }
    // Check integrality.
    if (IsMIP() && var.integer()) {
      if (fabs(value - round(value)) > tolerance) {
        ++num_errors;
        max_observed_error =
            std::max(max_observed_error, fabs(value - round(value)));
        LOG_IF(ERROR, log_errors)
            << "Non-integer value " << value << " for " << PrettyPrintVar(var);
      }
    }
  }
  if (!IsMIP() && HasIntegerVariables()) {
    LOG_IF(INFO, log_errors) << "Skipped variable integrality check, because "
                             << "a continuous relaxation of the model was "
                             << "solved (i.e., the selected solver does not "
                             << "support integer variables).";
  }

  // Verify constraints.
  const std::vector<double> activities = ComputeConstraintActivities();
  for (int i = 0; i < constraints_.size(); ++i) {
    const MPConstraint& constraint = *constraints_[i];
    const double activity = activities[i];
    // Re-compute the activity with a inaccurate summing algorithm.
    double inaccurate_activity = 0.0;
    for (const auto& entry : constraint.coefficients_) {
      inaccurate_activity += entry.first->solution_value() * entry.second;
    }
    // Catch NaNs.
    if (std::isnan(activity) || std::isnan(inaccurate_activity)) {
      ++num_errors;
      max_observed_error = infinity();
      LOG_IF(ERROR, log_errors)
          << "NaN value for " << PrettyPrintConstraint(constraint);
      continue;
    }
    // Check bounds.
    if (constraint.indicator_variable() == nullptr ||
        std::round(constraint.indicator_variable()->solution_value()) ==
            constraint.indicator_value()) {
      if (constraint.lb() != -infinity()) {
        if (activity < constraint.lb() - tolerance) {
          ++num_errors;
          max_observed_error =
              std::max(max_observed_error, constraint.lb() - activity);
          LOG_IF(ERROR, log_errors)
              << "Activity " << activity << " too low for "
              << PrettyPrintConstraint(constraint);
        } else if (inaccurate_activity < constraint.lb() - tolerance) {
          LOG_IF(WARNING, log_errors)
              << "Activity " << activity << ", computed with the (inaccurate)"
              << " standard sum of its terms, is too low for "
              << PrettyPrintConstraint(constraint);
        }
      }
      if (constraint.ub() != infinity()) {
        if (activity > constraint.ub() + tolerance) {
          ++num_errors;
          max_observed_error =
              std::max(max_observed_error, activity - constraint.ub());
          LOG_IF(ERROR, log_errors)
              << "Activity " << activity << " too high for "
              << PrettyPrintConstraint(constraint);
        } else if (inaccurate_activity > constraint.ub() + tolerance) {
          LOG_IF(WARNING, log_errors)
              << "Activity " << activity << ", computed with the (inaccurate)"
              << " standard sum of its terms, is too high for "
              << PrettyPrintConstraint(constraint);
        }
      }
    }
  }

  // Verify that the objective value wasn't reported incorrectly.
  const MPObjective& objective = Objective();
  AccurateSum<double> objective_sum;
  objective_sum.Add(objective.offset());
  double inaccurate_objective_value = objective.offset();
  for (const auto& entry : objective.coefficients_) {
    const double term = entry.first->solution_value() * entry.second;
    objective_sum.Add(term);
    inaccurate_objective_value += term;
  }
  const double actual_objective_value = objective_sum.Value();
  if (!AreWithinAbsoluteOrRelativeTolerances(
          objective.Value(), actual_objective_value, tolerance, tolerance)) {
    ++num_errors;
    max_observed_error = std::max(
        max_observed_error, fabs(actual_objective_value - objective.Value()));
    LOG_IF(ERROR, log_errors)
        << "Objective value " << objective.Value() << " isn't accurate"
        << ", it should be " << actual_objective_value
        << " (delta=" << actual_objective_value - objective.Value() << ").";
  } else if (!AreWithinAbsoluteOrRelativeTolerances(objective.Value(),
                                                    inaccurate_objective_value,
                                                    tolerance, tolerance)) {
    LOG_IF(WARNING, log_errors)
        << "Objective value " << objective.Value() << " doesn't correspond"
        << " to the value computed with the standard (and therefore inaccurate)"
        << " sum of its terms.";
  }
  if (num_errors > 0) {
    LOG_IF(ERROR, log_errors)
        << "There were " << num_errors << " errors above the tolerance ("
        << tolerance << "), the largest was " << max_observed_error;
    return false;
  }
  return true;
}

