sat_parameters.proto

// Copyright 2010-2024 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.

syntax = "proto2";

package operations_research.sat;

option java_package = "com.google.ortools.sat";
option java_multiple_files = true;

option csharp_namespace = "Google.OrTools.Sat";

// Contains the definitions for all the sat algorithm parameters and their
// default values.
//
// NEXT TAG: 284
message SatParameters {
  // In some context, like in a portfolio of search, it makes sense to name a
  // given parameters set for logging purpose.
  optional string name = 171 [default = ""];

  // ==========================================================================
  // Branching and polarity
  // ==========================================================================

  // Variables without activity (i.e. at the beginning of the search) will be
  // tried in this preferred order.
  enum VariableOrder {
    IN_ORDER = 0;  // As specified by the problem.
    IN_REVERSE_ORDER = 1;
    IN_RANDOM_ORDER = 2;
  }
  optional VariableOrder preferred_variable_order = 1 [default = IN_ORDER];

  // Specifies the initial polarity (true/false) when the solver branches on a
  // variable. This can be modified later by the user, or the phase saving
  // heuristic.
  //
  // Note(user): POLARITY_FALSE is usually a good choice because of the
  // "natural" way to express a linear boolean problem.
  enum Polarity {
    POLARITY_TRUE = 0;
    POLARITY_FALSE = 1;
    POLARITY_RANDOM = 2;
  }
  optional Polarity initial_polarity = 2 [default = POLARITY_FALSE];

  // If this is true, then the polarity of a variable will be the last value it
  // was assigned to, or its default polarity if it was never assigned since the
  // call to ResetDecisionHeuristic().
  //
  // Actually, we use a newer version where we follow the last value in the
  // longest non-conflicting partial assignment in the current phase.
  //
  // This is called 'literal phase saving'. For details see 'A Lightweight
  // Component Caching Scheme for Satisfiability Solvers' K. Pipatsrisawat and
  // A.Darwiche, In 10th International Conference on Theory and Applications of
  // Satisfiability Testing, 2007.
  optional bool use_phase_saving = 44 [default = true];

  // If non-zero, then we change the polarity heuristic after that many number
  // of conflicts in an arithmetically increasing fashion. So x the first time,
  // 2 * x the second time, etc...
  optional int32 polarity_rephase_increment = 168 [default = 1000];

  // The proportion of polarity chosen at random. Note that this take
  // precedence over the phase saving heuristic. This is different from
  // initial_polarity:POLARITY_RANDOM because it will select a new random
  // polarity each time the variable is branched upon instead of selecting one
  // initially and then always taking this choice.
  optional double random_polarity_ratio = 45 [default = 0.0];

  // A number between 0 and 1 that indicates the proportion of branching
  // variables that are selected randomly instead of choosing the first variable
  // from the given variable_ordering strategy.
  optional double random_branches_ratio = 32 [default = 0.0];

  // Whether we use the ERWA (Exponential Recency Weighted Average) heuristic as
  // described in "Learning Rate Based Branching Heuristic for SAT solvers",
  // J.H.Liang, V. Ganesh, P. Poupart, K.Czarnecki, SAT 2016.
  optional bool use_erwa_heuristic = 75 [default = false];

  // The initial value of the variables activity. A non-zero value only make
  // sense when use_erwa_heuristic is true. Experiments with a value of 1e-2
  // together with the ERWA heuristic showed slighthly better result than simply
  // using zero. The idea is that when the "learning rate" of a variable becomes
  // lower than this value, then we prefer to branch on never explored before
  // variables. This is not in the ERWA paper.
  optional double initial_variables_activity = 76 [default = 0.0];

  // When this is true, then the variables that appear in any of the reason of
  // the variables in a conflict have their activity bumped. This is addition to
  // the variables in the conflict, and the one that were used during conflict
  // resolution.
  optional bool also_bump_variables_in_conflict_reasons = 77 [default = false];

  // ==========================================================================
  // Conflict analysis
  // ==========================================================================

  // Do we try to minimize conflicts (greedily) when creating them.
  enum ConflictMinimizationAlgorithm {
    NONE = 0;
    SIMPLE = 1;
    RECURSIVE = 2;
    EXPERIMENTAL = 3;
  }
  optional ConflictMinimizationAlgorithm minimization_algorithm = 4
      [default = RECURSIVE];

  // Whether to expoit the binary clause to minimize learned clauses further.
  enum BinaryMinizationAlgorithm {
    NO_BINARY_MINIMIZATION = 0;
    BINARY_MINIMIZATION_FIRST = 1;
    BINARY_MINIMIZATION_FIRST_WITH_TRANSITIVE_REDUCTION = 4;
    BINARY_MINIMIZATION_WITH_REACHABILITY = 2;
    EXPERIMENTAL_BINARY_MINIMIZATION = 3;
  }
  optional BinaryMinizationAlgorithm binary_minimization_algorithm = 34
      [default = BINARY_MINIMIZATION_FIRST];

  // At a really low cost, during the 1-UIP conflict computation, it is easy to
  // detect if some of the involved reasons are subsumed by the current
  // conflict. When this is true, such clauses are detached and later removed
  // from the problem.
  optional bool subsumption_during_conflict_analysis = 56 [default = true];

  // ==========================================================================
  // Clause database management
  // ==========================================================================

  // Trigger a cleanup when this number of "deletable" clauses is learned.
  optional int32 clause_cleanup_period = 11 [default = 10000];

  // During a cleanup, we will always keep that number of "deletable" clauses.
  // Note that this doesn't include the "protected" clauses.
  optional int32 clause_cleanup_target = 13 [default = 0];

  // During a cleanup, if clause_cleanup_target is 0, we will delete the
  // clause_cleanup_ratio of "deletable" clauses instead of aiming for a fixed
  // target of clauses to keep.
  optional double clause_cleanup_ratio = 190 [default = 0.5];

  // Each time a clause activity is bumped, the clause has a chance to be
  // protected during the next cleanup phase. Note that clauses used as a reason
  // are always protected.
  enum ClauseProtection {
    PROTECTION_NONE = 0;    // No protection.
    PROTECTION_ALWAYS = 1;  // Protect all clauses whose activity is bumped.
    PROTECTION_LBD = 2;     // Only protect clause with a better LBD.
  }
  optional ClauseProtection clause_cleanup_protection = 58
      [default = PROTECTION_NONE];

  // All the clauses with a LBD (literal blocks distance) lower or equal to this
  // parameters will always be kept.
  optional int32 clause_cleanup_lbd_bound = 59 [default = 5];

  // The clauses that will be kept during a cleanup are the ones that come
  // first under this order. We always keep or exclude ties together.
  enum ClauseOrdering {
    // Order clause by decreasing activity, then by increasing LBD.
    CLAUSE_ACTIVITY = 0;
    // Order clause by increasing LBD, then by decreasing activity.
    CLAUSE_LBD = 1;
  }
  optional ClauseOrdering clause_cleanup_ordering = 60
      [default = CLAUSE_ACTIVITY];

  // Same as for the clauses, but for the learned pseudo-Boolean constraints.
  optional int32 pb_cleanup_increment = 46 [default = 200];
  optional double pb_cleanup_ratio = 47 [default = 0.5];

  // ==========================================================================
  // Variable and clause activities
  // ==========================================================================

  // Each time a conflict is found, the activities of some variables are
  // increased by one. Then, the activity of all variables are multiplied by
  // variable_activity_decay.
  //
  // To implement this efficiently, the activity of all the variables is not
  // decayed at each conflict. Instead, the activity increment is multiplied by
  // 1 / decay. When an activity reach max_variable_activity_value, all the
  // activity are multiplied by 1 / max_variable_activity_value.
  optional double variable_activity_decay = 15 [default = 0.8];
  optional double max_variable_activity_value = 16 [default = 1e100];

  // The activity starts at 0.8 and increment by 0.01 every 5000 conflicts until
  // 0.95. This "hack" seems to work well and comes from:
  //
  // Glucose 2.3 in the SAT 2013 Competition - SAT Competition 2013
  // http://edacc4.informatik.uni-ulm.de/SC13/solver-description-download/136
  optional double glucose_max_decay = 22 [default = 0.95];
  optional double glucose_decay_increment = 23 [default = 0.01];
  optional int32 glucose_decay_increment_period = 24 [default = 5000];

