fix(modeler): 保真扫描回归加固——refs袋+unresolved入门+规模门+stderr(评审I1/M1/M2/M5)
- I1 引用回归盲区:surface() 新增 refs 袋(sourceRef/targetRef/bpmnElement/ processRef/attachedToRef/default,与 sanitizeIds 同覆盖面)并入比对键;normWarn 的 unresolved reference 类 warning 计入硬门 failures(净化/往返破坏连线的直接信号)。 - M1:删除未使用的 realFail。 - M2:新增 EXPECTED_TOTAL 环境变量(默认 1512),实际语料数不符即判失败; decodeFailures.length 并入硬门(过滤集内本应 100% 干净 UTF8)。 - M5:崩溃分支透出子进程 e.stderr。 - M3/M4:命名空间字面前缀识别的语料前提、extElemText trim 口径各加注释声明。 重跑实测(node tools/modeler-fidelity-sweep.mjs):匹配 1512/1512 零丢失, 解析异常 0,解码失败 0,warnings=0,exit 0;EXPECTED_TOTAL=9999 时 exit 1 (规模门生效)。npx vitest run src/modeler 12/12 绿;npx tsc -b 干净(exit 0)。 Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This commit is contained in:
@@ -87,6 +87,9 @@ function sanitizeIds(xml) {
|
||||
}
|
||||
|
||||
// ── 命名空间嗅探:activiti 优先,其次 flowable,默认 activiti ─────────────────
|
||||
// M3 前提:靠字面量 uri/前缀识别命名空间,依赖语料使用规范前缀 activiti/flowable。
|
||||
// §0 实测本库确如此(activiti 1509 / flowable 3,无自定义前缀重映射),故成立;
|
||||
// 若未来语料出现 xmlns:act="http://activiti.org/bpmn" 之类别名,此处需扩正则。
|
||||
function sniffModdle(xml) {
|
||||
const hasActiviti = xml.includes('activiti.org/bpmn') || /\bactiviti:/.test(xml);
|
||||
const hasFlowable = xml.includes('flowable.org/bpmn') || /\bflowable:/.test(xml);
|
||||
@@ -101,6 +104,9 @@ function sniffModdle(xml) {
|
||||
const TAG_RE = /<([/!?]?)([\w:.-]+)((?:[^>"']|"[^"]*"|'[^']*')*)>/g;
|
||||
const ATTR_RE = /([\w:.-]+)\s*=\s*"([^"]*)"/g;
|
||||
const EXT_TEXT_RE = /<((?:activiti|flowable):[\w.-]+)((?:[^>"']|"[^"]*"|'[^']*')*)>([^<]*)<\//g;
|
||||
// I1:引用类属性单独入袋,捕捉 round-trip/净化重写破坏的连线(sourceRef/targetRef/
|
||||
// bpmnElement/processRef/attachedToRef/default)。与 sanitizeIds 的 ID_ATTR 同覆盖面。
|
||||
const REF_ATTRS = new Set(['sourceRef', 'targetRef', 'bpmnElement', 'processRef', 'attachedToRef', 'default']);
|
||||
|
||||
function surface(xml) {
|
||||
const extAttrs = []; // 原生元素上的 activiti:/flowable: 属性 p:name=value
|
||||
@@ -108,6 +114,7 @@ function surface(xml) {
|
||||
const extElemAttrs = [];// 扩展元素上的属性(含无前缀) tag@attr=value
|
||||
const extElemText = []; // 扩展叶元素的文本 tag#text=value
|
||||
const ids = []; // 全部 id="..."(含原生/DI/扩展)
|
||||
const refs = []; // 引用类属性 an=value(连线/DI 指向完整性,I1)
|
||||
|
||||
let m;
|
||||
TAG_RE.lastIndex = 0;
|
||||
@@ -123,6 +130,7 @@ function surface(xml) {
|
||||
const an = a[1];
|
||||
const av = decodeEntities(a[2]);
|
||||
if (an === 'id') ids.push(av);
|
||||
else if (REF_ATTRS.has(an)) refs.push(`${an}=${av}`);
|
||||
if (isExt) {