bool MPSolver::OutputIsEnabled() const { return !interface_->quiet(); }

void MPSolver::EnableOutput() { interface_->set_quiet(false); }

void MPSolver::SuppressOutput() { interface_->set_quiet(true); }

int64_t MPSolver::iterations() const { return interface_->iterations(); }

int64_t MPSolver::nodes() const { return interface_->nodes(); }

double MPSolver::ComputeExactConditionNumber() const {
  return interface_->ComputeExactConditionNumber();
}

bool MPSolver::OwnsVariable(const MPVariable* var) const {
  if (var == nullptr) return false;
  if (var->index() >= 0 && var->index() < variables_.size()) {
    // Then, verify that the variable with this index has the same address.
    return variables_[var->index()] == var;
  }
  return false;
}

bool MPSolver::ExportModelAsLpFormat(bool obfuscate,
                                     std::string* model_str) const {
  MPModelProto proto;
  ExportModelToProto(&proto);
  MPModelExportOptions options;
  options.obfuscate = obfuscate;
  const auto status_or =
      operations_research::ExportModelAsLpFormat(proto, options);
  *model_str = status_or.value_or("");
  return status_or.ok();
}

bool MPSolver::ExportModelAsMpsFormat(bool fixed_format, bool obfuscate,
                                      std::string* model_str) const {
  MPModelProto proto;
  ExportModelToProto(&proto);
  MPModelExportOptions options;
  options.obfuscate = obfuscate;
  const auto status_or =
      operations_research::ExportModelAsMpsFormat(proto, options);
  *model_str = status_or.value_or("");
  return status_or.ok();
}

void MPSolver::SetHint(std::vector<std::pair<const MPVariable*, double>> hint) {
  for (const auto& var_value_pair : hint) {
    CHECK(OwnsVariable(var_value_pair.first))
        << "hint variable does not belong to this solver";
  }
  solution_hint_ = std::move(hint);
}

void MPSolver::GenerateVariableNameIndex() const {
  if (variable_name_to_index_) return;
  variable_name_to_index_ = absl::flat_hash_map<std::string, int>();
  for (const MPVariable* const var : variables_) {
    gtl::InsertOrDie(&*variable_name_to_index_, var->name(), var->index());
  }
}

void MPSolver::GenerateConstraintNameIndex() const {
  if (constraint_name_to_index_) return;
  constraint_name_to_index_ = absl::flat_hash_map<std::string, int>();
  for (const MPConstraint* const cst : constraints_) {
    gtl::InsertOrDie(&*constraint_name_to_index_, cst->name(), cst->index());
  }
}

bool MPSolver::NextSolution() { return interface_->NextSolution(); }

void MPSolver::SetCallback(MPCallback* mp_callback) {
  interface_->SetCallback(mp_callback);
}

bool MPSolver::SupportsCallbacks() const {
  return interface_->SupportsCallbacks();
}

// Global counters.
absl::Mutex MPSolver::global_count_mutex_(absl::kConstInit);
int64_t MPSolver::global_num_variables_ = 0;
int64_t MPSolver::global_num_constraints_ = 0;

// static
int64_t MPSolver::global_num_variables() {
  // Why not ReaderMutexLock? See go/totw/197#when-are-shared-locks-useful.
  absl::MutexLock lock(&global_count_mutex_);
  return global_num_variables_;
}

// static
int64_t MPSolver::global_num_constraints() {
  // Why not ReaderMutexLock? See go/totw/197#when-are-shared-locks-useful.
  absl::MutexLock lock(&global_count_mutex_);
  return global_num_constraints_;
}

bool MPSolverResponseStatusIsRpcError(MPSolverResponseStatus status) {
  switch (status) {
    // Cases that don't yield an RPC error when they happen on the server.
    case MPSOLVER_OPTIMAL:
    case MPSOLVER_FEASIBLE:
    case MPSOLVER_INFEASIBLE:
    case MPSOLVER_NOT_SOLVED:
    case MPSOLVER_UNBOUNDED:
    case MPSOLVER_ABNORMAL:
    case MPSOLVER_UNKNOWN_STATUS:
      return false;
    // Cases that should never happen with the linear solver server. We prefer
    // to consider those as "not RPC errors".
    case MPSOLVER_MODEL_IS_VALID:
    case MPSOLVER_CANCELLED_BY_USER:
      return false;
    // Cases that yield an RPC error when they happen on the server.
    case MPSOLVER_MODEL_INVALID:
    case MPSOLVER_MODEL_INVALID_SOLUTION_HINT:
    case MPSOLVER_MODEL_INVALID_SOLVER_PARAMETERS:
    case MPSOLVER_SOLVER_TYPE_UNAVAILABLE:
    case MPSOLVER_INCOMPATIBLE_OPTIONS:
      return true;
  }
  LOG(DFATAL)
      << "MPSolverResponseStatusIsRpcError() called with invalid status "
      << "(value: " << status << ")";
  return false;
}