  // Clause activity parameters (same effect as the one on the variables).
  optional double clause_activity_decay = 17 [default = 0.999];
  optional double max_clause_activity_value = 18 [default = 1e20];

  // ==========================================================================
  // Restart
  // ==========================================================================

  // Restart algorithms.
  //
  // A reference for the more advanced ones is:
  // Gilles Audemard, Laurent Simon, "Refining Restarts Strategies for SAT
  // and UNSAT", Principles and Practice of Constraint Programming Lecture
  // Notes in Computer Science 2012, pp 118-126
  enum RestartAlgorithm {
    NO_RESTART = 0;

    // Just follow a Luby sequence times restart_period.
    LUBY_RESTART = 1;

    // Moving average restart based on the decision level of conflicts.
    DL_MOVING_AVERAGE_RESTART = 2;

    // Moving average restart based on the LBD of conflicts.
    LBD_MOVING_AVERAGE_RESTART = 3;

    // Fixed period restart every restart period.
    FIXED_RESTART = 4;
  }

  // The restart strategies will change each time the strategy_counter is
  // increased. The current strategy will simply be the one at index
  // strategy_counter modulo the number of strategy. Note that if this list
  // includes a NO_RESTART, nothing will change when it is reached because the
  // strategy_counter will only increment after a restart.
  //
  // The idea of switching of search strategy tailored for SAT/UNSAT comes from
  // Chanseok Oh with his COMiniSatPS solver, see http://cs.nyu.edu/~chanseok/.
  // But more generally, it seems REALLY beneficial to try different strategy.
  repeated RestartAlgorithm restart_algorithms = 61;
  optional string default_restart_algorithms = 70
      [default =
           "LUBY_RESTART,LBD_MOVING_AVERAGE_RESTART,DL_MOVING_AVERAGE_RESTART"];

  // Restart period for the FIXED_RESTART strategy. This is also the multiplier
  // used by the LUBY_RESTART strategy.
  optional int32 restart_period = 30 [default = 50];

  // Size of the window for the moving average restarts.
  optional int32 restart_running_window_size = 62 [default = 50];

  // In the moving average restart algorithms, a restart is triggered if the
  // window average times this ratio is greater that the global average.
  optional double restart_dl_average_ratio = 63 [default = 1.0];
  optional double restart_lbd_average_ratio = 71 [default = 1.0];

  // Block a moving restart algorithm if the trail size of the current conflict
  // is greater than the multiplier times the moving average of the trail size
  // at the previous conflicts.
  optional bool use_blocking_restart = 64 [default = false];
  optional int32 blocking_restart_window_size = 65 [default = 5000];
  optional double blocking_restart_multiplier = 66 [default = 1.4];

  // After each restart, if the number of conflict since the last strategy
  // change is greater that this, then we increment a "strategy_counter" that
  // can be use to change the search strategy used by the following restarts.
  optional int32 num_conflicts_before_strategy_changes = 68 [default = 0];

  // The parameter num_conflicts_before_strategy_changes is increased by that
  // much after each strategy change.
  optional double strategy_change_increase_ratio = 69 [default = 0.0];

  // ==========================================================================
  // Limits
  // ==========================================================================

  // Maximum time allowed in seconds to solve a problem.
  // The counter will starts at the beginning of the Solve() call.
  optional double max_time_in_seconds = 36 [default = inf];

  // Maximum time allowed in deterministic time to solve a problem.
  // The deterministic time should be correlated with the real time used by the
  // solver, the time unit being as close as possible to a second.
  optional double max_deterministic_time = 67 [default = inf];

  // Maximum number of conflicts allowed to solve a problem.
  //
  // TODO(user): Maybe change the way the conflict limit is enforced?
  // currently it is enforced on each independent internal SAT solve, rather
  // than on the overall number of conflicts across all solves. So in the
  // context of an optimization problem, this is not really usable directly by a
  // client.
  optional int64 max_number_of_conflicts = 37
      [default = 0x7FFFFFFFFFFFFFFF];  // kint64max

  // Maximum memory allowed for the whole thread containing the solver. The
  // solver will abort as soon as it detects that this limit is crossed. As a
  // result, this limit is approximative, but usually the solver will not go too
  // much over.
  //
  // TODO(user): This is only used by the pure SAT solver, generalize to CP-SAT.
  optional int64 max_memory_in_mb = 40 [default = 10000];

  // Stop the search when the gap between the best feasible objective (O) and
  // our best objective bound (B) is smaller than a limit.
  // The exact definition is:
  // - Absolute: abs(O - B)
  // - Relative: abs(O - B) / max(1, abs(O)).
  //
  // Important: The relative gap depends on the objective offset! If you
  // artificially shift the objective, you will get widely different value of
  // the relative gap.
  //
  // Note that if the gap is reached, the search status will be OPTIMAL. But
  // one can check the best objective bound to see the actual gap.
  //
  // If the objective is integer, then any absolute gap < 1 will lead to a true
  // optimal. If the objective is floating point, a gap of zero make little
  // sense so is is why we use a non-zero default value. At the end of the
  // search, we will display a warning if OPTIMAL is reported yet the gap is
  // greater than this absolute gap.
  optional double absolute_gap_limit = 159 [default = 1e-4];
  optional double relative_gap_limit = 160 [default = 0.0];

  // ==========================================================================
  // Other parameters
  // ==========================================================================

  // At the beginning of each solve, the random number generator used in some
  // part of the solver is reinitialized to this seed. If you change the random
  // seed, the solver may make different choices during the solving process.
  //
  // For some problems, the running time may vary a lot depending on small
  // change in the solving algorithm. Running the solver with different seeds
  // enables to have more robust benchmarks when evaluating new features.
  optional int32 random_seed = 31 [default = 1];

  // This is mainly here to test the solver variability. Note that in tests, if
  // not explicitly set to false, all 3 options will be set to true so that
  // clients do not rely on the solver returning a specific solution if they are
  // many equivalent optimal solutions.
  optional bool permute_variable_randomly = 178 [default = false];
  optional bool permute_presolve_constraint_order = 179 [default = false];
  optional bool use_absl_random = 180 [default = false];

  // Whether the solver should log the search progress. This is the maing
  // logging parameter and if this is false, none of the logging (callbacks,
  // log_to_stdout, log_to_response, ...) will do anything.
  optional bool log_search_progress = 41 [default = false];

  // Whether the solver should display per sub-solver search statistics.
  // This is only useful is log_search_progress is set to true, and if the
  // number of search workers is > 1. Note that in all case we display a bit
  // of stats with one line per subsolver.
  optional bool log_subsolver_statistics = 189 [default = false];

  // Add a prefix to all logs.
  optional string log_prefix = 185 [default = ""];

  // Log to stdout.
  optional bool log_to_stdout = 186 [default = true];

  // Log to response proto.
  optional bool log_to_response = 187 [default = false];

  // Whether to use pseudo-Boolean resolution to analyze a conflict. Note that
  // this option only make sense if your problem is modelized using
  // pseudo-Boolean constraints. If you only have clauses, this shouldn't change
  // anything (except slow the solver down).
  optional bool use_pb_resolution = 43 [default = false];

  // A different algorithm during PB resolution. It minimizes the number of
  // calls to ReduceCoefficients() which can be time consuming. However, the
  // search space will be different and if the coefficients are large, this may
  // lead to integer overflows that could otherwise be prevented.
  optional bool minimize_reduction_during_pb_resolution = 48 [default = false];

  // Whether or not the assumption levels are taken into account during the LBD
  // computation. According to the reference below, not counting them improves
  // the solver in some situation. Note that this only impact solves under
  // assumptions.
  //
  // Gilles Audemard, Jean-Marie Lagniez, Laurent Simon, "Improving Glucose for
  // Incremental SAT Solving with Assumptions: Application to MUS Extraction"
  // Theory and Applications of Satisfiability Testing - SAT 2013, Lecture Notes
  // in Computer Science Volume 7962, 2013, pp 309-317.
  optional bool count_assumption_levels_in_lbd = 49 [default = true];

  // ==========================================================================
  // Presolve
  // ==========================================================================

