feat(题目): 运行时错误告诉学生第几行、为什么出错（Python 看回溯，C 看信号），不泄露测试数据
原来学生只看到「运行时错误」四个字：主判题 output: false，Python 回溯没回来； C 的信号值在 info 里，而 info 只给管理员。 - Python 判出 RE 时，拿第一个失败测试点单独重跑一次（内联测试点 + output: true）， 从回溯取行号（扣掉模板前置行）、异常类型和按句式归的类别。主判题仍不开 output， 免得每个点最多 16MB 的输出回到 worker - C / C++ 取失败点的信号和退出码（段错误提示 scanf 漏 &、return 非 0 等） - 写进 statistic_info.runtime_error；不存异常原文（里面常带测试点输入），名字只在 出现在学生代码里时才存——raise Exception(input()) 实测只落下类型 - 前端按类型翻成中文、摆出出错那行并在编辑器标红；比赛提交不诊断（ACM 只报对错） - 没有诊断的 RE 不再说「代码格式写错了」 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
xuyueandClaude Opus 5.5 committed
1 parent
2a4405d2ea
commit
db70ab62e8
7 files changed
+391
-11
No files matched your search
@@ -19,6 +19,11 @@ import { judgeConfigFor } from "./languages"
|
||||
import { isAccepted, JudgeStatus, type JudgeStatusValue } from "./status"
|
||||
import { parseProblemTemplate } from "./template"
|
||||
import { runSqlCase } from "./sql"
|
||||
import {
|
||||
nativeRuntimeError,
|
||||
parsePythonTraceback,
|
||||
type RuntimeErrorInfo,
|
||||
} from "./runtime-diagnosis"
|
||||
import { readInfo } from "../services/test-case"
|
||||
import { readFile } from "node:fs/promises"
|
||||
import { resolve as resolvePath } from "node:path"
|
||||
@@ -63,12 +68,18 @@ function astRulesForLanguage(value: unknown, language: string): AstRule[] {
|
||||
})
|
||||
}
|
||||
|
||||
/**
|
||||
* `testCase` 是测试点目录名（正常判题），或者内联的测试点（诊断重跑，见
|
||||
* diagnoseRuntimeError）。`output` 只在诊断时打开：它让判题机把每个测试点的
|
||||
* 程序输出原样带回来，正常判题开着的话，死循环打印能把 worker 内存撑爆。
|
||||
*/
|
||||
async function requestJudge(
|
||||
language: string,
|
||||
code: string,
|
||||
timeLimit: number,
|
||||
memoryLimit: number,
|
||||
testCaseId: string,
|
||||
testCase: string | { input: string; output: string }[],
|
||||
output = false,
|
||||
) {
|
||||
const languageConfig = judgeConfigFor(language)
|
||||
if (!languageConfig) throw new Error(`Unsupported judge language: ${language}`)
|
||||
@@ -85,8 +96,8 @@ async function requestJudge(
|
||||
src: code,
|
||||
max_cpu_time: timeLimit,
|
||||
max_memory: 1024 * 1024 * memoryLimit,
|
||||
test_case_id: testCaseId,
|
||||
output: false,
|
||||
...(typeof testCase === "string" ? { test_case_id: testCase } : { test_case: testCase }),
|
||||
output,
|
||||
io_mode: {
|
||||
io_mode: "Standard IO",
|
||||
input: "input.txt",
|
||||
@@ -324,6 +335,48 @@ export async function failAbandonedSubmission(submissionId: string, error: unkno
|
||||
await markSystemError(submissionId, row.userId, error)
|
||||
}
|
||||
|
||||
/**
|
||||
* 运行时错误的诊断，结果写进 `statistic_info.runtime_error`（字段说明见契约）。
|
||||
*
|
||||
* C / C++ 直接用失败测试点的信号和退出码。Python 要拿回溯，得**重跑那一个测试点**：
|
||||
* 主判题不开 output（理由见 requestJudge），所以这里读出第一个失败测试点的输入，
|
||||
* 以内联测试点的形式单独再跑一次、打开 output。同样的代码、同样的输入，抛的是
|
||||
* 同一个异常。只多跑一个点，而且只在 Python 判出运行时错误时才跑。
|
||||
*
|
||||
* 诊断失败（读不到测试点、判题机没回回溯）就不写这个字段，学生看到的和原来一样。
|
||||
*/
|
||||
async function diagnoseRuntimeError(
|
||||
row: {
|
||||
submission: typeof schema.submission.$inferSelect
|
||||
problem: typeof schema.problem.$inferSelect
|
||||
},
|
||||
source: string,
|
||||
prependLines: number,
|
||||
failed: JudgeCase,
|
||||
): Promise<RuntimeErrorInfo | null> {
|
||||
if (row.submission.language !== "Python")
|
||||
return nativeRuntimeError(failed.signal, failed.exit_code)
|
||||
|
||||
const info = await readInfo(row.problem.testCaseId)