// ---------- MPSolverInterface ----------

const int MPSolverInterface::kDummyVariableIndex = 0;

// TODO(user): Initialize objective value and bound to +/- inf (depending on
// optimization direction).
MPSolverInterface::MPSolverInterface(MPSolver* const solver)
    : solver_(solver),
      sync_status_(MODEL_SYNCHRONIZED),
      result_status_(MPSolver::NOT_SOLVED),
      maximize_(false),
      last_constraint_index_(0),
      last_variable_index_(0),
      objective_value_(0.0),
      best_objective_bound_(0.0),
      quiet_(true) {}

MPSolverInterface::~MPSolverInterface() {}

void MPSolverInterface::Write(const std::string& filename) {
  LOG(WARNING) << "Writing model not implemented in this solver interface.";
}

void MPSolverInterface::ExtractModel() {
  switch (sync_status_) {
    case MUST_RELOAD: {
      ExtractNewVariables();
      ExtractNewConstraints();
      ExtractObjective();

      last_constraint_index_ = solver_->constraints_.size();
      last_variable_index_ = solver_->variables_.size();
      sync_status_ = MODEL_SYNCHRONIZED;
      break;
    }
    case MODEL_SYNCHRONIZED: {
      // Everything has already been extracted.
      DCHECK_EQ(last_constraint_index_, solver_->constraints_.size());
      DCHECK_EQ(last_variable_index_, solver_->variables_.size());
      break;
    }
    case SOLUTION_SYNCHRONIZED: {
      // Nothing has changed since last solve.
      DCHECK_EQ(last_constraint_index_, solver_->constraints_.size());
      DCHECK_EQ(last_variable_index_, solver_->variables_.size());
      break;
    }
  }
}

// TODO(user): remove this method.
void MPSolverInterface::ResetExtractionInformation() {
  sync_status_ = MUST_RELOAD;
  last_constraint_index_ = 0;
  last_variable_index_ = 0;
  solver_->variable_is_extracted_.assign(solver_->variables_.size(), false);
  solver_->constraint_is_extracted_.assign(solver_->constraints_.size(), false);
}

bool MPSolverInterface::CheckSolutionIsSynchronized() const {
  if (sync_status_ != SOLUTION_SYNCHRONIZED) {
    LOG(DFATAL)
        << "The model has been changed since the solution was last computed."
        << " MPSolverInterface::sync_status_ = " << sync_status_;
    return false;
  }
  return true;
}

// Default version that can be overwritten by a solver-specific
// version to accommodate for the quirks of each solver.
bool MPSolverInterface::CheckSolutionExists() const {
  if (result_status_ != MPSolver::OPTIMAL &&
      result_status_ != MPSolver::FEASIBLE) {
    LOG(DFATAL) << "No solution exists. MPSolverInterface::result_status_ = "
                << result_status_;
    return false;
  }
  return true;
}

double MPSolverInterface::objective_value() const {
  if (!CheckSolutionIsSynchronizedAndExists()) return 0;
  return objective_value_;
}

double MPSolverInterface::best_objective_bound() const {
  const double trivial_worst_bound =
      maximize_ ? -std::numeric_limits<double>::infinity()
                : std::numeric_limits<double>::infinity();
  if (!IsMIP()) {
    VLOG(1) << "Best objective bound only available for discrete problems.";
    return trivial_worst_bound;
  }
  if (!CheckSolutionIsSynchronized()) {
    return trivial_worst_bound;
  }
  // Special case for empty model.
  if (solver_->variables_.empty() && solver_->constraints_.empty()) {
    return solver_->Objective().offset();
  }
  return best_objective_bound_;
}

void MPSolverInterface::InvalidateSolutionSynchronization() {
  if (sync_status_ == SOLUTION_SYNCHRONIZED) {
    sync_status_ = MODEL_SYNCHRONIZED;
  }
}

double MPSolverInterface::ComputeExactConditionNumber() const {
  // Override this method in interfaces that actually support it.
  LOG(DFATAL) << "ComputeExactConditionNumber not implemented for "
              << ProtoEnumToString<MPModelRequest::SolverType>(
                     static_cast<MPModelRequest::SolverType>(
                         solver_->ProblemType()));
  return 0.0;
}