  // During presolve, only try to perform the bounded variable elimination (BVE)
  // of a variable x if the number of occurrences of x times the number of
  // occurrences of not(x) is not greater than this parameter.
  optional int32 presolve_bve_threshold = 54 [default = 500];

  // During presolve, we apply BVE only if this weight times the number of
  // clauses plus the number of clause literals is not increased.
  optional int32 presolve_bve_clause_weight = 55 [default = 3];

  // The maximum "deterministic" time limit to spend in probing. A value of
  // zero will disable the probing.
  //
  // TODO(user): Clean up. The first one is used in CP-SAT, the other in pure
  // SAT presolve.
  optional double probing_deterministic_time_limit = 226 [default = 1.0];
  optional double presolve_probing_deterministic_time_limit = 57
      [default = 30.0];

  // Whether we use an heuristic to detect some basic case of blocked clause
  // in the SAT presolve.
  optional bool presolve_blocked_clause = 88 [default = true];

  // Whether or not we use Bounded Variable Addition (BVA) in the presolve.
  optional bool presolve_use_bva = 72 [default = true];

  // Apply Bounded Variable Addition (BVA) if the number of clauses is reduced
  // by stricly more than this threshold. The algorithm described in the paper
  // uses 0, but quick experiments showed that 1 is a good value. It may not be
  // worth it to add a new variable just to remove one clause.
  optional int32 presolve_bva_threshold = 73 [default = 1];

  // In case of large reduction in a presolve iteration, we perform multiple
  // presolve iterations. This parameter controls the maximum number of such
  // presolve iterations.
  optional int32 max_presolve_iterations = 138 [default = 3];

  // Whether we presolve the cp_model before solving it.
  optional bool cp_model_presolve = 86 [default = true];

  // How much effort do we spend on probing. 0 disables it completely.
  optional int32 cp_model_probing_level = 110 [default = 2];

  // Whether we also use the sat presolve when cp_model_presolve is true.
  optional bool cp_model_use_sat_presolve = 93 [default = true];

  // If true, we detect variable that are unique to a table constraint and only
  // there to encode a cost on each tuple. This is usually the case when a WCSP
  // (weighted constraint program) is encoded into CP-SAT format.
  //
  // This can lead to a dramatic speed-up for such problems but is still
  // experimental at this point.
  optional bool detect_table_with_cost = 216 [default = false];

  // How much we try to "compress" a table constraint. Compressing more leads to
  // less Booleans and faster propagation but can reduced the quality of the lp
  // relaxation. Values goes from 0 to 3 where we always try to fully compress a
  // table. At 2, we try to automatically decide if it is worth it.
  optional int32 table_compression_level = 217 [default = 2];

  // If true, expand all_different constraints that are not permutations.
  // Permutations (#Variables = #Values) are always expanded.
  optional bool expand_alldiff_constraints = 170 [default = false];

  // If true, expand the reservoir constraints by creating booleans for all
  // possible precedences between event and encoding the constraint.
  optional bool expand_reservoir_constraints = 182 [default = true];

  // If the number of expressions in the lin_max is less that the max size
  // parameter, model expansion replaces target = max(xi) by linear constraint
  // with the introduction of new booleans bi such that bi => target == xi.
  //
  // This is mainly for experimenting compared to a custom lin_max propagator.
  optional int32 max_lin_max_size_for_expansion = 280 [default = 0];

  // If true, it disable all constraint expansion.
  // This should only be used to test the presolve of expanded constraints.
  optional bool disable_constraint_expansion = 181 [default = false];

  // Linear constraint with a complex right hand side (more than a single
  // interval) need to be expanded, there is a couple of way to do that.
  optional bool encode_complex_linear_constraint_with_integer = 223
      [default = false];

  // During presolve, we use a maximum clique heuristic to merge together
  // no-overlap constraints or at most one constraints. This code can be slow,
  // so we have a limit in place on the number of explored nodes in the
  // underlying graph. The internal limit is an int64, but we use double here to
  // simplify manual input.
  optional double merge_no_overlap_work_limit = 145 [default = 1e12];
  optional double merge_at_most_one_work_limit = 146 [default = 1e8];

  // How much substitution (also called free variable aggregation in MIP
  // litterature) should we perform at presolve. This currently only concerns
  // variable appearing only in linear constraints. For now the value 0 turns it
  // off and any positive value performs substitution.
  optional int32 presolve_substitution_level = 147 [default = 1];

  // If true, we will extract from linear constraints, enforcement literals of
  // the form "integer variable at bound => simplified constraint". This should
  // always be beneficial except that we don't always handle them as efficiently
  // as we could for now. This causes problem on manna81.mps (LP relaxation not
  // as tight it seems) and on neos-3354841-apure.mps.gz (too many literals
  // created this way).
  optional bool presolve_extract_integer_enforcement = 174 [default = false];

  // A few presolve operations involve detecting constraints included in other
  // constraint. Since there can be a quadratic number of such pairs, and
  // processing them usually involve scanning them, the complexity of these
  // operations can be big. This enforce a local deterministic limit on the
  // number of entries scanned. Default is 1e8.
  //
  // A value of zero will disable these presolve rules completely.
  optional int64 presolve_inclusion_work_limit = 201 [default = 100000000];

  // If true, we don't keep names in our internal copy of the user given model.
  optional bool ignore_names = 202 [default = true];

  // Run a max-clique code amongst all the x != y we can find and try to infer
  // set of variables that are all different. This allows to close neos16.mps
  // for instance. Note that we only run this code if there is no all_diff
  // already in the model so that if a user want to add some all_diff, we assume
  // it is well done and do not try to add more.
  //
  // This will also detect and add no_overlap constraints, if all the relations
  // x != y have "offsets" between them. I.e. x > y + offset.
  optional bool infer_all_diffs = 233 [default = true];

  // Try to find large "rectangle" in the linear constraint matrix with
  // identical lines. If such rectangle is big enough, we can introduce a new
  // integer variable corresponding to the common expression and greatly reduce
  // the number of non-zero.
  optional bool find_big_linear_overlap = 234 [default = true];

  // ==========================================================================
  // Inprocessing
  // ==========================================================================

  // Enable or disable "inprocessing" which is some SAT presolving done at
  // each restart to the root level.
  optional bool use_sat_inprocessing = 163 [default = true];

  // Proportion of deterministic time we should spend on inprocessing.
  // At each "restart", if the proportion is below this ratio, we will do some
  // inprocessing, otherwise, we skip it for this restart.
  optional double inprocessing_dtime_ratio = 273 [default = 0.2];

  // The amount of dtime we should spend on probing for each inprocessing round.
  optional double inprocessing_probing_dtime = 274 [default = 1.0];

  // Parameters for an heuristic similar to the one described in "An effective
  // learnt clause minimization approach for CDCL Sat Solvers",
  // https://www.ijcai.org/proceedings/2017/0098.pdf
  //
  // This is the amount of dtime we should spend on this technique during each
  // inprocessing phase.
  //
  // The minimization technique is the same as the one used to minimize core in
  // max-sat. We also minimize problem clauses and not just the learned clause
  // that we keep forever like in the paper.
  optional double inprocessing_minimization_dtime = 275 [default = 1.0];

  // ==========================================================================
  // Multithread
  // ==========================================================================

  // Specify the number of parallel workers (i.e. threads) to use during search.
  // This should usually be lower than your number of available cpus +
  // hyperthread in your machine.
  //
  // A value of 0 means the solver will try to use all cores on the machine.
  // A number of 1 means no parallelism.
  //
  // Note that 'num_workers' is the preferred name, but if it is set to zero,
  // we will still read the deprecated 'num_search_worker'.
  //
  // As of 2020-04-10, if you're using SAT via MPSolver (to solve integer
  // programs) this field is overridden with a value of 8, if the field is not
  // set *explicitly*. Thus, always set this field explicitly or via
  // MPSolver::SetNumThreads().
  optional int32 num_workers = 206 [default = 0];
  optional int32 num_search_workers = 100 [default = 0];

  // Obsolete parameter. No-op.
  optional int32 min_num_lns_workers = 211 [default = 2];