|
||||
const inputName = info?.test_cases?.[failed.test_case]?.input_name
|
||||
if (!inputName) return null
|
||||
const input = await readFile(
|
||||
resolvePath(config.testCaseDirectory, row.problem.testCaseId, inputName),
|
||||
"utf8",
|
||||
)
|
||||
const response = await requestJudge(
|
||||
row.submission.language,
|
||||
source,
|
||||
row.problem.timeLimit,
|
||||
row.problem.memoryLimit,
|
||||
[{ input, output: "" }],
|
||||
true,
|
||||
)
|
||||
const output = Array.isArray(response.data) ? response.data[0]?.output : null
|
||||
if (typeof output !== "string") return null
|
||||
return parsePythonTraceback(output, row.submission.code, prependLines)
|
||||
}
|
||||
|
||||
export async function judgeSubmission(job: JudgeJobData) {
|
||||
const [row] = await db
|
||||
.select({
|
||||
@@ -404,6 +457,22 @@ export async function judgeSubmission(job: JudgeJobData) {
|
||||
)?.error_message
|
||||
if (typeof failedMessage === "string") statisticInfo.err_info = failedMessage
|
||||
|
||||
// 比赛不诊断：ACM 只报对错，和「通过几个测试点」、AI 提示不给比赛提交同一个口径
|
||||
if (
|
||||
result === JudgeStatus.RUNTIME_ERROR &&
|
||||
firstFailure &&
|
||||
row.submission.contestId === null
|
||||
) {
|
||||
const prependLines = template ? template.prepend.split("\n").length : 0
|
||||
const diagnosis = await diagnoseRuntimeError(row, source, prependLines, firstFailure).catch(
|
||||
(error: unknown) => {
|
||||
console.warn(`[judge] 运行时错误诊断失败 ${row.submission.id}:`, error)
|
||||
return null
|
||||
},
|
||||
)
|
||||
if (diagnosis) statisticInfo.runtime_error = diagnosis
|
||||
}
|
||||
|
||||
if (result === JudgeStatus.ACCEPTED) {
|
||||
const rules = astRulesForLanguage(row.problem.astRules, row.submission.language)
|
||||
if (rules.length > 0) {
|
||||
|
||||
@@ -0,0 +1,110 @@
|
||||
import type { StatisticInfo } from "@oj2/contract"
|
||||
|
||||
/**
|
||||
* 运行时错误的诊断：把判题机的原始结果收成 `statistic_info.runtime_error`，
|
||||
* 前端再翻成中文（apps/web/src/oj/problem/utils/runtimeError.ts）。
|
||||
*
|
||||
* 原来学生拿到的只有「运行时错误」四个字：主判题是 `output: false`，Python 的回溯
|
||||
* 根本没回来；C 的信号值倒是有，但在 `info` 里，而 `info` 只给管理员。
|
||||
*
|
||||
* Python 的回溯要靠**重跑一次**拿（run.ts 的 diagnoseRuntimeError）：只拿第一个
|
||||
* 失败的测试点、打开 output 单独跑。不给主判题开 output —— 那样每个测试点的完整
|
||||
* 输出（每个最多 16MB）都会回到 worker，一个死循环打印就能把内存撑爆。
|
||||
*
|
||||
* **原文消息一律不存**。Python 的异常消息里常带着测试点的输入，比如
|
||||
* `invalid literal for int() with base 10: '1 2'`，给学生看就等于把隐藏数据放出去 ——
|
||||
* 故意写 `raise Exception(input())` 就能一个点一个点地套。所以只存：
|
||||
* - 异常类型和归好类的 `kind`（按消息的**句式**判，不取消息里的值）
|
||||
* - NameError / AttributeError 点名的名字，而且只在它**原样出现在学生代码里**时才存
|
||||
*/
|
||||
export type RuntimeErrorInfo = NonNullable<StatisticInfo["runtime_error"]>
|
||||
|
||||
const IDENTIFIER = /^[A-Za-z_]\w*$/
|
||||
|
||||
/** 按消息句式细分。键是异常类型，值是 [句式, kind]，从上往下取第一个对上的 */
|
||||
const KINDS: Record<string, [RegExp, string][]> = {
|
||||
ValueError: [
|
||||
[/^invalid literal for int\(\)/, "int-parse"],
|
||||
[/^could not convert string to float/, "float-parse"],
|
||||
[/^(not enough|too many) values to unpack/, "unpack"],
|
||||
[/^math domain error/, "math-domain"],
|
||||
],
|
||||
TypeError: [
|
||||
[/^can only concatenate str/, "str-concat"],
|
||||
[/^unsupported operand type/, "operand"],
|
||||
[/cannot be interpreted as an integer/, "not-int"],
|
||||