void MPSolverInterface::SetCommonParameters(const MPSolverParameters& param) {
  // TODO(user): Overhaul the code that sets parameters to enable changing
  // GLOP parameters without issuing warnings.
  // By default, we let GLOP keep its own default tolerance, much more accurate
  // than for the rest of the solvers.
  //
  if (solver_->ProblemType() != MPSolver::GLOP_LINEAR_PROGRAMMING) {
    SetPrimalTolerance(
        param.GetDoubleParam(MPSolverParameters::PRIMAL_TOLERANCE));
    SetDualTolerance(param.GetDoubleParam(MPSolverParameters::DUAL_TOLERANCE));
  }
  SetPresolveMode(param.GetIntegerParam(MPSolverParameters::PRESOLVE));
  // TODO(user): In the future, we could distinguish between the
  // algorithm to solve the root LP and the algorithm to solve node
  // LPs. Not sure if underlying solvers support it.
  int value = param.GetIntegerParam(MPSolverParameters::LP_ALGORITHM);
  if (value != MPSolverParameters::kDefaultIntegerParamValue) {
    SetLpAlgorithm(value);
  }
}

void MPSolverInterface::SetMIPParameters(const MPSolverParameters& param) {
  if (solver_->ProblemType() != MPSolver::GLOP_LINEAR_PROGRAMMING) {
    SetRelativeMipGap(
        param.GetDoubleParam(MPSolverParameters::RELATIVE_MIP_GAP));
  }
}

void MPSolverInterface::SetUnsupportedDoubleParam(
    MPSolverParameters::DoubleParam param) {
  LOG(WARNING) << "Trying to set an unsupported parameter: " << param << ".";
}
void MPSolverInterface::SetUnsupportedIntegerParam(
    MPSolverParameters::IntegerParam param) {
  LOG(WARNING) << "Trying to set an unsupported parameter: " << param << ".";
}
void MPSolverInterface::SetDoubleParamToUnsupportedValue(
    MPSolverParameters::DoubleParam param, double value) {
  LOG(WARNING) << "Trying to set a supported parameter: " << param
               << " to an unsupported value: " << value;
}
void MPSolverInterface::SetIntegerParamToUnsupportedValue(
    MPSolverParameters::IntegerParam param, int value) {
  LOG(WARNING) << "Trying to set a supported parameter: " << param
               << " to an unsupported value: " << value;
}

absl::Status MPSolverInterface::SetNumThreads(int num_threads) {
  return absl::UnimplementedError(
      absl::StrFormat("SetNumThreads() not supported by %s.", SolverVersion()));
}

bool MPSolverInterface::SetSolverSpecificParametersAsString(
    const std::string& parameters) {
  if (parameters.empty()) {
    return true;
  }

  LOG(WARNING) << "SetSolverSpecificParametersAsString() not supported by "
               << SolverVersion();
  return false;
}

// ---------- MPSolverParameters ----------

const double MPSolverParameters::kDefaultRelativeMipGap = 1e-4;
// For the primal and dual tolerances, choose the same default as CLP and GLPK.
const double MPSolverParameters::kDefaultPrimalTolerance =
    operations_research::kDefaultPrimalTolerance;
const double MPSolverParameters::kDefaultDualTolerance = 1e-7;
const MPSolverParameters::PresolveValues MPSolverParameters::kDefaultPresolve =
    MPSolverParameters::PRESOLVE_ON;
const MPSolverParameters::IncrementalityValues
    MPSolverParameters::kDefaultIncrementality =
        MPSolverParameters::INCREMENTALITY_ON;

const double MPSolverParameters::kDefaultDoubleParamValue = -1.0;
const int MPSolverParameters::kDefaultIntegerParamValue = -1;
const double MPSolverParameters::kUnknownDoubleParamValue = -2.0;
const int MPSolverParameters::kUnknownIntegerParamValue = -2;

// The constructor sets all parameters to their default value.
MPSolverParameters::MPSolverParameters()
    : relative_mip_gap_value_(kDefaultRelativeMipGap),
      primal_tolerance_value_(kDefaultPrimalTolerance),
      dual_tolerance_value_(kDefaultDualTolerance),
      presolve_value_(kDefaultPresolve),
      scaling_value_(kDefaultIntegerParamValue),
      lp_algorithm_value_(kDefaultIntegerParamValue),
      incrementality_value_(kDefaultIncrementality),
      lp_algorithm_is_default_(true) {}

void MPSolverParameters::SetDoubleParam(MPSolverParameters::DoubleParam param,
                                        double value) {
  switch (param) {
    case RELATIVE_MIP_GAP: {
      relative_mip_gap_value_ = value;
      break;
    }
    case PRIMAL_TOLERANCE: {
      primal_tolerance_value_ = value;
      break;
    }
    case DUAL_TOLERANCE: {
      dual_tolerance_value_ = value;
      break;
    }
    default: {
      LOG(ERROR) << "Trying to set an unknown parameter: " << param << ".";
    }
  }
}