  // In multi-thread, the solver can be mainly seen as a portfolio of solvers
  // with different parameters. This field indicates the names of the parameters
  // that are used in multithread.
  //
  // See cp_model_search.cc to see a list of the names and the default value (if
  // left empty) that looks like:
  // - default_lp           (linearization_level:1)
  // - fixed                (only if fixed search specified or scheduling)
  // - no_lp                (linearization_level:0)
  // - max_lp               (linearization_level:2)
  // - pseudo_costs         (only if objective, change search heuristic)
  // - reduced_costs        (only if objective, change search heuristic)
  // - quick_restart        (kind of probing)
  // - quick_restart_no_lp  (kind of probing with linearization_level:0)
  // - lb_tree_search       (to improve lower bound, MIP like tree search)
  // - probing              (continuous probing and shaving)
  //
  // Also, note that some set of parameters will be ignored if they do not make
  // sense. For instance if there is no objective, pseudo_cost or reduced_cost
  // search will be ignored. Core based search will only work if the objective
  // has many terms. If there is no fixed strategy fixed will be ignored. And so
  // on.
  //
  // The order is important, as only the first usable "num_workers -
  // min_num_lns_workers" subsolvers will be scheduled. You can see in the log
  // which one are selected for a given run. All the others will be LNS if there
  // is an objective, or randomized SAT search for pure satisfiability problems.
  repeated string subsolvers = 207;

  // A convenient way to add more workers types.
  // These will be added at the beginning of the list.
  repeated string extra_subsolvers = 219;

  // Rather than fully specifying subsolvers, it is often convenient to just
  // remove the ones that are not useful on a given problem.
  repeated string ignore_subsolvers = 209;

  // It is possible to specify additional subsolver configuration. These can be
  // referred by their params.name() in the fields above. Note that only the
  // specified field will "overwrite" the ones of the base parameter. If a
  // subsolver_params has the name of an existing subsolver configuration, the
  // named parameters will be merged into the subsolver configuration.
  repeated SatParameters subsolver_params = 210;

  // TODO(user): Also define like for subsolvers the list of "active"
  // type of neighborhood used.

  // Experimental. If this is true, then we interleave all our major search
  // strategy and distribute the work amongst num_workers.
  //
  // The search is deterministic (independently of num_workers!), and we
  // schedule and wait for interleave_batch_size task to be completed before
  // synchronizing and scheduling the next batch of tasks.
  optional bool interleave_search = 136 [default = false];
  optional int32 interleave_batch_size = 134 [default = 0];

  // Allows objective sharing between workers.
  optional bool share_objective_bounds = 113 [default = true];

  // Allows sharing of the bounds of modified variables at level 0.
  optional bool share_level_zero_bounds = 114 [default = true];

  // Allows sharing of new learned binary clause between workers.
  optional bool share_binary_clauses = 203 [default = true];

  // ==========================================================================
  // Debugging parameters
  // ==========================================================================

  // We have two different postsolve code. The default one should be better and
  // it allows for a more powerful presolve, but it can be useful to postsolve
  // using the full solver instead.
  optional bool debug_postsolve_with_full_solver = 162 [default = false];

  // If positive, try to stop just after that many presolve rules have been
  // applied. This is mainly useful for debugging presolve.
  optional int32 debug_max_num_presolve_operations = 151 [default = 0];

  // Crash if we do not manage to complete the hint into a full solution.
  optional bool debug_crash_on_bad_hint = 195 [default = false];

  // ==========================================================================
  // Max-sat parameters
  // ==========================================================================

  // For an optimization problem, whether we follow some hints in order to find
  // a better first solution. For a variable with hint, the solver will always
  // try to follow the hint. It will revert to the variable_branching default
  // otherwise.
  optional bool use_optimization_hints = 35 [default = true];

  // If positive, we spend some effort on each core:
  // - At level 1, we use a simple heuristic to try to minimize an UNSAT core.
  // - At level 2, we use propagation to minimize the core but also identify
  //   literal in at most one relationship in this core.
  optional int32 core_minimization_level = 50 [default = 2];

  // Whether we try to find more independent cores for a given set of
  // assumptions in the core based max-SAT algorithms.
  optional bool find_multiple_cores = 84 [default = true];

  // If true, when the max-sat algo find a core, we compute the minimal number
  // of literals in the core that needs to be true to have a feasible solution.
  // This is also called core exhaustion in more recent max-SAT papers.
  optional bool cover_optimization = 89 [default = true];

  // In what order do we add the assumptions in a core-based max-sat algorithm
  enum MaxSatAssumptionOrder {
    DEFAULT_ASSUMPTION_ORDER = 0;
    ORDER_ASSUMPTION_BY_DEPTH = 1;
    ORDER_ASSUMPTION_BY_WEIGHT = 2;
  }
  optional MaxSatAssumptionOrder max_sat_assumption_order = 51
      [default = DEFAULT_ASSUMPTION_ORDER];

  // If true, adds the assumption in the reverse order of the one defined by
  // max_sat_assumption_order.
  optional bool max_sat_reverse_assumption_order = 52 [default = false];

  // What stratification algorithm we use in the presence of weight.
  enum MaxSatStratificationAlgorithm {
    // No stratification of the problem.
    STRATIFICATION_NONE = 0;

    // Start with literals with the highest weight, and when SAT, add the
    // literals with the next highest weight and so on.
    STRATIFICATION_DESCENT = 1;

    // Start with all literals. Each time a core is found with a given minimum
    // weight, do not consider literals with a lower weight for the next core
    // computation. If the subproblem is SAT, do like in STRATIFICATION_DESCENT
    // and just add the literals with the next highest weight.
    STRATIFICATION_ASCENT = 2;
  }
  optional MaxSatStratificationAlgorithm max_sat_stratification = 53
      [default = STRATIFICATION_DESCENT];

  // ==========================================================================
  // Constraint programming parameters
  // ==========================================================================

  // Some search decisions might cause a really large number of propagations to
  // happen when integer variables with large domains are only reduced by 1 at
  // each step. If we propagate more than the number of variable times this
  // parameters we try to take counter-measure. Setting this to 0.0 disable this
  // feature.
  //
  // TODO(user): Setting this to something like 10 helps in most cases, but the
  // code is currently buggy and can cause the solve to enter a bad state where
  // no progress is made.
  optional double propagation_loop_detection_factor = 221 [default = 10.0];

  // When this is true, then a disjunctive constraint will try to use the
  // precedence relations between time intervals to propagate their bounds
  // further. For instance if task A and B are both before C and task A and B
  // are in disjunction, then we can deduce that task C must start after
  // duration(A) + duration(B) instead of simply max(duration(A), duration(B)),
  // provided that the start time for all task was currently zero.
  //
  // This always result in better propagation, but it is usually slow, so
  // depending on the problem, turning this off may lead to a faster solution.
  optional bool use_precedences_in_disjunctive_constraint = 74 [default = true];

  // Create one literal for each disjunction of two pairs of tasks. This slows
  // down the solve time, but improves the lower bound of the objective in the
  // makespan case. This will be triggered if the number of intervals is less or
  // equal than the parameter and if use_strong_propagation_in_disjunctive is
  // true.
  optional int32 max_size_to_create_precedence_literals_in_disjunctive = 229
      [default = 60];

  // Enable stronger and more expensive propagation on no_overlap constraint.
  optional bool use_strong_propagation_in_disjunctive = 230 [default = false];

  // Whether we try to branch on decision "interval A before interval B" rather
  // than on intervals bounds. This usually works better, but slow down a bit
  // the time to find the first solution.
  //
  // These parameters are still EXPERIMENTAL, the result should be correct, but
  // it some corner cases, they can cause some failing CHECK in the solver.
  optional bool use_dynamic_precedence_in_disjunctive = 263 [default = false];
  optional bool use_dynamic_precedence_in_cumulative = 268 [default = false];

  // When this is true, the cumulative constraint is reinforced with overload
  // checking, i.e., an additional level of reasoning based on energy. This
  // additional level supplements the default level of reasoning as well as
  // timetable edge finding.
  //
  // This always result in better propagation, but it is usually slow, so
  // depending on the problem, turning this off may lead to a faster solution.
  optional bool use_overload_checker_in_cumulative = 78 [default = false];

  // When this is true, the cumulative constraint is reinforced with timetable
  // edge finding, i.e., an additional level of reasoning based on the
  // conjunction of energy and mandatory parts. This additional level
  // supplements the default level of reasoning as well as overload_checker.
  //
  // This always result in better propagation, but it is usually slow, so
  // depending on the problem, turning this off may lead to a faster solution.
  optional bool use_timetable_edge_finding_in_cumulative = 79 [default = false];