[/not supported between instances of/, "compare"],
|
||||
[/object is not callable/, "not-callable"],
|
||||
[/object is not subscriptable/, "not-subscriptable"],
|
||||
[/missing \d+ required positional argument/, "missing-arg"],
|
||||
[/^(list|string|tuple) indices must be integers/, "index-type"],
|
||||
],
|
||||
IndexError: [[/out of range/, "index"]],
|
||||
}
|
||||
|
||||
/** 名字在学生自己的代码里原样出现过才算数，否则它可能来自输入（`eval(input())`） */
|
||||
function nameFromCode(name: string | undefined, code: string) {
|
||||
if (!name || !IDENTIFIER.test(name)) return undefined
|
||||
return new RegExp(`\\b${name}\\b`).test(code) ? name : undefined
|
||||
}
|
||||
|
||||
/**
|
||||
* 从程序输出里解析 Python 回溯。stdout 和 stderr 在判题机里写的是同一个文件，
|
||||
* 所以回溯接在正常输出后面；只看末尾，取最后一段 Traceback（链式异常取最外层）。
|
||||
*
|
||||
* `prependLines` 是题目模板拼在学生代码前面的行数：判题跑的是
|
||||
* `${prepend}\n${code}\n${append}`，回溯里的行号要扣掉它才对得上编辑器。
|
||||
*/
|
||||
export function parsePythonTraceback(
|
||||
output: string,
|
||||
code: string,
|
||||
prependLines: number,
|
||||
): RuntimeErrorInfo | null {
|
||||
const tail = output.slice(-16_000)
|
||||
const start = tail.lastIndexOf("Traceback (most recent call last):")
|
||||
if (start < 0) return null
|
||||
const lines = tail.slice(start).trimEnd().split("\n")
|
||||
|
||||
const last = lines[lines.length - 1]!
|
||||
const head = last.match(/^([A-Za-z_][\w.]*)(?::\s?(.*))?$/)
|
||||
if (!head) return null
|
||||
const type = head[1]!.split(".").pop()!
|
||||
const message = head[2] ?? ""
|
||||
|
||||
// 最后一个落在 solution.py 里的栈帧：错误发生在标准库里（statistics.mean([]) 之类）时，
|
||||
// 最后一帧是库文件，学生要看的是自己调它的那一行
|
||||
let sourceLine: number | null = null
|
||||
for (const line of lines) {
|
||||
const frame = line.match(/^\s*File "([^"]*)", line (\d+)/)
|
||||
if (frame && frame[1]!.endsWith("solution.py")) sourceLine = Number(frame[2])
|
||||
}
|
||||
const codeLines = code.split("\n").length
|
||||
const lineInCode = sourceLine === null ? null : sourceLine - prependLines
|
||||
const info: RuntimeErrorInfo = {
|
||||
line: lineInCode !== null && lineInCode >= 1 && lineInCode <= codeLines ? lineInCode : null,
|
||||
type,
|
||||
}
|
||||
|
||||
const kind = KINDS[type]?.find(([pattern]) => pattern.test(message))?.[1]
|
||||
if (kind) info.kind = kind
|
||||
|
||||
if (type === "NameError" || type === "UnboundLocalError") {
|
||||
const match = message.match(/^(?:name|cannot access local variable) '([^']+)'/)
|
||||
const name = nameFromCode(match?.[1], code)
|
||||
if (name) info.name = name
|
||||
const suggestion = message.match(/Did you mean: '([^']+)'\?/)?.[1]
|
||||
if (suggestion && IDENTIFIER.test(suggestion)) info.suggestion = suggestion
|
||||
} else if (type === "AttributeError") {
|
||||
const name = nameFromCode(message.match(/has no attribute '([^']+)'/)?.[1], code)
|
||||
if (name) info.name = name
|
||||
}
|
||||
return info
|
||||
}
|
||||
|
||||
/** C / C++：没有回溯，信号和退出码就是全部线索。两个都为 0 时没什么可说的 */
|
||||
export function nativeRuntimeError(signal: unknown, exitCode: unknown): RuntimeErrorInfo | null {
|
||||
const info: RuntimeErrorInfo = {}
|
||||
if (typeof signal === "number" && signal > 0) info.signal = signal
|
||||
if (typeof exitCode === "number" && exitCode !== 0) info.exit_code = exitCode
|
||||
return info.signal || info.exit_code ? info : null
|
||||
}
|
||||
Reference in new issue
Block a user