fix(brand-label): 品牌标签清洗抽公共件,六模块读写两端统一过出口

PublishBrandLabel 从 publish 私有件搬入 common/util/BrandLabel(三条守卫尺子随迁:
唯一副本、实体写出不许绕过、解析端不许重复实现),品牌列的读入与出件两端都走这一处。

读入端接上:asincache 缓存落库、invalidasin 人工黑名单(先洗再判空——键对不上是静默失效,
比脏值出件更难发现)、appearancepatent resolveBrand(两条取值分支同一个串)、collectdata
入库、shopdatacrawl copyRow/sameRow/rowDedupKey(键与值分家会让同一行跨分片重复出件)、
撞款解析与 re-baseline、dedupe 按表头名只洗品牌列。
写出端接上:publish/similarasin×2/appearancepatent/collectdata/撞款 csv+xlsx/shopdatacrawl。

rows-snapshot golden 里品牌格去掉首尾空格:正本 strip 自 2026-10-08 就含 trim,
Java/Python/MySQL 三处同形,不是这次新加的行为。

门禁:后端 mvn -o test 4594 例 0 红。BrandTaskService(含四张 sheet 出件与解析端匹配)
因同仓另一路在飞改动未动,见 docs/audit/BACKLOG.md B5。
This commit is contained in:
2026-10-20 22:35:25 +08:00
parent e855801237
commit cdef922251
34 changed files with 456 additions and 49 deletions
@@ -1,4 +1,4 @@
package com.nanri.aiimage.modules.publish.service;
package com.nanri.aiimage.common.util;
import static org.assertj.core.api.Assertions.assertThat;
import static org.assertj.core.api.Assertions.fail;
@@ -14,7 +14,7 @@ import java.util.stream.Stream;
import org.junit.jupiter.api.Test;
/**
* 上架品牌值的清洗出口。
* 品牌标签前缀清洗出口。
*
* <p>2026-10-08 的事故链是:Excel 里带 "Brand:" 标签前缀 → WIPO 把它当 Solr 字段限定符
* ({@code undefined field Brand},单日 19,247 次 500)→ 上游错误信封不是合法 Base64、被客户端
@@ -24,29 +24,32 @@ import org.junit.jupiter.api.Test;
* {@code PublishTaskService.replaceRows} 会用客户端回传的分片整表删后重插,
* 脏值照样落库(现网 7 个脏值来源任务里 5 个的行创建时间 == 任务完成时刻,即由这条路径写入)。
* 所以护栏从"注释里提醒调用方"改成"落在唯一写库出口 + 可实跑的用例"。
*
* <p>本类原在 {@code modules/publish/service} 下,因品牌列的读写两端散在多个模块而出搬为公共件;
* 三条守卫尺子(唯一副本、出口不被绕过、解析端不重复实现)随迁。
*/
class PublishBrandLabelTest {
class BrandLabelTest {
private static final String OUTLET_CLASS = "PublishBrandLabel.java";
private static final String OUTLET_CLASS = "BrandLabel.java";
/** 标签正则的特征片段:出现第二处就等于开了第二条会漂移的副本。 */
private static final String REGEX_SIGNATURE = "marque|marka";
@Test
void stripsLabelPrefixButKeepsLookalikes() {
assertThat(PublishBrandLabel.strip("Brand: NIKE")).isEqualTo("NIKE");
assertThat(PublishBrandLabel.strip(" brand : NIKE ")).isEqualTo("NIKE");
assertThat(PublishBrandLabel.strip("品牌:耐克")).isEqualTo("耐克");
assertThat(PublishBrandLabel.strip("MARCA: X")).isEqualTo("X");
assertThat(BrandLabel.strip("Brand: NIKE")).isEqualTo("NIKE");
assertThat(BrandLabel.strip(" brand : NIKE ")).isEqualTo("NIKE");
assertThat(BrandLabel.strip("品牌:耐克")).isEqualTo("耐克");
assertThat(BrandLabel.strip("MARCA: X")).isEqualTo("X");
// "Brands:" 不是标签;值中间的冒号属于品牌自身内容,只剥行首一次
assertThat(PublishBrandLabel.strip("Brands: NIKE")).isEqualTo("Brands: NIKE");
assertThat(PublishBrandLabel.strip("Brand: NIKE: X")).isEqualTo("NIKE: X");
assertThat(PublishBrandLabel.strip("NIKE Brand")).isEqualTo("NIKE Brand");
assertThat(BrandLabel.strip("Brands: NIKE")).isEqualTo("Brands: NIKE");
assertThat(BrandLabel.strip("Brand: NIKE: X")).isEqualTo("NIKE: X");
assertThat(BrandLabel.strip("NIKE Brand")).isEqualTo("NIKE Brand");