void MPSolverParameters::SetIntegerParam(MPSolverParameters::IntegerParam param,
                                         int value) {
  switch (param) {
    case PRESOLVE: {
      if (value != PRESOLVE_OFF && value != PRESOLVE_ON) {
        LOG(ERROR) << "Trying to set a supported parameter: " << param
                   << " to an unknown value: " << value;
      }
      presolve_value_ = value;
      break;
    }
    case SCALING: {
      if (value != SCALING_OFF && value != SCALING_ON) {
        LOG(ERROR) << "Trying to set a supported parameter: " << param
                   << " to an unknown value: " << value;
      }
      scaling_value_ = value;
      break;
    }
    case LP_ALGORITHM: {
      if (value != DUAL && value != PRIMAL && value != BARRIER) {
        LOG(ERROR) << "Trying to set a supported parameter: " << param
                   << " to an unknown value: " << value;
      }
      lp_algorithm_value_ = value;
      lp_algorithm_is_default_ = false;
      break;
    }
    case INCREMENTALITY: {
      if (value != INCREMENTALITY_OFF && value != INCREMENTALITY_ON) {
        LOG(ERROR) << "Trying to set a supported parameter: " << param
                   << " to an unknown value: " << value;
      }
      incrementality_value_ = value;
      break;
    }
    default: {
      LOG(ERROR) << "Trying to set an unknown parameter: " << param << ".";
    }
  }
}

void MPSolverParameters::ResetDoubleParam(
    MPSolverParameters::DoubleParam param) {
  switch (param) {
    case RELATIVE_MIP_GAP: {
      relative_mip_gap_value_ = kDefaultRelativeMipGap;
      break;
    }
    case PRIMAL_TOLERANCE: {
      primal_tolerance_value_ = kDefaultPrimalTolerance;
      break;
    }
    case DUAL_TOLERANCE: {
      dual_tolerance_value_ = kDefaultDualTolerance;
      break;
    }
    default: {
      LOG(ERROR) << "Trying to reset an unknown parameter: " << param << ".";
    }
  }
}

void MPSolverParameters::ResetIntegerParam(
    MPSolverParameters::IntegerParam param) {
  switch (param) {
    case PRESOLVE: {
      presolve_value_ = kDefaultPresolve;
      break;
    }
    case SCALING: {
      scaling_value_ = kDefaultIntegerParamValue;
      break;
    }
    case LP_ALGORITHM: {
      lp_algorithm_is_default_ = true;
      break;
    }
    case INCREMENTALITY: {
      incrementality_value_ = kDefaultIncrementality;
      break;
    }
    default: {
      LOG(ERROR) << "Trying to reset an unknown parameter: " << param << ".";
    }
  }
}

void MPSolverParameters::Reset() {
  ResetDoubleParam(RELATIVE_MIP_GAP);
  ResetDoubleParam(PRIMAL_TOLERANCE);
  ResetDoubleParam(DUAL_TOLERANCE);
  ResetIntegerParam(PRESOLVE);
  ResetIntegerParam(SCALING);
  ResetIntegerParam(LP_ALGORITHM);
  ResetIntegerParam(INCREMENTALITY);
}

double MPSolverParameters::GetDoubleParam(
    MPSolverParameters::DoubleParam param) const {
  switch (param) {
    case RELATIVE_MIP_GAP: {
      return relative_mip_gap_value_;
    }
    case PRIMAL_TOLERANCE: {
      return primal_tolerance_value_;
    }
    case DUAL_TOLERANCE: {
      return dual_tolerance_value_;
    }
    default: {
      LOG(ERROR) << "Trying to get an unknown parameter: " << param << ".";
      return kUnknownDoubleParamValue;
    }
  }
}

int MPSolverParameters::GetIntegerParam(
    MPSolverParameters::IntegerParam param) const {
  switch (param) {
    case PRESOLVE: {
      return presolve_value_;
    }
    case LP_ALGORITHM: {
      if (lp_algorithm_is_default_) return kDefaultIntegerParamValue;
      return lp_algorithm_value_;
    }
    case INCREMENTALITY: {
      return incrementality_value_;
    }
    case SCALING: {
      return scaling_value_;
    }
    default: {
      LOG(ERROR) << "Trying to get an unknown parameter: " << param << ".";
      return kUnknownIntegerParamValue;
    }
  }
}

}  // namespace operations_research