Fix file loading issues
Browse files- backend/app.py +4 -1
- backend/solver.py +7 -5
backend/app.py
CHANGED
|
@@ -9,6 +9,9 @@ BASE_DIR = os.path.dirname(os.path.abspath(__file__))
|
|
| 9 |
example_path = os.path.join(BASE_DIR, "Example.txt")
|
| 10 |
with open(example_path, "r", encoding="utf-8") as f:
|
| 11 |
DEFAULT_SYS_CONTENT = f.read()
|
|
|
|
|
|
|
|
|
|
| 12 |
|
| 13 |
# Only initialize once
|
| 14 |
app = Flask(__name__, static_folder='static', static_url_path='')
|
|
@@ -49,7 +52,7 @@ def solve():
|
|
| 49 |
if puzzle_index is None or puzzle_text is None or expected_solution is None:
|
| 50 |
return jsonify({"success": False, "error": "Missing puzzle data"}), 400
|
| 51 |
|
| 52 |
-
result = solve_puzzle(puzzle_index, puzzle_text, expected_solution, sys_content)
|
| 53 |
return jsonify({"success": True, "result": result})
|
| 54 |
|
| 55 |
@app.route("/default_sys_content", methods=["GET"])
|
|
|
|
| 9 |
example_path = os.path.join(BASE_DIR, "Example.txt")
|
| 10 |
with open(example_path, "r", encoding="utf-8") as f:
|
| 11 |
DEFAULT_SYS_CONTENT = f.read()
|
| 12 |
+
sat_cnt_path = os.path.join(BASE_DIR, "Sat_cnt.txt")
|
| 13 |
+
with open(sat_cnt_path, "r", encoding="utf-8") as f:
|
| 14 |
+
SAT_CNT_CONTENT = f.read()
|
| 15 |
|
| 16 |
# Only initialize once
|
| 17 |
app = Flask(__name__, static_folder='static', static_url_path='')
|
|
|
|
| 52 |
if puzzle_index is None or puzzle_text is None or expected_solution is None:
|
| 53 |
return jsonify({"success": False, "error": "Missing puzzle data"}), 400
|
| 54 |
|
| 55 |
+
result = solve_puzzle(puzzle_index, puzzle_text, expected_solution, sys_content, SAT_CNT_CONTENT)
|
| 56 |
return jsonify({"success": True, "result": result})
|
| 57 |
|
| 58 |
@app.route("/default_sys_content", methods=["GET"])
|
backend/solver.py
CHANGED
|
@@ -8,7 +8,7 @@ from deep_convert import deep_convert
|
|
| 8 |
from openai import OpenAI
|
| 9 |
from dotenv import load_dotenv
|
| 10 |
|
| 11 |
-
def solve_puzzle(index, puzzle, expected_solution, sys_content):
|
| 12 |
"""
|
| 13 |
使用 sys_content + puzzle 调用模型,生成 Z3 Python 代码,
|
| 14 |
运行并比对输出是否与 expected_solution 相同。
|
|
@@ -146,14 +146,16 @@ def solve_puzzle(index, puzzle, expected_solution, sys_content):
|
|
| 146 |
if satisfied(constraint) == "unsat":
|
| 147 |
problematic_constraints += f"In line {i + 1}, the {cnt_cons}-th constraint: {constraint}. Not satisfied.\n"
|
| 148 |
|
| 149 |
-
with open("Sat_cnt.txt", "r") as f:
|
| 150 |
-
|
|
|
|
|
|
|
| 151 |
cnt_cons = 0
|
| 152 |
problematic_constraints += "\nMultiple solutions judge:\n"
|
| 153 |
code_by_line_experiment = code_to_run.split("\n")[:last_cons_line + 1]
|
| 154 |
code_by_line_experiment.append("\n")
|
| 155 |
def run_result():
|
| 156 |
-
experiment_code = "\n".join(code_by_line_experiment) +
|
| 157 |
sat_checker = experiment_code.strip()
|
| 158 |
result = subprocess.run(
|
| 159 |
[sys.executable, "-c", sat_checker],
|
|
@@ -163,7 +165,7 @@ def solve_puzzle(index, puzzle, expected_solution, sys_content):
|
|
| 163 |
)
|
| 164 |
return result.stdout.strip()
|
| 165 |
res = run_result()
|
| 166 |
-
# print("\n".join(code_by_line_experiment) +
|
| 167 |
# print(res)
|
| 168 |
if not res or res == "error":
|
| 169 |
problematic_constraints += "Unable to judge."
|
|
|
|
| 8 |
from openai import OpenAI
|
| 9 |
from dotenv import load_dotenv
|
| 10 |
|
| 11 |
+
def solve_puzzle(index, puzzle, expected_solution, sys_content, sat_cnt_content):
|
| 12 |
"""
|
| 13 |
使用 sys_content + puzzle 调用模型,生成 Z3 Python 代码,
|
| 14 |
运行并比对输出是否与 expected_solution 相同。
|
|
|
|
| 146 |
if satisfied(constraint) == "unsat":
|
| 147 |
problematic_constraints += f"In line {i + 1}, the {cnt_cons}-th constraint: {constraint}. Not satisfied.\n"
|
| 148 |
|
| 149 |
+
# with open("Sat_cnt.txt", "r") as f:
|
| 150 |
+
# sat_cnt = f.read()
|
| 151 |
+
# Cannot directly load because the server strucutre
|
| 152 |
+
|
| 153 |
cnt_cons = 0
|
| 154 |
problematic_constraints += "\nMultiple solutions judge:\n"
|
| 155 |
code_by_line_experiment = code_to_run.split("\n")[:last_cons_line + 1]
|
| 156 |
code_by_line_experiment.append("\n")
|
| 157 |
def run_result():
|
| 158 |
+
experiment_code = "\n".join(code_by_line_experiment) + sat_cnt_content
|
| 159 |
sat_checker = experiment_code.strip()
|
| 160 |
result = subprocess.run(
|
| 161 |
[sys.executable, "-c", sat_checker],
|
|
|
|
| 165 |
)
|
| 166 |
return result.stdout.strip()
|
| 167 |
res = run_result()
|
| 168 |
+
# print("\n".join(code_by_line_experiment) + sat_cnt_content)
|
| 169 |
# print(res)
|
| 170 |
if not res or res == "error":
|
| 171 |
problematic_constraints += "Unable to judge."
|