Hard Leetcode, Easy with Constraint Solvers

The author shows that many leetcode-style problems become straightforward when expressed as constraint models and solved with tools like MiniZinc. Examples include coin change, stock trading, 3-number zero-sum, and largest histogram rectangle, all encoded with concise constraints and objectives. While solvers may not match the runtime of bespoke optimal algorithms, they excel in ease of modeling and adaptability when requirements evolve.

Key Points

• Many "hard" leetcode-style problems are naturally optimization or satisfaction problems and are easy to model in a constraint solver.
• MiniZinc models for coin change, stock profit, 3-number sum-to-zero, and largest histogram rectangle show concise formulations using variables, constraints, and objectives.
• Solvers may be slower than hand-tuned optimal algorithms and have unpredictable runtime, reflecting a capability/tractability tradeoff.
• Their key advantage is extensibility: adding new constraints (e.g., multi-trade stock rules and holding limits) is simple compared to rewriting algorithms.
• Using solvers for interview-style problems is both practical and pedagogically valuable, offering opportunities to teach modeling and optimization techniques.

Sentiment

Mixed: many appreciate solvers as powerful real-world tools and would view their use positively if accompanied by solid coding and trade-off reasoning, but a substantial number argue they’re inappropriate for typical interviews, which still expect hand-crafted algorithms, clear complexity analysis, and scalability.

In Agreement

• Constraint solvers (SAT/SMT/CP/MIP) are underutilized, practical tools that can quickly yield correct solutions and are easy to adapt as requirements change.
• Using a solver can be a positive interview signal if the candidate also demonstrates coding ability and can discuss trade-offs; some interviewers would be impressed or even consider it a strong hire signal.
• A pragmatic workflow is to prototype with a solver, then profile and replace with tailored algorithms only if it becomes a bottleneck.
• Constraint modeling broadens problem-solving patterns and can be educational; resources like Håkan Kjellerstrand’s MiniZinc collection are valuable.
• Interviews should emphasize problem decomposition, communication, and iterative refinement with clarifying questions rather than rote ‘trick’ solutions.

Opposed

• In interviews, the goal is to demonstrate algorithmic understanding and implementation; outsourcing to a solver subverts the assessment and will draw scaling or ‘implement the solver’ follow-ups.
• General-purpose solvers can be slow, unscalable (e.g., large variable counts), and lack predictable runtime/space guarantees, making them unsuitable for many cases.
• LeetCode-style questions aim to test cleverness or mastery of known patterns (e.g., dynamic programming); a solver doesn’t show that understanding.
• Using heavyweight solvers for simple tasks can be overengineering and raises concerns about maintainability and introducing baroque solutions.
• Some believe interviews should favor simple, efficient bespoke algorithms (e.g., O(n) stack for histogram) and that using solvers avoids critical performance reasoning.