原来答案错误只有四个字。6901 条历史答案错误在公开样例上重跑过：73% 在样例上就错了，
其中约三分之一只差格式（没输出、input() 提示文字、空格换行、中英文标点、小数位数）。
- judge/run.ts 的 checkSamples：判出答案错误后以内联测试点重跑题目样例（output: true），
记下第一个没过的样例和学生在它上面的输出，写进 statistic_info.sample_check；
样例全过则只记 passed。全是公开信息，不碰隐藏测试点；比赛不跑（有期末考试）
- 前端按差异归类（数字先于「多了内容」判，免得 007 被当成多打了文字），
并排显示输入 / 正确输出 / 你的输出，标出第一处不同的行；只差空白时把空格、换行画出来，
末尾空白不画（判题本来就忽略）
- 提示给具体值（「正确输出是 84.10，你输出的是 84.1」），input("…") 单独提示
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
xuyueandClaude Opus 5.5 committed
1 parent
db70ab62e8
commit
7e964fdbdd
6 files changed
+446
-2
No files matched your search
@@ -16,6 +16,7 @@ import SubmissionResultTag from "shared/components/SubmissionResultTag.vue"
|
||||
import { useProblemStore } from "oj/store/problem"
|
||||
import PythonErrorExplain from "./PythonErrorExplain.vue"
|
||||
import RuntimeErrorExplain from "./RuntimeErrorExplain.vue"
|
||||
import WrongAnswerExplain from "./WrongAnswerExplain.vue"
|
||||
import { useAIStream } from "shared/composables/aiStream"
|
||||
import { submitHintFeedback } from "oj/api"
|
||||
import { MdPreview } from "md-editor-v3"
|
||||
@@ -113,6 +114,13 @@ const runtimeError = computed(() => {
|
||||
return submission.statisticInfo?.runtime_error ?? null
|
||||
})
|
||||
|
||||
/** 答案错误时公开样例上的对比，交给 WrongAnswerExplain；比赛提交没有（见后端 checkSamples） */
|
||||
const sampleCheck = computed(() => {
|
||||
const submission = props.submission
|
||||
if (!submission || submission.result !== SubmissionStatus.wrong_answer) return null
|
||||
return submission.statisticInfo?.sample_check ?? null
|
||||
})
|
||||
|
||||
const msg = computed(() => {
|
||||
if (!props.submission) return ""
|
||||
// 走 PythonErrorExplain / RuntimeErrorExplain 那两张中文卡片
|
||||
@@ -285,6 +293,7 @@ const columns: DataTableColumn<JudgeCaseResult>[] = [
|
||||
v-if="
|
||||
pythonCompileError ||
|
||||
runtimeError ||
|
||||
sampleCheck ||
|
||||
msg ||
|
||||
infoTable.length ||
|
||||
submission.statisticInfo?.ast_results?.length
|
||||
@@ -292,6 +301,12 @@ const columns: DataTableColumn<JudgeCaseResult>[] = [
|
||||
>
|
||||
<PythonErrorExplain v-if="pythonCompileError" :err-info="pythonCompileError" />
|
||||
<RuntimeErrorExplain v-if="runtimeError" :info="runtimeError" :code="submission.code" />
|
||||
<WrongAnswerExplain
|
||||
v-if="sampleCheck"
|
||||
:check="sampleCheck"
|
||||
:code="submission.code"
|
||||
:language="submission.language"
|
||||
/>
|
||||
<n-card v-if="submission.statisticInfo?.ast_results?.length" embedded>
|
||||
<n-flex vertical :size="8">
|
||||
<n-flex
|
||||
|
||||
@@ -0,0 +1,171 @@
|
||||
<script setup lang="ts">
|
||||
import { JUDGE_STATUS, SubmissionStatus } from "utils/constants"
|
||||
import {
|
||||
classifyDiff,
|
||||
explainDiff,
|
||||
firstDifferentLine,
|
||||
