Fix issue #4133: FinalizeAllowedVehicles incorrectly excludes vehicles with negative transits

[cyril23]

Fixes:

• OR-Tools v9.9 regression: incorrect negative transit variables (with modulo > capacity) handling in heterogeneous VRP #4133

Problem

FinalizeAllowedVehicles() uses abs(transit) when checking if a vehicle can serve a node, which incorrectly excludes vehicles when dimensions use negative transit values for algorithmic purposes (reload dimensions, load tracking).

Detailed analysis: #4133 (comment)

Solution

Skip capacity checks for non-positive transits, since only positive transits represent actual cargo/capacity requirements:

const int64_t transit = transit_evaluator(node);
if (transit <= 0 || transit <= capacity) continue;

Changes

File: ortools/routing/routing.cc (after merge with main branch - file was moved from ortools/constraint_solver/routing.cc)

1. Pre-processing loop (lines 1414-1421 lines 1413-1420)

for (int callback_index : dimension->class_evaluators_) {
  // Only consider positive transits for capacity checks.
  // Negative transits are used for reload/load tracking and should not
  // trigger vehicle exclusions based on capacity.
  const int64_t transit = UnaryTransitCallbackOrNull(callback_index)(node);
  if (transit > 0) {
    max_node_transit = std::max(max_node_transit, transit);
  }
}

2. Main capacity check (lines 1444-1448 lines 1443-1447)

// Fix for issue #4133: Skip capacity check for negative transits.
// Negative transits are used for reload dimensions and load tracking,
// not actual cargo requirements.
const int64_t transit = transit_evaluator(node);
if (transit <= 0 || transit <= capacity) continue;

Testing

Built and tested with the reproduction script from the issue comment:

Before fix (v9.12.4544):

Dropped customer stops: 13
Tours found: 1

After fix:

Dropped customer stops: 0
Tours found: 4

✅ All customer stops served, all vehicle classes utilized correctly

Backward compatibility: Only affects models using negative transits in unary dimensions. Standard capacity models are unaffected.

edit: Merge Note

This PR has been rebased on the current main branch (v10.0), which includes a routing module reorganization where routing files were moved from ortools/constraint_solver/ to ortools/routing/. The fix has been successfully applied to the new file location.

[google-cla]

Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

View this failed invocation of the CLA check for more information.

For the most up to date status, view the checks section at the bottom of the pull request.

[cyril23]

OK I've done the CLA and resolved the merge conflict. Ready for review.

[lperron]

no, this correct is currently is correct.

cumul var needs to be 0 <= var <= max_value

if abs(transit) > max_value, the node can never be visited by that vehicle.

What the code does is change the variable ordering which leads to different solutions.

But the proposed fix is incorrect.