  // Max number of intervals for the timetable_edge_finding algorithm to
  // propagate. A value of 0 disables the constraint.
  optional int32 max_num_intervals_for_timetable_edge_finding = 260
      [default = 100];

  // If true, detect and create constraint for integer variable that are "after"
  // a set of intervals in the same cumulative constraint.
  //
  // Experimental: by default we just use "direct" precedences. If
  // exploit_all_precedences is true, we explore the full precedence graph. This
  // assumes we have a DAG otherwise it fails.
  optional bool use_hard_precedences_in_cumulative = 215 [default = false];
  optional bool exploit_all_precedences = 220 [default = false];

  // When this is true, the cumulative constraint is reinforced with propagators
  // from the disjunctive constraint to improve the inference on a set of tasks
  // that are disjunctive at the root of the problem. This additional level
  // supplements the default level of reasoning.
  //
  // Propagators of the cumulative constraint will not be used at all if all the
  // tasks are disjunctive at root node.
  //
  // This always result in better propagation, but it is usually slow, so
  // depending on the problem, turning this off may lead to a faster solution.
  optional bool use_disjunctive_constraint_in_cumulative = 80 [default = true];

  // When this is true, the no_overlap_2d constraint is reinforced with
  // propagators from the cumulative constraints. It consists of ignoring the
  // position of rectangles in one position and projecting the no_overlap_2d on
  // the other dimension to create a cumulative constraint. This is done on both
  // axis. This additional level supplements the default level of reasoning.
  optional bool use_timetabling_in_no_overlap_2d = 200 [default = false];

  // When this is true, the no_overlap_2d constraint is reinforced with
  // energetic reasoning. This additional level supplements the default level of
  // reasoning.
  optional bool use_energetic_reasoning_in_no_overlap_2d = 213
      [default = false];

  // When this is true, the no_overlap_2d constraint is reinforced with
  // an energetic reasoning that uses an area-based energy. This can be combined
  // with the two other overlap heuristics above.
  optional bool use_area_energetic_reasoning_in_no_overlap_2d = 271
      [default = false];

  // If the number of pairs to look is below this threshold, do an extra step of
  // propagation in the no_overlap_2d constraint by looking at all pairs of
  // intervals.
  optional int32 max_pairs_pairwise_reasoning_in_no_overlap_2d = 276
      [default = 1250];

  // When set, it activates a few scheduling parameters to improve the lower
  // bound of scheduling problems. This is only effective with multiple workers
  // as it modifies the reduced_cost, lb_tree_search, and probing workers.
  optional bool use_dual_scheduling_heuristics = 214 [default = true];

  // The search branching will be used to decide how to branch on unfixed nodes.
  enum SearchBranching {
    // Try to fix all literals using the underlying SAT solver's heuristics,
    // then generate and fix literals until integer variables are fixed. New
    // literals on integer variables are generated using the fixed search
    // specified by the user or our default one.
    AUTOMATIC_SEARCH = 0;

    // If used then all decisions taken by the solver are made using a fixed
    // order as specified in the API or in the CpModelProto search_strategy
    // field.
    FIXED_SEARCH = 1;

    // Simple portfolio search used by LNS workers.
    PORTFOLIO_SEARCH = 2;

    // If used, the solver will use heuristics from the LP relaxation. This
    // exploit the reduced costs of the variables in the relaxation.
    LP_SEARCH = 3;

    // If used, the solver uses the pseudo costs for branching. Pseudo costs
    // are computed using the historical change in objective bounds when some
    // decision are taken. Note that this works whether we use an LP or not.
    PSEUDO_COST_SEARCH = 4;

    // Mainly exposed here for testing. This quickly tries a lot of randomized
    // heuristics with a low conflict limit. It usually provides a good first
    // solution.
    PORTFOLIO_WITH_QUICK_RESTART_SEARCH = 5;

    // Mainly used internally. This is like FIXED_SEARCH, except we follow the
    // solution_hint field of the CpModelProto rather than using the information
    // provided in the search_strategy.
    HINT_SEARCH = 6;

    // Similar to FIXED_SEARCH, but differ in how the variable not listed into
    // the fixed search heuristics are branched on. This will always start the
    // search tree according to the specified fixed search strategy, but will
    // complete it using the default automatic search.
    PARTIAL_FIXED_SEARCH = 7;

    // Randomized search. Used to increase entropy in the search.
    RANDOMIZED_SEARCH = 8;
  }
  optional SearchBranching search_branching = 82 [default = AUTOMATIC_SEARCH];

  // Conflict limit used in the phase that exploit the solution hint.
  optional int32 hint_conflict_limit = 153 [default = 10];

  // If true, the solver tries to repair the solution given in the hint. This
  // search terminates after the 'hint_conflict_limit' is reached and the solver
  // switches to regular search. If false, then  we do a FIXED_SEARCH using the
  // hint until the hint_conflict_limit is reached.
  optional bool repair_hint = 167 [default = false];

  // If true, variables appearing in the solution hints will be fixed to their
  // hinted value.
  optional bool fix_variables_to_their_hinted_value = 192 [default = false];

  // If true, search will continuously probe Boolean variables, and integer
  // variable bounds. This parameter is set to true in parallel on the probing
  // worker.
  optional bool use_probing_search = 176 [default = false];

  // Use extended probing (probe bool_or, at_most_one, exactly_one).
  optional bool use_extended_probing = 269 [default = true];

  // How many combinations of pairs or triplets of variables we want to scan.
  optional int32 probing_num_combinations_limit = 272 [default = 20000];

  // Add a shaving phase (where the solver tries to prove that the lower or
  // upper bound of a variable are infeasible) to the probing search.
  optional bool use_shaving_in_probing_search = 204 [default = true];

  // Specifies the amount of deterministic time spent of each try at shaving a
  // bound in the shaving search.
  optional double shaving_search_deterministic_time = 205 [default = 0.001];

  // If true, search will search in ascending max objective value (when
  // minimizing) starting from the lower bound of the objective.
  optional bool use_objective_lb_search = 228 [default = false];

  // This search differs from the previous search as it will not use assumptions
  // to bound the objective, and it will recreate a full model with the
  // hardcoded objective value.
  optional bool use_objective_shaving_search = 253 [default = false];

  // The solver ignores the pseudo costs of variables with number of recordings
  // less than this threshold.
  optional int64 pseudo_cost_reliability_threshold = 123 [default = 100];

  // The default optimization method is a simple "linear scan", each time trying
  // to find a better solution than the previous one. If this is true, then we
  // use a core-based approach (like in max-SAT) when we try to increase the
  // lower bound instead.
  optional bool optimize_with_core = 83 [default = false];

  // Do a more conventional tree search (by opposition to SAT based one) where
  // we keep all the explored node in a tree. This is meant to be used in a
  // portfolio and focus on improving the objective lower bound. Keeping the
  // whole tree allow us to report a better objective lower bound coming from
  // the worst open node in the tree.
  optional bool optimize_with_lb_tree_search = 188 [default = false];

  // If non-negative, perform a binary search on the objective variable in order
  // to find an [min, max] interval outside of which the solver proved unsat/sat
  // under this amount of conflict. This can quickly reduce the objective domain
  // on some problems.
  optional int32 binary_search_num_conflicts = 99 [default = -1];

  // This has no effect if optimize_with_core is false. If true, use a different
  // core-based algorithm similar to the max-HS algo for max-SAT. This is a
  // hybrid MIP/CP approach and it uses a MIP solver in addition to the CP/SAT
  // one. This is also related to the PhD work of tobyodavies@
  // "Automatic Logic-Based Benders Decomposition with MiniZinc"
  // http://aaai.org/ocs/index.php/AAAI/AAAI17/paper/view/14489
  optional bool optimize_with_max_hs = 85 [default = false];

  // Parameters for an heuristic similar to the one described in the paper:
  // "Feasibility Jump: an LP-free Lagrangian MIP heuristic", Bjørnar
  // Luteberget, Giorgio Sartor, 2023, Mathematical Programming Computation.
  optional bool use_feasibility_jump = 265 [default = true];