|
||||
extElemAttrs.push(`${name}@${an}=${av}`);
|
||||
} else if (/^(activiti|flowable):/.test(an)) {
|
||||
@@ -132,13 +140,15 @@ function surface(xml) {
|
||||
if (isExt) extElemTags.push(name);
|
||||
}
|
||||
|
||||
// M4:extElemText 对叶元素文本做 trim 归一(去两端空白),故不比对纯空白/缩进差异。
|
||||
// 扩展叶文本(监听器 field 值、表单选项 name 等)非空白敏感,且前后同样 trim,对称无偏。
|
||||
EXT_TEXT_RE.lastIndex = 0;
|
||||
while ((m = EXT_TEXT_RE.exec(xml))) {
|
||||
const t = decodeEntities(m[3]).trim();
|
||||
if (t) extElemText.push(`${m[1]}#text=${t}`);
|
||||
}
|
||||
|
||||
return { extAttrs, extElemTags, extElemAttrs, extElemText, ids };
|
||||
return { extAttrs, extElemTags, extElemAttrs, extElemText, ids, refs };
|
||||
}
|
||||
|
||||
function bag(arr) {
|
||||
@@ -206,9 +216,15 @@ async function main() {
|
||||
continue;
|
||||
}
|
||||
|
||||
const unresolvedRefs = [];
|
||||
for (const w of warnings) {
|
||||
const key = normWarn(String(w && w.message || w));
|
||||
const msg = String(w && w.message || w);
|
||||
const key = normWarn(msg);
|
||||
warnCounts.set(key, (warnCounts.get(key) || 0) + 1);
|
||||
if (key === 'unresolved reference') {
|
||||
const mr = msg.match(/unresolved reference <([^>]+)>/i);
|
||||
unresolvedRefs.push(mr ? mr[1] : msg.slice(0, 60));
|
||||
}
|
||||
}
|
||||
|
||||
// 5) 数据级零丢失比对:sanitized(前) vs out(后)
|
||||
@@ -216,20 +232,31 @@ async function main() {
|
||||
const after = surface(out);
|
||||
const diffs = {};
|
||||
let lost = false;
|
||||
for (const k of ['extAttrs', 'extElemTags', 'extElemAttrs', 'extElemText', 'ids']) {
|
||||
for (const k of ['extAttrs', 'extElemTags', 'extElemAttrs', 'extElemText', 'ids', 'refs']) {
|
||||
const d = diffBag(before[k], after[k]);
|
||||
if (d.missing.length || d.added.length) {
|
||||
diffs[k] = d;
|
||||
lost = true;
|
||||
}
|
||||
}
|
||||
if (lost) failures.push({ id: doc.id, name: doc.name, ns, diffs });
|
||||
else ok++;
|
||||
// I1②:未解析引用计入硬门(净化/往返破坏连线的直接信号,不再只当 warning)
|
||||
if (lost || unresolvedRefs.length) {
|
||||
failures.push({ id: doc.id, name: doc.name, ns, diffs, unresolved: unresolvedRefs });
|
||||
} else {
|
||||
ok++;
|
||||
}
|
||||
}
|
||||
|
||||
report({ total: docs.length, ok, failures, decodeFailures, parseErrors, warnCounts, nsCounts });
|
||||
// 硬门:真实 BPMN 全部零丢失(解码失败=非目标语料,单列不计入失败门)
|
||||
process.exit(failures.length === 0 && parseErrors.length === 0 ? 0 : 1);
|
||||
const expectedTotal = Number(process.env.EXPECTED_TOTAL || 1512);
|
||||