// 标签只有出现在行首才是"字段名";句中出现的要原样留着,否则把 ^ 锚摘掉(变成"逢 Brand: 就裁")这条就会红
assertThat(PublishBrandLabel.strip("Super Brand: Nike"))
assertThat(BrandLabel.strip("Super Brand: Nike"))
.isEqualTo("Super Brand: Nike");
assertThat(PublishBrandLabel.strip("BrandNIKE")).isEqualTo("BrandNIKE");
assertThat(PublishBrandLabel.strip("")).isEmpty();
assertThat(PublishBrandLabel.strip(null)).isNull();
assertThat(BrandLabel.strip("BrandNIKE")).isEqualTo("BrandNIKE");
assertThat(BrandLabel.strip("")).isEmpty();
assertThat(BrandLabel.strip(null)).isNull();
}
@Test
@@ -54,8 +57,8 @@ class PublishBrandLabelTest {
// 现网 1,533 个脏值清洗后 0 个变空串(实测),但 "Brand:" 这种纯标签值确实会洗成空。
// 钉住这一条是因为 PublishTaskService.isBlankRow 要求八个字段同时为空才算空行,
// 单靠 brand 变空不会让一行消失 —— 改了这条判据就会静默丢行。
assertThat(PublishBrandLabel.strip("Brand:")).isEmpty();
assertThat(PublishBrandLabel.strip("Brand: ")).isEmpty();
assertThat(BrandLabel.strip("Brand:")).isEmpty();
assertThat(BrandLabel.strip("Brand: ")).isEmpty();
}
@Test
@@ -74,7 +77,7 @@ class PublishBrandLabelTest {
// 只有写进实体的那一行算出口;DTO 之间的搬运(copyRow/toRowDto)不是
if (line.matches("^entity\\.setBrand\\(.*\\);")) {
written.add(file.getFileName() + ": " + line);
if (!line.contains("PublishBrandLabel.strip(")) {
if (!line.contains("BrandLabel.strip(")) {
offenders.add(file.getFileName() + ": " + line);
}
}
@@ -83,7 +86,7 @@ class PublishBrandLabelTest {
// 取样集非空自检:扫描 0 命中时下面这条恒绿,等于没护栏
assertThat(written).as("publish 模块里写 PublishItemEntity.brand 的行应被扫到").isNotEmpty();
assertThat(offenders)
.as("绕过 PublishBrandLabel.strip 的 brand 写入点(客户端回传 replaceRows 会整表重插,"
.as("绕过 BrandLabel.strip 的 brand 写入点(客户端回传 replaceRows 会整表重插,"
+ "少堵一处就等于每完成一个任务把脏值写回去一次)")
.isEmpty();
}
@@ -104,7 +107,7 @@ class PublishBrandLabelTest {
@Test
void parseSideAlsoDelegatesInsteadOfReimplementing() throws IOException {
String text = read(sourceOf("PublishWorkbookService.java"));
assertThat(text).contains("row.setBrand(PublishBrandLabel.strip(values.get(3)));");
assertThat(text).contains("row.setBrand(BrandLabel.strip(values.get(3)));");
assertThat(text).doesNotContain("stripBrandLabel");
}
@@ -29,6 +29,7 @@ import static org.junit.jupiter.api.Assertions.assertTrue;
*/
class AppearancePatentRowOutcomeTest {
private static final int COL_BRAND = 4;
private static final int COL_TITLE = 6;
private static final int COL_URL = 7;
private static final int COL_TITLE_RISK = 9;
@@ -355,4 +356,36 @@ class AppearancePatentRowOutcomeTest {
}
throw new AssertionError("原因 sheet 里没有 ASIN=" + asin + " 的行");
}
// ---------- 4. 品牌列只认洗过标签前缀的值 ----------
@Test
void brandCellNeverCarriesTheAmazonBylinePrefix() throws Exception {
// 品牌列有两个来源:用户上传表的「品牌」列(这里就是 "Brand: Acme")与客户端回传行的 Brand,
// 详情页整行复制来的署名前缀既脏出件、又会在品牌库查询里被当字段限定符打成上游 500。
List<AppearancePatentParsedRowVo> all = List.of(sheetRowOnlyId("1", "B0A"));
AppearancePatentResultRowDto resultRow = judgedRow("1", "B0A", "T", "https://img/1.jpg");
Map<String, AppearancePatentResultRowDto> byRowKey = new LinkedHashMap<>();