type SampleCheck,
|
||||
} from "oj/problem/utils/wrongAnswer"
|
||||
|
||||
/**
|
||||
* 答案错误时的样例对比（数据见后端 judge/run.ts 的 checkSamples）。
|
||||
* 全是公开信息：样例写在题面上，「你的输出」是学生自己的程序在公开输入上的输出。
|
||||
*/
|
||||
const props = defineProps<{
|
||||
check: SampleCheck
|
||||
code: string
|
||||
language: string
|
||||
}>()
|
||||
|
||||
/** 后端每段截到 2000 字，到了上限就是被截过 */
|
||||
const TEXT_LIMIT = 2000
|
||||
|
||||
const onSample = computed(() => props.check.result === SubmissionStatus.wrong_answer)
|
||||
const kind = computed(() =>
|
||||
onSample.value ? classifyDiff(props.check.expected ?? "", props.check.output ?? "") : null,
|
||||
)
|
||||
const hint = computed(() => {
|
||||
if (props.check.passed) return ""
|
||||
if (!kind.value) {
|
||||
const name = JUDGE_STATUS[props.check.result as keyof typeof JUDGE_STATUS]?.name ?? "出错了"
|
||||
return `你的程序在这个例子上就「${name}」了，先让它在例子上跑通。`
|
||||
}
|
||||
return explainDiff(kind.value, props.check, props.code, props.language)
|
||||
})
|
||||
const diffLine = computed(() =>
|
||||
firstDifferentLine(props.check.expected ?? "", props.check.output ?? ""),
|
||||
)
|
||||
|
||||
/**
|
||||
* 只差空白的时候把空白画出来，不然两边看着一模一样。
|
||||
*
|
||||
* 末尾的空白先去掉：判题比的是整个输出去掉末尾空白之后的样子（JudgeServer 的
|
||||
* stripped_output_md5），print 自己带的那个换行不算错。画出来的话，学生会以为
|
||||
* 是结尾那个「↵」惹的祸。
|
||||
*/
|
||||
function lines(text: string) {
|
||||
const visible = kind.value === "whitespace"
|
||||
const parts = text.replace(/\s+$/, "").split("\n")
|
||||
return parts.map((line, i) => {
|
||||
if (!visible) return line
|
||||
const ending = i < parts.length - 1 ? "↵" : ""
|
||||
return line.replace(/ /g, "·").replace(/\t/g, "→") + ending
|
||||
})
|
||||
}
|
||||
</script>
|
||||
|
||||
<template>
|
||||
<n-card embedded class="explain-card">
|
||||
<div v-if="check.passed" class="explain">
|
||||
题目里的例子都对了，没通过的是隐藏的测试点。想想特殊情况：最大的数、最小的数、0、负数，或者题目里专门提到的情况。
|
||||
</div>
|
||||
<n-flex v-else vertical :size="12">
|
||||
<div class="explain">
|
||||
<b>例子 {{ (check.index ?? 0) + 1 }} 没有通过：</b>{{ hint }}
|
||||
</div>
|
||||
<div class="blocks">
|
||||
<div class="block input">
|
||||
<div class="label">输入</div>
|
||||
<div class="text">
|
||||
<div
|
||||
v-for="(line, i) in (check.input ?? '').split('\n')"
|
||||
:key="i"
|
||||
class="line"
|
||||
v-text="line"
|
||||
/>
|
||||
</div>
|
||||
</div>
|
||||
<div class="block">
|
||||
<div class="label">正确输出</div>
|
||||
<div class="text">
|
||||
<div
|
||||
v-for="(line, i) in lines(check.expected ?? '')"
|
||||
:key="i"
|
||||
class="line"
|
||||
:class="{ diff: onSample && i === diffLine }"
|
||||
v-text="line"
|
||||
/>
|
||||
</div>
|
||||
</div>
|
||||
<div class="block">
|
||||
<div class="label">你的输出</div>
|
||||
<div v-if="check.output" class="text">
|
||||
<div
|
||||
