Guess the Output — C++

A small “guess the output” training game built from single-file C++ programs and a static Next.js frontend.

This repository is a proof of concept: many problems + explanations were generated by an AI (LLM), so you should always verify correctness (especially around undefined behavior, compiler differences, and subtle language rules).

Live Demo: https://gto.demo.blanpied.fr/

Overview

• Each problem is a single C++ file in problems/src/.
• A problem can:
  • Print some stdout (normal run),
  • Fail to compile (compilation error),
  • Crash/fail at runtime (runtime error),
  • Intentionally demonstrate undefined behavior (UB).
• Players either:
  • Type the exact expected stdout, or
  • Choose a generic outcome (Compilation error / Runtime error / Undefined behavior).
• Expected outputs (or errors) are generated automatically by compiling/running the programs and exporting JSON consumed by the web UI.

Quick start

Prerequisites

• C++ compiler (e.g., g++)
• Python 3
• Node.js + npm (only if you want to run the web UI locally)

Generate expected results (run from repo root)

python3 problems/run_all.py

This generates/updates:

• web/data/problems.generated.json
• web/data/problems.index.json

Build/run a single problem (example)

cd problems
make problem NAME=p0001
make run NAME=p0001

Run the web app locally (optional)

cd web
npm install
npm run dev
# open http://localhost:3000

Notes:

• The frontend is static (no backend) to keep hosting simple.
• Do not hand-edit web/data/*.generated.json—regenerate it via the Python script.

Project structure

• problems/ — C++ problem sources and build utilities
  • problems/src/ — C++ single-file problems (p0001.cpp, ...)
  • problems/Makefile — build & run helpers
  • problems/run_all.py — generator script (produces web/data/*)
  • problems/problems.json — problem metadata (id, title, difficulty, explanation, concepts)
• web/ — Next.js static frontend
  • web/src/ — frontend source
  • web/data/ — generated JSON used by the UI

Creating problems

Rules (game design)

• One problem = one C++ program file.
• Keep programs compact; they may be tricky/non-idiomatic on purpose to test real understanding.
• Outcomes supported by the game:
  • stdout output,
  • compilation error,
  • runtime error,
  • undefined behavior (must be marked in metadata).

Add a new problem

1. Create a new file in problems/src/ (e.g., p0042.cpp).
2. Add a matching entry in problems/problems.json (see schema below).
3. Run:

   python3 problems/run_all.py

4. Open the web app and confirm it shows the new problem + expected result.

Data pipeline (JSON)

Source metadata: problems/problems.json

This file is the “authoritative” list of problems and explanations. It is a JSON array where each object has:

• id (string): "p0001", "p0002", …
• title (string)
• difficulty (number): integer in [1–5]
• concepts (string[]): keyword tags (used for filtering/presets)
• explanation (string): Markdown shown after a correct answer
• stdin (string, optional): fixed stdin content if the program reads input
• UB (bool, optional): true if the problem intentionally triggers undefined behavior

Generated runtime data: web/data/problems.generated.json

After running python3 problems/run_all.py, each problem object is enriched with generated fields (notably the code and the measured outcome), including:

• result.errorType: "no-error" | "compilation-error" | "runtime-error" | "undefined-behavior"
• result.stdout: only when errorType is "no-error"
• result.errorMessage: styled tokens (only when there is an error)

The web UI reads this generated file to display the code, the “expected” outcome, and the explanation together.

Training configuration (in the UI)

The home page offers presets / filters to build a training session. In plain terms:

• Difficulty range (difficultyMin → difficultyMax)
• Attempts per problem (maxAttemptsPerProblem, unlimited when null)
• Concept tags filter (concepts)
• Problems per session (problemsPerSession, endless when null)
• Timers: either per-problem (problemTimer) or whole-session (sessionTimer) (mutually exclusive)
• Ordering: random vs progressive
• Optional “show output difference” to help compare guesses vs expected output

License & contact

This repository is shared under the GNU General Public License v3.0 (GPL-3.0). See LICENSE for details.

Author: Timothée Blanpied — February 2026