  // Disable every other type of subsolver, setting this turns CP-SAT into a
  // pure local-search solver.
  optional bool test_feasibility_jump = 240 [default = false];

  // On each restart, we randomly choose if we use decay (with this parameter)
  // or no decay.
  optional double feasibility_jump_decay = 242 [default = 0.95];

  // How much do we linearize the problem in the local search code.
  optional int32 feasibility_jump_linearization_level = 257 [default = 2];

  // This is a factor that directly influence the work before each restart.
  // Setting this to zero disable restart, and increasing it lead to longer
  // restarts.
  optional int32 feasibility_jump_restart_factor = 258 [default = 1];

  // Probability for a variable to have a non default value upon restarts or
  // perturbations.
  optional double feasibility_jump_var_randomization_probability = 247
      [default = 0.0];

  // Max distance between the default value and the pertubated value relative to
  // the range of the domain of the variable.
  optional double feasibility_jump_var_perburbation_range_ratio = 248
      [default = 0.2];

  // When stagnating, feasibility jump will either restart from a default
  // solution (with some possible randomization), or randomly pertubate the
  // current solution. This parameter selects the first option.
  optional bool feasibility_jump_enable_restarts = 250 [default = true];

  // Maximum size of no_overlap or no_overlap_2d constraint for a quadratic
  // expansion.
  optional int32 feasibility_jump_max_expanded_constraint_size = 264
      [default = 100];

  // This will create incomplete subsolvers (that are not LNS subsolvers)
  // that use the feasibility jump code to find improving solution, treating
  // the objective improvement as a hard constraint.
  optional int32 num_violation_ls = 244 [default = 0];

  // How long violation_ls should wait before perturbating a solution.
  optional int32 violation_ls_perturbation_period = 249 [default = 100];

  // Probability of using compound move search each restart.
  // TODO(user): Add reference to paper when published.
  optional double violation_ls_compound_move_probability = 259 [default = 0.5];

  // Enables experimental workstealing-like shared tree search.
  // If non-zero, start this many complete worker threads to explore a shared
  // search tree. These workers communicate objective bounds and simple decision
  // nogoods relating to the shared prefix of the tree, and will avoid exploring
  // the same subtrees as one another.
  optional int32 shared_tree_num_workers = 235 [default = 0];

  // Set on shared subtree workers. Users should not set this directly.
  optional bool use_shared_tree_search = 236 [default = false];

  // After their assigned prefix, shared tree workers will branch on the
  // objective with this probability. Higher numbers cause the shared tree
  // search to focus on improving the lower bound over finding primal solutions.
  optional double shared_tree_worker_objective_split_probability = 237
      [default = 0.5];

  // Minimum number of restarts before a worker will replace a subtree
  // that looks "bad" based on the average LBD of learned clauses.
  optional int32 shared_tree_worker_min_restarts_per_subtree = 282
      [default = 32];

  // How many open leaf nodes should the shared tree maintain per worker.
  optional double shared_tree_open_leaves_per_worker = 281 [default = 2.0];

  // In order to limit total shared memory and communication overhead, limit the
  // total number of nodes that may be generated in the shared tree. If the
  // shared tree runs out of unassigned leaves, workers act as portfolio
  // workers. Note: this limit includes interior nodes, not just leaves.
  optional int32 shared_tree_max_nodes_per_worker = 238 [default = 128];

  enum SharedTreeSplitStrategy {
    // Uses the default strategy, currently equivalent to
    // SPLIT_STRATEGY_DISCREPANCY.
    SPLIT_STRATEGY_AUTO = 0;
    // Only accept splits if the node to be split's depth+discrepancy is minimal
    // for the desired number of leaves.
    // The preferred child for discrepancy calculation is the one with the
    // lowest objective lower bound or the original branch direction if the
    // bounds are equal. This rule allows twice as many workers to work in the
    // preferred subtree as non-preferred.
    SPLIT_STRATEGY_DISCREPANCY = 1;
    // Only split nodes with an objective lb equal to the global lb. If there is
    // no objective, this is equivalent to SPLIT_STRATEGY_FIRST_PROPOSAL.
    SPLIT_STRATEGY_OBJECTIVE_LB = 2;
    // Attempt to keep the shared tree balanced.
    SPLIT_STRATEGY_BALANCED_TREE = 3;
    // Workers race to split their subtree, the winner's proposal is accepted.
    SPLIT_STRATEGY_FIRST_PROPOSAL = 4;
  }
  optional SharedTreeSplitStrategy shared_tree_split_strategy = 239
      [default = SPLIT_STRATEGY_AUTO];

  // Whether we enumerate all solutions of a problem without objective. Note
  // that setting this to true automatically disable some presolve reduction
  // that can remove feasible solution. That is it has the same effect as
  // setting keep_all_feasible_solutions_in_presolve.
  //
  // TODO(user): Do not do that and let the user choose what behavior is best by
  // setting keep_all_feasible_solutions_in_presolve ?
  optional bool enumerate_all_solutions = 87 [default = false];

  // If true, we disable the presolve reductions that remove feasible solutions
  // from the search space. Such solution are usually dominated by a "better"
  // solution that is kept, but depending on the situation, we might want to
  // keep all solutions.
  //
  // A trivial example is when a variable is unused. If this is true, then the
  // presolve will not fix it to an arbitrary value and it will stay in the
  // search space.
  optional bool keep_all_feasible_solutions_in_presolve = 173 [default = false];

  // If true, add information about the derived variable domains to the
  // CpSolverResponse. It is an option because it makes the response slighly
  // bigger and there is a bit more work involved during the postsolve to
  // construct it, but it should still have a low overhead. See the
  // tightened_variables field in CpSolverResponse for more details.
  optional bool fill_tightened_domains_in_response = 132 [default = false];

  // If true, the final response addition_solutions field will be filled with
  // all solutions from our solutions pool.
  //
  // Note that if both this field and enumerate_all_solutions is true, we will
  // copy to the pool all of the solution found. So if solution_pool_size is big
  // enough, you can get all solutions this way instead of using the solution
  // callback.
  //
  // Note that this only affect the "final" solution, not the one passed to the
  // solution callbacks.
  optional bool fill_additional_solutions_in_response = 194 [default = false];

  // If true, the solver will add a default integer branching strategy to the
  // already defined search strategy. If not, some variable might still not be
  // fixed at the end of the search. For now we assume these variable can just
  // be set to their lower bound.
  optional bool instantiate_all_variables = 106 [default = true];

  // If true, then the precedences propagator try to detect for each variable if
  // it has a set of "optional incoming arc" for which at least one of them is
  // present. This is usually useful to have but can be slow on model with a lot
  // of precedence.
  optional bool auto_detect_greater_than_at_least_one_of = 95 [default = true];

  // For an optimization problem, stop the solver as soon as we have a solution.
  optional bool stop_after_first_solution = 98 [default = false];

  // Mainly used when improving the presolver. When true, stops the solver after
  // the presolve is complete (or after loading and root level propagation).
  optional bool stop_after_presolve = 149 [default = false];
  optional bool stop_after_root_propagation = 252 [default = false];

  // LNS parameters.