v-for="(line, i) in lines(check.output)"
|
||||
:key="i"
|
||||
class="line"
|
||||
:class="{ diff: onSample && i === diffLine }"
|
||||
v-text="line"
|
||||
/>
|
||||
</div>
|
||||
<div v-else class="text empty">（什么都没有输出）</div>
|
||||
<n-text v-if="check.output && check.output.length >= TEXT_LIMIT" depth="3" class="cut">
|
||||
输出太长，只显示了前面一部分
|
||||
</n-text>
|
||||
</div>
|
||||
</div>
|
||||
</n-flex>
|
||||
</n-card>
|
||||
</template>
|
||||
|
||||
<style scoped>
|
||||
.explain-card {
|
||||
max-width: 560px;
|
||||
}
|
||||
|
||||
.explain {
|
||||
font-size: 16px;
|
||||
line-height: 1.7;
|
||||
}
|
||||
|
||||
.blocks {
|
||||
display: grid;
|
||||
grid-template-columns: 1fr 1fr;
|
||||
gap: 8px;
|
||||
}
|
||||
|
||||
.block {
|
||||
min-width: 0;
|
||||
}
|
||||
|
||||
.block.input {
|
||||
grid-column: 1 / -1;
|
||||
}
|
||||
|
||||
.label {
|
||||
font-size: 13px;
|
||||
opacity: 0.7;
|
||||
margin-bottom: 4px;
|
||||
}
|
||||
|
||||
.text {
|
||||
margin: 0;
|
||||
font-family: ui-monospace, "SF Mono", Menlo, Consolas, monospace;
|
||||
padding: 8px 10px;
|
||||
border-radius: 4px;
|
||||
font-size: 14px;
|
||||
line-height: 1.6;
|
||||
max-height: 200px;
|
||||
overflow: auto;
|
||||
background-color: rgba(128, 128, 128, 0.1);
|
||||
}
|
||||
|
||||
/* 每行用 v-text 填，不写 {{ line }}：模板里标签两边的换行缩进会被 Vue 压成一个空格，
|
||||
配上 white-space: pre 就是每行开头多一个空格 —— 在一个专门比对空格的地方，这不能有 */
|
||||
.line {
|
||||
white-space: pre;
|
||||
min-height: 1.6em;
|
||||
}
|
||||
|
||||
.diff {
|
||||
background-color: rgba(208, 48, 80, 0.2);
|
||||
}
|
||||
|
||||
.empty {
|
||||
opacity: 0.6;
|
||||
}
|
||||
|
||||
.cut {
|
||||
font-size: 12px;
|
||||
}
|
||||
</style>
|
||||
@@ -76,7 +76,7 @@ export const PUNCTUATION_MAP: Record<string, string> = {
|
||||
* 几乎一样，「把「：」换成「:」」这句话学生看不出区别。「冒号是中文的，要换成英文的」
|
||||
* 看得懂，要换哪一个由卡片里那行代码的标红来指。
|
||||
*/
|
||||
const PUNCTUATION_NAMES: Record<string, string> = {
|
||||
export const PUNCTUATION_NAMES: Record<string, string> = {
|
||||
"（": "左括号",
|
||||
"）": "右括号",
|
||||
"，": "逗号",
|
||||
|
||||
@@ -0,0 +1,162 @@
|
||||
import type { StatisticInfo } from "@oj2/contract"
|
||||
import { PUNCTUATION_MAP, PUNCTUATION_NAMES } from "./pythonError"
|
||||
|
||||
/**
|
||||
* 答案错误时，比较学生在**公开样例**上的输出和正确输出，归一类、给一句中文提示。
|
||||
* 数据来自后端 judge/run.ts 的 checkSamples（`statistic_info.sample_check`）。
|
||||
*
|
||||
* 分类是拿 2025 秋以来 5055 条「样例上就错了」的答案错误定的：
|
||||
* 格式和内容都不同 44%、数值不对 15%、没有输出 10%、多了文字 9%、只差空格换行 9%、
|
||||
* 少了内容 5%、中英文标点 4%、数字写法（小数位数）2.5%、大小写 1%。
|
||||
* 前几类之外的「几乎对了」合起来约三分之一，一句话就能改对。
|
||||
*
|
||||
* 样例是公开的，这里可以放心地把具体的值说出来（「正确输出是 84.10，你输出的是 84.1」）。
|
||||
*/
|
||||
export type SampleCheck = NonNullable<StatisticInfo["sample_check"]>
|
||||
|
||||
export type DiffKind =
|
||||
| "no-output"
|
||||
| "whitespace"
|
||||