report({ total: docs.length, ok, failures, decodeFailures, parseErrors, warnCounts, nsCounts, expectedTotal });
|
||||
// 硬门:真实 BPMN 全部零丢失 + 无解析异常 + 无解码失败 + 语料规模符合预期。
|
||||
// (M2:解码失败与规模不符均判失败——过滤集内本应 100% 干净 UTF8,且恒 EXPECTED_TOTAL 篇)
|
||||
const gateOk =
|
||||
failures.length === 0 &&
|
||||
parseErrors.length === 0 &&
|
||||
decodeFailures.length === 0 &&
|
||||
docs.length === expectedTotal;
|
||||
process.exit(gateOk ? 0 : 1);
|
||||
}
|
||||
|
||||
// warning 归一:抽取属性/元素名做聚合键
|
||||
@@ -246,10 +273,11 @@ function normWarn(msg) {
|
||||
function report(r) {
|
||||
const line = '─'.repeat(60);
|
||||
process.stdout.write(`${line}\n保真扫描结果\n${line}\n`);
|
||||
const realFail = r.failures.length + r.parseErrors.length;
|
||||
const realTotal = r.total - r.decodeFailures.length;
|
||||
process.stdout.write(`匹配 ${r.ok}/${realTotal}(真实 BPMN,已排除 ${r.decodeFailures.length} 解码失败)\n`);
|
||||
process.stdout.write(`总文档 ${r.total}|零丢失 ${r.ok}|丢失 ${r.failures.length}|解析异常 ${r.parseErrors.length}|解码失败 ${r.decodeFailures.length}\n`);
|
||||
process.stdout.write(`匹配 ${r.ok}/${r.total} 零丢失(真实 BPMN)\n`);
|
||||
process.stdout.write(`总文档 ${r.total}(预期 ${r.expectedTotal})|零丢失 ${r.ok}|丢失 ${r.failures.length}|解析异常 ${r.parseErrors.length}|解码失败 ${r.decodeFailures.length}\n`);
|
||||
if (r.total !== r.expectedTotal) {
|
||||
process.stdout.write(`⚠ 语料规模 ${r.total} ≠ 预期 ${r.expectedTotal}(EXPECTED_TOTAL)→ 判失败\n`);
|
||||
}
|
||||
|
||||
process.stdout.write(`\n命名空间加载分布:\n`);
|
||||
for (const [ns, c] of [...r.nsCounts].sort((a, b) => b[1] - a[1])) {
|
||||
@@ -280,6 +308,9 @@ function report(r) {
|
||||
if (d.missing.length) process.stdout.write(` ${k} 丢失: ${d.missing.slice(0, 6).join(' ; ')}${d.missing.length > 6 ? ' …' : ''}\n`);
|
||||
if (d.added.length) process.stdout.write(` ${k} 新增: ${d.added.slice(0, 6).join(' ; ')}${d.added.length > 6 ? ' …' : ''}\n`);
|
||||
}
|
||||
if (f.unresolved && f.unresolved.length) {
|
||||
process.stdout.write(` 未解析引用(I1): ${f.unresolved.slice(0, 6).join(' ; ')}${f.unresolved.length > 6 ? ' …' : ''}\n`);
|
||||
}
|
||||
}
|
||||
if (r.failures.length > 30) process.stdout.write(` … 另有 ${r.failures.length - 30} 个失败文档\n`);
|
||||
}
|
||||
@@ -287,6 +318,8 @@ function report(r) {
|
||||
}
|
||||
|
||||
main().catch((e) => {
|
||||
// M5:psql 等子进程失败时把其 stderr 透出来(execFileSync 把 stderr 挂在 e.stderr)
|
||||
if (e && e.stderr) process.stderr.write(String(e.stderr));
|
||||
process.stderr.write(`扫描崩溃: ${e && e.stack || e}\n`);
|
||||
process.exit(2);
|
||||
});
|
||||
|
||||
Reference in New Issue
Block a user