  // Testing parameters used to disable all lns workers.
  optional bool use_lns = 283 [default = true];

  // Experimental parameters to disable everything but lns.
  optional bool use_lns_only = 101 [default = false];

  // Size of the top-n different solutions kept by the solver.
  // This parameter must be > 0.
  // Currently this only impact the "base" solution chosen for a LNS fragment.
  optional int32 solution_pool_size = 193 [default = 3];

  // Turns on relaxation induced neighborhood generator.
  optional bool use_rins_lns = 129 [default = true];

  // Adds a feasibility pump subsolver along with lns subsolvers.
  optional bool use_feasibility_pump = 164 [default = true];

  // Turns on neighborhood generator based on local branching LP. Based on Huang
  // et al., "Local Branching Relaxation Heuristics for Integer Linear
  // Programs", 2023.
  optional bool use_lb_relax_lns = 255 [default = false];

  // Rounding method to use for feasibility pump.
  enum FPRoundingMethod {
    // Rounds to the nearest integer value.
    NEAREST_INTEGER = 0;

    // Counts the number of linear constraints restricting the variable in the
    // increasing values (up locks) and decreasing values (down locks). Rounds
    // the variable in the direction of lesser locks.
    LOCK_BASED = 1;

    // Similar to lock based rounding except this only considers locks of active
    // constraints from the last lp solve.
    ACTIVE_LOCK_BASED = 3;

    // This is expensive rounding algorithm. We round variables one by one and
    // propagate the bounds in between. If none of the rounded values fall in
    // the continuous domain specified by lower and upper bound, we use the
    // current lower/upper bound (whichever one is closest) instead of rounding
    // the fractional lp solution value. If both the rounded values are in the
    // domain, we round to nearest integer.
    PROPAGATION_ASSISTED = 2;
  }
  optional FPRoundingMethod fp_rounding = 165 [default = PROPAGATION_ASSISTED];

  // If true, registers more lns subsolvers with different parameters.
  optional bool diversify_lns_params = 137 [default = false];

  // Randomize fixed search.
  optional bool randomize_search = 103 [default = false];

  // Search randomization will collect the top
  // 'search_random_variable_pool_size' valued variables, and pick one randomly.
  // The value of the variable is specific to each strategy.
  optional int64 search_random_variable_pool_size = 104 [default = 0];

  // Experimental code: specify if the objective pushes all tasks toward the
  // start of the schedule.
  optional bool push_all_tasks_toward_start = 262 [default = false];

  // If true, we automatically detect variables whose constraint are always
  // enforced by the same literal and we mark them as optional. This allows
  // to propagate them as if they were present in some situation.
  //
  // TODO(user): This is experimental and seems to lead to wrong optimal in
  // some situation. It should however gives correct solutions. Fix.
  optional bool use_optional_variables = 108 [default = false];

  // The solver usually exploit the LP relaxation of a model. If this option is
  // true, then whatever is infered by the LP will be used like an heuristic to
  // compute EXACT propagation on the IP. So with this option, there is no
  // numerical imprecision issues.
  optional bool use_exact_lp_reason = 109 [default = true];

  // This can be beneficial if there is a lot of no-overlap constraints but a
  // relatively low number of different intervals in the problem. Like 1000
  // intervals, but 1M intervals in the no-overlap constraints covering them.
  optional bool use_combined_no_overlap = 133 [default = false];

  // All at_most_one constraints with a size <= param will be replaced by a
  // quadratic number of binary implications.
  optional int32 at_most_one_max_expansion_size = 270 [default = 3];

  // Indicates if the CP-SAT layer should catch Control-C (SIGINT) signals
  // when calling solve. If set, catching the SIGINT signal will terminate the
  // search gracefully, as if a time limit was reached.
  optional bool catch_sigint_signal = 135 [default = true];

  // Stores and exploits "implied-bounds" in the solver. That is, relations of
  // the form literal => (var >= bound). This is currently used to derive
  // stronger cuts.
  optional bool use_implied_bounds = 144 [default = true];

  // Whether we try to do a few degenerate iteration at the end of an LP solve
  // to minimize the fractionality of the integer variable in the basis. This
  // helps on some problems, but not so much on others. It also cost of bit of
  // time to do such polish step.
  optional bool polish_lp_solution = 175 [default = false];

  // The internal LP tolerances used by CP-SAT. These applies to the internal
  // and scaled problem. If the domains of your variables are large it might be
  // good to use lower tolerances. If your problem is binary with low
  // coefficients, it might be good to use higher ones to speed-up the lp
  // solves.
  optional double lp_primal_tolerance = 266 [default = 1e-7];
  optional double lp_dual_tolerance = 267 [default = 1e-7];

  // Temporary flag util the feature is more mature. This convert intervals to
  // the newer proto format that support affine start/var/end instead of just
  // variables.
  optional bool convert_intervals = 177 [default = true];

  // Whether we try to automatically detect the symmetries in a model and
  // exploit them. Currently, at level 1 we detect them in presolve and try
  // to fix Booleans. At level 2, we also do some form of dynamic symmetry
  // breaking during search.
  optional int32 symmetry_level = 183 [default = 2];

  // The new linear propagation code treat all constraints at once and use
  // an adaptation of Bellman-Ford-Tarjan to propagate constraint in a smarter
  // order and potentially detect propagation cycle earlier.
  optional bool new_linear_propagation = 224 [default = true];

  // Linear constraints that are not pseudo-Boolean and that are longer than
  // this size will be split into sqrt(size) intermediate sums in order to have
  // faster propation in the CP engine.
  optional int32 linear_split_size = 256 [default = 100];

  // ==========================================================================
  // Linear programming relaxation
  // ==========================================================================

  // A non-negative level indicating the type of constraints we consider in the
  // LP relaxation. At level zero, no LP relaxation is used. At level 1, only
  // the linear constraint and full encoding are added. At level 2, we also add
  // all the Boolean constraints.
  optional int32 linearization_level = 90 [default = 1];

  // A non-negative level indicating how much we should try to fully encode
  // Integer variables as Boolean.
  optional int32 boolean_encoding_level = 107 [default = 1];

  // When loading a*x + b*y ==/!= c when x and y are both fully encoded.
  // The solver may decide to replace the linear equation by a set of clauses.
  // This is triggered if the sizes of the domains of x and y are below the
  // threshold.
  optional int32 max_domain_size_when_encoding_eq_neq_constraints = 191
      [default = 16];

  // The limit on the number of cuts in our cut pool. When this is reached we do
  // not generate cuts anymore.
  //
  // TODO(user): We should probably remove this parameters, and just always
  // generate cuts but only keep the best n or something.
  optional int32 max_num_cuts = 91 [default = 10000];

  // Control the global cut effort. Zero will turn off all cut. For now we just
  // have one level. Note also that most cuts are only used at linearization
  // level >= 2.
  optional int32 cut_level = 196 [default = 1];

  // For the cut that can be generated at any level, this control if we only
  // try to generate them at the root node.
  optional bool only_add_cuts_at_level_zero = 92 [default = false];

  // When the LP objective is fractional, do we add the cut that forces the
  // linear objective expression to be greater or equal to this fractional value
  // rounded up? We can always do that since our objective is integer, and
  // combined with MIR heuristic to reduce the coefficient of such cut, it can
  // help.
  optional bool add_objective_cut = 197 [default = false];

  // Whether we generate and add Chvatal-Gomory cuts to the LP at root node.
  // Note that for now, this is not heavily tuned.
  optional bool add_cg_cuts = 117 [default = true];

  // Whether we generate MIR cuts at root node.
  // Note that for now, this is not heavily tuned.
  optional bool add_mir_cuts = 120 [default = true];

  // Whether we generate Zero-Half cuts at root node.
  // Note that for now, this is not heavily tuned.
  optional bool add_zero_half_cuts = 169 [default = true];

  // Whether we generate clique cuts from the binary implication graph. Note
  // that as the search goes on, this graph will contains new binary clauses
  // learned by the SAT engine.
  optional bool add_clique_cuts = 172 [default = true];

  // Whether we generate RLT cuts. This is still experimental but can help on
  // binary problem with a lot of clauses of size 3.
  optional bool add_rlt_cuts = 279 [default = true];

  // Cut generator for all diffs can add too many cuts for large all_diff
  // constraints. This parameter restricts the large all_diff constraints to
  // have a cut generator.
  optional int32 max_all_diff_cut_size = 148 [default = 64];

  // For the lin max constraints, generates the cuts described in "Strong
  // mixed-integer programming formulations for trained neural networks" by Ross
  // Anderson et. (https://arxiv.org/pdf/1811.01988.pdf)
  optional bool add_lin_max_cuts = 152 [default = true];

  // In the integer rounding procedure used for MIR and Gomory cut, the maximum
  // "scaling" we use (must be positive). The lower this is, the lower the
  // integer coefficients of the cut will be. Note that cut generated by lower
  // values are not necessarily worse than cut generated by larger value. There
  // is no strict dominance relationship.
  //
  // Setting this to 2 result in the "strong fractional rouding" of Letchford
  // and Lodi.
  optional int32 max_integer_rounding_scaling = 119 [default = 600];

  // If true, we start by an empty LP, and only add constraints not satisfied
  // by the current LP solution batch by batch. A constraint that is only added
  // like this is known as a "lazy" constraint in the literature, except that we
  // currently consider all constraints as lazy here.
  optional bool add_lp_constraints_lazily = 112 [default = true];