| "punctuation"
|
||||
| "case"
|
||||
| "extra"
|
||||
| "missing"
|
||||
| "number-format"
|
||||
| "values"
|
||||
| "other"
|
||||
|
||||
const NUMBER = /-?\d+(?:\.\d+)?/g
|
||||
|
||||
function squash(text: string) {
|
||||
return text.replace(/\s+/g, " ").trim()
|
||||
}
|
||||
|
||||
function toEnglishPunctuation(text: string) {
|
||||
return [...text].map((ch) => PUNCTUATION_MAP[ch] ?? ch).join("")
|
||||
}
|
||||
|
||||
export function classifyDiff(expected: string, output: string): DiffKind {
|
||||
const e = squash(expected)
|
||||
const o = squash(output)
|
||||
if (!o) return "no-output"
|
||||
if (e === o) return "whitespace"
|
||||
if (squash(toEnglishPunctuation(expected)) === squash(toEnglishPunctuation(output))) {
|
||||
return "punctuation"
|
||||
}
|
||||
if (e.toLowerCase() === o.toLowerCase()) return "case"
|
||||
// 数字先于「多了内容」判：007 里包含 7、3.0 里包含 3，先判包含就会把它们当成
|
||||
// 「多打了说明文字」，给出完全不相干的建议
|
||||
const en = e.match(NUMBER) ?? []
|
||||
const on = o.match(NUMBER) ?? []
|
||||
const sameShape = e.replace(NUMBER, "N") === o.replace(NUMBER, "N") && en.length === on.length
|
||||
if (sameShape && en.length) {
|
||||
const sameValues = en.every((value, i) => Math.abs(Number(value) - Number(on[i])) < 1e-9)
|
||||
return sameValues ? "number-format" : "values"
|
||||
}
|
||||
if (e && o.includes(e)) return "extra"
|
||||
if (o && e.includes(o)) return "missing"
|
||||
return "other"
|
||||
}
|
||||
|
||||
/** 第一处不一样的标点：[正确输出里的, 你输出的] */
|
||||
function firstPunctuationPair(expected: string, output: string) {
|
||||
const e = [...squash(expected)]
|
||||
const o = [...squash(output)]
|
||||
for (let i = 0; i < Math.min(e.length, o.length); i++) {
|
||||
if (e[i] !== o[i]) return [e[i]!, o[i]!] as const
|
||||
}
|
||||
return null
|
||||
}
|
||||
|
||||
function punctuationHint(expected: string, output: string) {
|
||||
const pair = firstPunctuationPair(expected, output)
|
||||
if (pair) {
|
||||
const [want, got] = pair
|
||||
// 中文的那个才在名字表里
|
||||
const name = PUNCTUATION_NAMES[got] ?? PUNCTUATION_NAMES[want]
|
||||
if (name && PUNCTUATION_MAP[got] === want) {
|
||||
return `内容对了，但标点不一样：正确输出用的是英文的${name}，你输出的是中文的${name}。输出里的标点要和题目完全一样。`
|
||||
}
|
||||
if (name && PUNCTUATION_MAP[want] === got) {
|
||||
return `内容对了，但标点不一样：正确输出用的是中文的${name}，你输出的是英文的${name}。输出里的标点要和题目完全一样。`
|
||||
}
|
||||
}
|
||||
return "内容对了，但标点不一样：输出里的标点要和题目完全一样，注意分清中文标点和英文标点。"
|
||||
}
|
||||
|
||||
function numberFormatHint(expected: string, output: string, language: string) {
|
||||
const en = squash(expected).match(NUMBER) ?? []
|
||||
const on = squash(output).match(NUMBER) ?? []
|
||||
const i = en.findIndex((value, k) => value !== on[k])
|
||||
if (i < 0) return "数字的写法和正确输出不一样。"
|
||||
const want = en[i]!
|
||||
const got = on[i]!