byRowKey.put(AppearancePatentSheetBuilder.rowKey(all.get(0)), resultRow);
assertEquals("Acme", brandCell(all, byRowKey), "回传行没带品牌时,源表值也要剥掉标签");
resultRow.setBrand("Brand: Nike");
assertEquals("Nike", brandCell(all, byRowKey));
// "Brands:" 是真实品牌写法、不是字段标签;值中间的冒号属于品牌本体,只能剥行首一次
resultRow.setBrand("Brands: Nike");
assertEquals("Brands: Nike", brandCell(all, byRowKey));
// 只剩标签的回传值等于这行没带品牌:回退源表列,而不是把洗出来的空串当成有效值写进格子
resultRow.setBrand("Brand:");
assertEquals("Acme", brandCell(all, byRowKey));
}
private static String brandCell(List<AppearancePatentParsedRowVo> rows,
Map<String, AppearancePatentResultRowDto> byRowKey) throws Exception {
try (Workbook workbook = render(rows, byRowKey)) {
return cell(workbook.getSheet("外观专利检测结果"), 1, COL_BRAND);
}
}
}
@@ -175,6 +175,23 @@ class AsinResultCacheServiceTest {
assertEquals(255, captor.getValue().get(0).getBrand().length());
}
/**
* 缓存行来自客户端回传的采集结果,品牌那格常常是亚马逊详情页署名行原样("Brand: XXX")。
* 这张缓存之后被 AsinCacheHitVo 喂给多个下游任务,入库前不剥一次,脏值就从这里向下游复播。
*/
@Test
@SuppressWarnings({"rawtypes", "unchecked"})
void writeStoresBrandWithoutTheBylineLabel() {
AsinCacheWriteRowDto dto = writeRow("OK", "9.90");
dto.setBrand("Brand: NIKE");
service.write(List.of(dto));
ArgumentCaptor<List<AsinResultCacheEntity>> captor = ArgumentCaptor.forClass(List.class);
verify(mapper).upsertBatch(captor.capture());
assertEquals("NIKE", captor.getValue().get(0).getBrand());
}
@Test
void writeSplitsByBatchSizeAndSurvivesFailingBatch() {
when(mapper.upsertBatch(anyList())).thenReturn(100).thenThrow(new RuntimeException("锁等待超时"));
@@ -2,6 +2,7 @@ package com.nanri.aiimage.modules.collectdata.service;
import com.fasterxml.jackson.databind.ObjectMapper;
import com.nanri.aiimage.modules.collectdata.model.dto.CollectDataSummaryRowDto;
import com.nanri.aiimage.modules.collectdata.model.vo.CollectDataResultRowVo;
import org.apache.poi.ss.usermodel.Sheet;
import org.apache.poi.ss.usermodel.Workbook;
import org.apache.poi.ss.usermodel.WorkbookFactory;
@@ -34,4 +35,30 @@ class CollectDataExcelAssemblyServiceTest {
assertThat(sheet.getRow(1).getCell(6).getNumericCellValue()).isEqualTo(3);
}
}
/**
* 出件这一端的品牌列必须是没有标签前缀的:这份结果文件是用户的下一份上传物,
* 漏剥一次就把脏值送回链路里,清洗的账也随之作废;改口径之前落库的历史明细里也仍带着前缀。
*/
@Test
void stripsAmazonBylinePrefixFromBrandColumn() throws Exception {
File output = tempDir.resolve("collect-data-brand-label.xlsx").toFile();
new CollectDataExcelAssemblyService().writeWorkbook(output,
List.of(row("Brand: Nike", "B000000001"), row("品牌:耐克", "B000000002")),
List.of(), List.of());
try (Workbook workbook = WorkbookFactory.create(output)) {
Sheet detail = workbook.getSheet("采集数据结果");
assertThat(detail.getRow(1).getCell(0).getStringCellValue()).isEqualTo("Nike");
assertThat(detail.getRow(2).getCell(0).getStringCellValue()).isEqualTo("耐克");
}
}
private static CollectDataResultRowVo row(String brand, String asin) {
CollectDataResultRowVo row = new CollectDataResultRowVo();