  // Even at the root node, we do not want to spend too much time on the LP if
  // it is "difficult". So we solve it in "chunks" of that many iterations. The
  // solve will be continued down in the tree or the next time we go back to the
  // root node.
  optional int32 root_lp_iterations = 227 [default = 2000];

  // While adding constraints, skip the constraints which have orthogonality
  // less than 'min_orthogonality_for_lp_constraints' with already added
  // constraints during current call. Orthogonality is defined as 1 -
  // cosine(vector angle between constraints). A value of zero disable this
  // feature.
  optional double min_orthogonality_for_lp_constraints = 115 [default = 0.05];

  // Max number of time we perform cut generation and resolve the LP at level 0.
  optional int32 max_cut_rounds_at_level_zero = 154 [default = 1];

  // If a constraint/cut in LP is not active for that many consecutive OPTIMAL
  // solves, remove it from the LP. Note that it might be added again later if
  // it become violated by the current LP solution.
  optional int32 max_consecutive_inactive_count = 121 [default = 100];

  // These parameters are similar to sat clause management activity parameters.
  // They are effective only if the number of generated cuts exceed the storage
  // limit. Default values are based on a few experiments on miplib instances.
  optional double cut_max_active_count_value = 155 [default = 1e10];
  optional double cut_active_count_decay = 156 [default = 0.8];

  // Target number of constraints to remove during cleanup.
  optional int32 cut_cleanup_target = 157 [default = 1000];

  // Add that many lazy constraints (or cuts) at once in the LP. Note that at
  // the beginning of the solve, we do add more than this.
  optional int32 new_constraints_batch_size = 122 [default = 50];

  // All the "exploit_*" parameters below work in the same way: when branching
  // on an IntegerVariable, these parameters affect the value the variable is
  // branched on. Currently the first heuristic that triggers win in the order
  // in which they appear below.
  //
  // TODO(user): Maybe do like for the restart algorithm, introduce an enum
  // and a repeated field that control the order on which these are applied?

  // If true and the Lp relaxation of the problem has an integer optimal
  // solution, try to exploit it. Note that since the LP relaxation may not
  // contain all the constraints, such a solution is not necessarily a solution
  // of the full problem.
  optional bool exploit_integer_lp_solution = 94 [default = true];

  // If true and the Lp relaxation of the problem has a solution, try to exploit
  // it. This is same as above except in this case the lp solution might not be
  // an integer solution.
  optional bool exploit_all_lp_solution = 116 [default = true];

  // When branching on a variable, follow the last best solution value.
  optional bool exploit_best_solution = 130 [default = false];

  // When branching on a variable, follow the last best relaxation solution
  // value. We use the relaxation with the tightest bound on the objective as
  // the best relaxation solution.
  optional bool exploit_relaxation_solution = 161 [default = false];

  // When branching an a variable that directly affect the objective,
  // branch on the value that lead to the best objective first.
  optional bool exploit_objective = 131 [default = true];

  // Infer products of Boolean or of Boolean time IntegerVariable from the
  // linear constrainst in the problem. This can be used in some cuts, altough
  // for now we don't really exploit it.
  optional bool detect_linearized_product = 277 [default = false];

  // ==========================================================================
  // MIP -> CP-SAT (i.e. IP with integer coeff) conversion parameters that are
  // used by our automatic "scaling" algorithm.
  //
  // Note that it is hard to do a meaningful conversion automatically and if
  // you have a model with continuous variables, it is best if you scale the
  // domain of the variable yourself so that you have a relevant precision for
  // the application at hand. Same for the coefficients and constraint bounds.
  // ==========================================================================

  // We need to bound the maximum magnitude of the variables for CP-SAT, and
  // that is the bound we use. If the MIP model expect larger variable value in
  // the solution, then the converted model will likely not be relevant.
  optional double mip_max_bound = 124 [default = 1e7];

  // All continuous variable of the problem will be multiplied by this factor.
  // By default, we don't do any variable scaling and rely on the MIP model to
  // specify continuous variable domain with the wanted precision.
  optional double mip_var_scaling = 125 [default = 1.0];

  // If this is false, then mip_var_scaling is only applied to variables with
  // "small" domain. If it is true, we scale all floating point variable
  // independenlty of their domain.
  optional bool mip_scale_large_domain = 225 [default = false];

  // If true, some continuous variable might be automatically scaled. For now,
  // this is only the case where we detect that a variable is actually an
  // integer multiple of a constant. For instance, variables of the form k * 0.5
  // are quite frequent, and if we detect this, we will scale such variable
  // domain by 2 to make it implied integer.
  optional bool mip_automatically_scale_variables = 166 [default = true];

  // If one try to solve a MIP model with CP-SAT, because we assume all variable
  // to be integer after scaling, we will not necessarily have the correct
  // optimal. Note however that all feasible solutions are valid since we will
  // just solve a more restricted version of the original problem.
  //
  // This parameters is here to prevent user to think the solution is optimal
  // when it might not be. One will need to manually set this to false to solve
  // a MIP model where the optimal might be different.
  //
  // Note that this is tested after some MIP presolve steps, so even if not
  // all original variable are integer, we might end up with a pure IP after
  // presolve and after implied integer detection.
  optional bool only_solve_ip = 222 [default = false];

  // When scaling constraint with double coefficients to integer coefficients,
  // we will multiply by a power of 2 and round the coefficients. We will choose
  // the lowest power such that we have no potential overflow (see
  // mip_max_activity_exponent) and the worst case constraint activity error
  // does not exceed this threshold.
  //
  // Note that we also detect constraint with rational coefficients and scale
  // them accordingly when it seems better instead of using a power of 2.
  //
  // We also relax all constraint bounds by this absolute value. For pure
  // integer constraint, if this value if lower than one, this will not change
  // anything. However it is needed when scaling MIP problems.
  //
  // If we manage to scale a constraint correctly, the maximum error we can make
  // will be twice this value (once for the scaling error and once for the
  // relaxed bounds). If we are not able to scale that well, we will display
  // that fact but still scale as best as we can.
  optional double mip_wanted_precision = 126 [default = 1e-6];

  // To avoid integer overflow, we always force the maximum possible constraint
  // activity (and objective value) according to the initial variable domain to
  // be smaller than 2 to this given power. Because of this, we cannot always
  // reach the "mip_wanted_precision" parameter above.
  //
  // This can go as high as 62, but some internal algo currently abort early if
  // they might run into integer overflow, so it is better to keep it a bit
  // lower than this.
  optional int32 mip_max_activity_exponent = 127 [default = 53];

  // As explained in mip_precision and mip_max_activity_exponent, we cannot
  // always reach the wanted precision during scaling. We use this threshold to
  // enphasize in the logs when the precision seems bad.
  optional double mip_check_precision = 128 [default = 1e-4];

  // Even if we make big error when scaling the objective, we can always derive
  // a correct lower bound on the original objective by using the exact lower
  // bound on the scaled integer version of the objective. This should be fast,
  // but if you don't care about having a precise lower bound, you can turn it
  // off.
  optional bool mip_compute_true_objective_bound = 198 [default = true];

  // Any finite values in the input MIP must be below this threshold, otherwise
  // the model will be reported invalid. This is needed to avoid floating point
  // overflow when evaluating bounds * coeff for instance. We are a bit more
  // defensive, but in practice, users shouldn't use super large values in a
  // MIP.
  optional double mip_max_valid_magnitude = 199 [default = 1e20];

  // By default, any variable/constraint bound with a finite value and a
  // magnitude greater than the mip_max_valid_magnitude will result with a
  // invalid model. This flags change the behavior such that such bounds are
  // silently transformed to +∞ or -∞.
  //
  // It is recommended to keep it at false, and create valid bounds.
  optional bool mip_treat_high_magnitude_bounds_as_infinity = 278
      [default = false];

  // Any value in the input mip with a magnitude lower than this will be set to
  // zero. This is to avoid some issue in LP presolving.
  optional double mip_drop_tolerance = 232 [default = 1e-16];

  // When solving a MIP, we do some basic floating point presolving before
  // scaling the problem to integer to be handled by CP-SAT. This control how
  // much of that presolve we do. It can help to better scale floating point
  // model, but it is not always behaving nicely.
  optional int32 mip_presolve_level = 261 [default = 2];
}