|
||||
const decimals = want.split(".")[1]?.length ?? 0
|
||||
const head = `正确输出是 ${want}，你输出的是 ${got}。`
|
||||
// 007 和 7：逆序数、取位数一类题的老毛病，跟小数没关系
|
||||
if (/^-?0\d/.test(got)) {
|
||||
const how = language === "Python" ? "用 int() 转成整数再输出" : "按整数（%d）输出"
|
||||
return `${head}数字前面多了 0。要输出的是一个数，不是一串数字字符：${how}，前面的 0 就没了。`
|
||||
}
|
||||
if (decimals > 0) {
|
||||
const how =
|
||||
language === "Python"
|
||||
? `Python 可以写 f"{x:.${decimals}f}"`
|
||||
: `C 语言可以写 printf("%.${decimals}f", x)`
|
||||
return `${head}题目要求保留 ${decimals} 位小数，${how}。`
|
||||
}
|
||||
if (!got.includes(".")) return `${head}数字的写法和正确输出不一样。`
|
||||
if (language === "Python") {
|
||||
return `${head}这里要输出整数：Python 里 / 算出来的总是小数，整除要用 //，或者用 int() 转换。`
|
||||
}
|
||||
return `${head}这里要输出整数，不要带小数点：printf 用 %d 输出整数。`
|
||||
}
|
||||
|
||||
/** Python 的 input("请输入……")：提示文字会被打印进输出，判题时必然对不上 */
|
||||
export function hasInputPrompt(code: string) {
|
||||
return /\binput\(\s*[fr]?["'][^"'\n]*\S[^"'\n]*["']\s*\)/.test(code)
|
||||
}
|
||||
|
||||
export function explainDiff(
|
||||
kind: DiffKind,
|
||||
check: SampleCheck,
|
||||
code: string,
|
||||
language: string,
|
||||
): string {
|
||||
const expected = check.expected ?? ""
|
||||
const output = check.output ?? ""
|
||||
const printName = language === "Python" ? "print" : "printf"
|
||||
switch (kind) {
|
||||
case "no-output":
|
||||
return `你的程序什么都没有输出。检查一下是不是忘了写 ${printName}，或者 ${printName} 写在了执行不到的地方（比如某个 if 的里面）。`
|
||||
case "whitespace":
|
||||
return "内容对了，只是空格或换行不一样。下面把空格画成了「·」、换行画成了「↵」，对照着看看哪一行多了或少了。"
|
||||
case "punctuation":
|
||||
return punctuationHint(expected, output)
|
||||
case "case":
|
||||
return "内容对了，只是字母的大小写不一样。输出要和题目的大小写完全一样。"
|
||||
case "extra":
|
||||
if (language === "Python" && hasInputPrompt(code)) {
|
||||
return 'input() 括号里的提示文字也会被输出，判题时就对不上了。把 input("……") 里的文字删掉，只写 input()。'
|
||||
}
|
||||
return "你的输出比正确输出多了一些内容。题目要输出什么就只输出什么，不要加「请输入」「结果是」这类提示或说明文字。"
|
||||
case "missing":
|
||||
return "你的输出比正确输出少了一部分，对照看看漏了什么。"
|
||||
case "number-format":
|
||||
return numberFormatHint(expected, output, language)
|
||||
case "values":
|
||||
return "输出的格式对了，但算出来的数不对。拿这个例子的输入，自己一步一步算一遍，看看程序在哪一步和你想的不一样。"
|
||||
case "other":
|
||||
return "对照一下你的输出和正确输出，找找从哪里开始不一样。"
|
||||
}
|
||||
}
|
||||
|
||||
/** 第一处不一样的行号，0 起；完全一样（只差结尾空白）时为 -1 */
|
||||
export function firstDifferentLine(expected: string, output: string) {
|
||||
const e = expected.replace(/\s+$/, "").split("\n")
|
||||
const o = output.replace(/\s+$/, "").split("\n")
|
||||
for (let i = 0; i < Math.max(e.length, o.length); i++) {
|
||||
if (e[i] !== o[i]) return i
|
||||
}
|
||||
return -1
|
||||
}
|
||||
Reference in new issue
Block a user