row.setBrand(brand);
row.setAsin(asin);
return row;
}
}
@@ -5,6 +5,8 @@ import com.baomidou.mybatisplus.core.MybatisConfiguration;
import com.baomidou.mybatisplus.core.conditions.update.LambdaUpdateWrapper;
import com.baomidou.mybatisplus.core.metadata.TableInfoHelper;
import org.apache.ibatis.builder.MapperBuilderAssistant;
import com.nanri.aiimage.modules.collectdata.model.dto.CollectDataSubmitRowDto;
import com.nanri.aiimage.modules.collectdata.model.vo.CollectDataResultRowVo;
import com.nanri.aiimage.modules.collectdata.model.vo.CollectDataDashboardVo;
import com.nanri.aiimage.modules.collectdata.model.vo.CollectDataTaskBatchVo;
import com.nanri.aiimage.modules.task.mapper.FileResultMapper;
@@ -219,6 +221,25 @@ class CollectDataServiceTest {
verify(taskFileJobService, never()).enqueueAssembleResult(any(), any(), any(), any());
}
// 亚马逊详情页把卖家名渲染成 "Brand: XXX",用户复制、客户端就原样回传这种署名值。
// 分片回传是脏值进服务端的唯一入口:品牌检查、结果明细落库、出件读的都是这里产出的 VO,
// 所以标签必须在进模型这一端就被剥掉(口径只在 BrandLabel 一处)。
@Test
void submittedBrandLosesTheAmazonBylinePrefixAtIngest() throws Exception {
CollectDataSubmitRowDto byline = objectMapper.readValue(
"{\"品牌\":\"Brand: NIKE\"}", CollectDataSubmitRowDto.class);
CollectDataSubmitRowDto lookalike = objectMapper.readValue(
"{\"brand_name\":\"Brands: Nike\"}", CollectDataSubmitRowDto.class);
List<CollectDataResultRowVo> rows = ReflectionTestUtils.invokeMethod(
service, "normalizeSubmitRows", List.of(byline, lookalike));
assertThat(rows).hasSize(2);
assertThat(rows.get(0).getBrand()).isEqualTo("NIKE");
// "Brands:" 是真实品牌写法,不是字段标签
assertThat(rows.get(1).getBrand()).isEqualTo("Brands: Nike");
}
private static FileTaskEntity staleTask(long taskId) {
FileTaskEntity task = new FileTaskEntity();
task.setId(taskId);
@@ -93,6 +93,26 @@ class DedupeRunServiceCleanRulesTest {
assertEquals(List.of("7", "B0FFFFFF01", "英国", "9.99", "品牌F", "在售"), readOutputRow(output, 1));
}
/**
* 去重出件的品牌列也必须剥掉亚马逊署名标签前缀:源文件是用户从详情页复制再上传的,
* 结果文件又是他的下一份上传物。源文件列集由用户勾选,所以只能按表头认出「品牌」这一列,
* 其余所选列一律原文透传(这里用同样带前缀的「状态」列钉住不越界)。
* 摘掉 buildCandidateRow 里按表头认出的 strip,下面第一条断言读回来的就是 "Brand: NIKE"。
*/
@Test
@DisplayName("品牌列出件剥掉行首标签前缀,其余所选列原文透传")
void stripsBrandLabelPrefixOnlyFromBrandColumn(@TempDir Path tempDir) throws Exception {
File output = runClean(tempDir, List.of(
new String[]{"1", "B0BRAND001", "英国", "10.00", "FBM", "Brand: NIKE", "Brand: 在售", "2026-09-01"},
new String[]{"2", "B0BRAND002", "英国", "11.00", "FBM", "Brands: NIKE", "在售", "2026-09-01"}),
List.of("id", "ASIN", "国家", "价格", "品牌", "状态"));
assertEquals(List.of("NIKE", "Brands: NIKE"), readOutputColumn(output, 4),
"品牌列只剥行首一个标签,「Brands:」是真实品牌写法");
assertEquals(List.of("Brand: 在售", "在售"), readOutputColumn(output, 5),
"非品牌列不参与清洗");
}
/** 纯规则单测:保留主链接开关打开时,整数行不再受子链接影响。 */
@Test
@DisplayName("开启保留主链接时,整数行不受子链接影响")
@@ -298,6 +298,46 @@ class InvalidAsinDataServiceTest {
return captor.getValue();
}
/**
* 品牌是这张表的匹配键(下游按它判"这个 ASIN 属于被屏蔽品牌")。人工把详情页署名行整段
* 粘进来会存成 "brand: nike",而采集侧现在给 "nike" —— 对不上时屏蔽**静默失效**,不报错,
* 比脏值出件更难发现。摘掉 create 里的 BrandLabel.strip,这两条一条读到 "brand: nike"、
* 一条根本不抛。
*/
@Test
@SuppressWarnings({"rawtypes", "unchecked"})
void createStripsTheBylineLabelFromTheMatchKey() {
InvalidAsinDataCreateRequest request = new InvalidAsinDataCreateRequest();
request.setDataValue("B012345678");
request.setBrand("Brand: Acme");
request.setGroupId(9L);
when(shopManageGroupService.getAccessibleById(9L, 7L, true)).thenReturn(group(9L, "group-a"));
when(invalidAsinDataMapper.selectOne(any())).thenReturn(null);
when(invalidAsinDataMapper.insert(any(InvalidAsinDataEntity.class))).thenAnswer(invocation -> {
invocation.getArgument(0, InvalidAsinDataEntity.class).setId(91L);
return 1;
});
when(invalidAsinDataMapper.selectById(91L)).thenReturn(data(91L, "MANUAL", 9L));
service.create(request, 7L);
ArgumentCaptor<InvalidAsinDataEntity> captor = ArgumentCaptor.forClass(InvalidAsinDataEntity.class);
verify(invalidAsinDataMapper).insert(captor.capture());
assertEquals("acme", captor.getValue().getBrand(), "标签前缀必须在落库前就剥掉,键值才是同一个串");
}
@Test
void createRejectsALabelOnlyBrand() {
// 先洗再判空:洗完全空=用户其实没填品牌,不能让它带着空串进唯一键
InvalidAsinDataCreateRequest request = new InvalidAsinDataCreateRequest();
request.setDataValue("B012345678");
request.setBrand("Brand:");
request.setGroupId(9L);
when(shopManageGroupService.getAccessibleById(9L, 7L, true)).thenReturn(group(9L, "group-a"));
assertThrows(BusinessException.class, () -> service.create(request, 7L));
}
private ShopManageGroupEntity group(Long id, String groupName) {
ShopManageGroupEntity group = new ShopManageGroupEntity();
group.setId(id);
@@ -106,6 +106,31 @@ class ShopDataCrawlRowNormalizerTest {
assertEquals(" B001 ", source.getAsin());
}
/**
* 客户端从亚马逊详情页署名行原样回传,品牌值带着 "Brand: " 标签前缀。
* 读入(copyRow)与比对(sameRow / rowDedupKey)必须用同一个清洗后的字符串:
* 只洗其中一处,跨分片合并时已洗过的存量行与未洗的回传行会算出两个键,同一行重复出件。
* 摘掉 ShopDataCrawlRowNormalizer.brand 里的 strip,下面三条断言全部回退。
*/
@Test
void test_brand_label_prefix_stripped_on_ingest_and_in_dedup_key() {
ShopDataCrawlRowDto byline = row("2026-07-25", "B001", "Brand: NIKE", "img",
"10", "20", "30", "40", "50", "60");
ShopDataCrawlRowDto plain = row("2026-07-25", "B001", "NIKE", "img",
"10", "20", "30", "40", "50", "60");
assertEquals("NIKE", ShopDataCrawlRowNormalizer.copyRow(byline).getBrand(), "回传行进模型即剥标签");
assertTrue(ShopDataCrawlRowNormalizer.sameRow(byline, plain), "署名行与干净行是同一行");
assertEquals(ShopDataCrawlRowNormalizer.rowDedupKey(plain),
ShopDataCrawlRowNormalizer.rowDedupKey(byline), "去重键按剥完标签的品牌计算");
// "Brands:" 是真实品牌写法,不是字段标签
assertEquals("Brands: NIKE",
ShopDataCrawlRowNormalizer.copyRow(row("2026-07-25", "B001", "Brands: NIKE", "img",
"10", "20", "30", "40", "50", "60")).getBrand(),
"近似标签原样保留");
}
// ---- rowEmpty ----
@Test
@@ -128,6 +128,29 @@ class ShopDataCrawlSheetBuilderTest {
}
}
/**
* 出件的品牌末列必须剥掉亚马逊署名标签前缀:这份结果文件是用户的下一份上传物,
* 而改口径之前落库的累计文件快照与 row_payload 里仍带着 "Brand: " 前缀。
* 摘掉 writeDataRowValues 里的 strip,下面第一条断言读回来的就是 "Brand: NIKE"。
*/
@Test
void test_write_data_row_brand_label_prefix_stripped() throws Exception {
try (Workbook wb = new XSSFWorkbook()) {
Sheet sheet = ShopDataCrawlSheetBuilder.createStreamingSheet(wb, 0);
CellStyle[] none = new CellStyle[ShopDataCrawlSheetBuilder.HEADERS.size()];
Row dirty = sheet.createRow(1);
ShopDataCrawlSheetBuilder.writeDataRowValues(dirty,
row("2026-07-25", "B012345678", "Brand: NIKE", "https://img/x.jpg"), none);
Row lookalike = sheet.createRow(2);
ShopDataCrawlSheetBuilder.writeDataRowValues(lookalike,
row("2026-07-25", "B012345679", "Brands: NIKE", "https://img/x.jpg"), none);
assertEquals("NIKE", dirty.getCell(9).getStringCellValue(), "品牌末列剥掉行首标签");
assertEquals("Brands: NIKE", lookalike.getCell(9).getStringCellValue(), "近似标签是真实品牌写法,不剥");
assertEquals("2026-07-25", lookalike.getCell(0).getStringCellValue(), "其余列不参与品牌清洗");
}
}
@Test
void test_write_data_row_null_values_blank_cells() throws Exception {
try (Workbook wb = new XSSFWorkbook()) {
@@ -82,6 +82,9 @@ class ShopDataCrawlSnapshotTest {
private static ShopDataCrawlRowDto sparseRow() {
ShopDataCrawlRowDto row = new ShopDataCrawlRowDto();
row.setAsin("B099999999");
// 首尾空格刻意留着:品牌列出件要过 BrandLabel.strip,它除剥 "Brand:" 标签外还 trim
// (口径自 2026-10-08 就是这两手,Java/Python 正本与 MySQL 清洗脚本同形)。
// golden 里这格是无空格的 "Brand With Spaces" —— 快照改动是这条,不是有人手抖改了基线。
row.setBrand(" Brand With Spaces ");
return row;
}
@@ -12,6 +12,7 @@ import com.nanri.aiimage.modules.shopduplicatecheck.model.payload.DuplicateShop;
import com.nanri.aiimage.modules.shopduplicatecheck.service.ShopDataDuplicateCheckScanService.DuplicateScanView;
import org.junit.jupiter.api.BeforeEach;
import org.junit.jupiter.api.Test;
import org.mockito.ArgumentCaptor;
import java.util.List;
@@ -87,6 +88,37 @@ class ShopDataDuplicateCheckImportServiceTest {
verify(scanMapper).insert(any(ShopDataDuplicateScanEntity.class));
}
/** 基线 payload:郭悦晴那行的品牌带着改口径之前的标签前缀(这一店不在本次导入的店铺里,会被原样重插)。 */
private DuplicateScanView legacyBrandBaselineView() {
DuplicateOccurrence kept = new DuplicateOccurrence("B09PNTZM4L", "2026-08-19 05:48", "10.1",
"Brand: HANYTON", "郭悦晴", "G2", List.of("DE"), "DE");
DuplicateShop shop = new DuplicateShop("郭悦晴", "G2", List.of("DE"), 1, 1);
return new DuplicateScanView("2026-09-04 00:00:00", null, new DuplicateScanPayload(
List.of(shop), List.of(new DuplicateItem("B09PNTZM4L", 1, 1, List.of(kept)))));
}
/**
* 导入的基线行读自上一份落库 payload,改口径之前它带着 "Brand: " 前缀;原样重插就是替用户把
* 脏值写回扫描表一次(下载再上传同构的问题)。摘掉 applyToLatest 里的 strip,下面这条会红。
*/
@Test
void addImportDoesNotReseedLegacyBrandLabelIntoNewScanRow() throws Exception {
when(scanService.loadLatest()).thenReturn(legacyBrandBaselineView());
when(scanMapper.insert(any(ShopDataDuplicateScanEntity.class))).thenAnswer(invocation -> {
invocation.<ShopDataDuplicateScanEntity>getArgument(0).setId(99L);
return 1;
});
service.importExcel("小曾", "UK", workbook(), false);
ArgumentCaptor<ShopDataDuplicateScanEntity> inserted =
ArgumentCaptor.forClass(ShopDataDuplicateScanEntity.class);
verify(scanMapper).insert(inserted.capture());
assertThat(inserted.getValue().getPayloadJson())
.doesNotContain("Brand:")
.contains("HANYTON");
}
@Test
void importWithoutBaselineFails() {
when(scanService.loadLatest()).thenReturn(null);
@@ -4,13 +4,18 @@ import com.nanri.aiimage.modules.shopduplicatecheck.TestFixtures;
import com.nanri.aiimage.modules.shopduplicatecheck.mapper.ShopDuplicateCheckSourceMapper;
import com.nanri.aiimage.modules.shopduplicatecheck.model.payload.DuplicateItem;
import com.nanri.aiimage.modules.shopduplicatecheck.model.payload.DuplicateOccurrence;
import com.nanri.aiimage.modules.shopduplicatecheck.model.payload.DuplicateScanPayload;
import com.nanri.aiimage.modules.shopduplicatecheck.model.payload.DuplicateScanSummary;
import com.nanri.aiimage.modules.shopduplicatecheck.model.payload.DuplicateShop;
import org.apache.poi.ss.usermodel.Sheet;
import org.apache.poi.xssf.usermodel.XSSFWorkbook;
import org.junit.jupiter.api.BeforeEach;
import org.junit.jupiter.api.Test;
import java.io.ByteArrayInputStream;
import java.io.ByteArrayOutputStream;
import java.nio.charset.StandardCharsets;
import java.util.ArrayList;
import java.util.List;
import java.util.Map;
import java.util.Set;
@@ -265,6 +270,39 @@ class ShopDataDuplicateCheckQueryServiceTest {
assertThat(lines[1]).doesNotContain(",UK,");
}
/**
* 导出文件的品牌列必须是剥过标签前缀的:扫描行是改口径之前落库的,payload 里的品牌仍带
* "Brand: "(亚马逊详情页复制来的署名),而出件是用户的下一份上传物——这一格不洗,
* 下载再上传就把脏值送回链路里。摘掉 buildExportRows 里的 strip,这一格读回来就是 "Brand: NIKE"。
*/
@Test
void exportBrandColumnLosesTheAmazonBylinePrefix() throws Exception {
ShopDataDuplicateCheckScanService.DuplicateScanView legacyPayload =
new ShopDataDuplicateCheckScanService.DuplicateScanView(
"2026-09-04 03:10:00", null, new DuplicateScanPayload(List.of(), List.of(
new DuplicateItem("B0BRAND01", 2, 2, List.of(
legacyOccurrence("ShopA", "Brand: NIKE"),
legacyOccurrence("ShopB", "Brands: NIKE"))))));
ByteArrayOutputStream out = new ByteArrayOutputStream();
service.writeExportXlsx(legacyPayload, null, "all", noFilter(), out);
List<String> brands = new ArrayList<>();
try (XSSFWorkbook workbook = new XSSFWorkbook(new ByteArrayInputStream(out.toByteArray()))) {
Sheet sheet = workbook.getSheetAt(0);
for (int r = 1; r <= sheet.getLastRowNum(); r++) {
brands.add(sheet.getRow(r).getCell(7).getStringCellValue());
}
}
// "Brands:" 是真实品牌写法,只剥行首的那个标签
assertThat(brands).containsExactlyInAnyOrder("NIKE", "Brands: NIKE");
}
private static DuplicateOccurrence legacyOccurrence(String shopName, String brand) {
return new DuplicateOccurrence("B0BRAND01", "2026-08-30", "GBP 9.99", brand,
shopName, "GroupA", List.of("UK"), "UK");
}
@Test
void groupFilterNarrowsToGroupMembers() {
// 分组筛选:GroupA 只保留 ShopA/ShopB 键。
@@ -49,4 +49,22 @@ class ShopDataDuplicateCheckScanServiceTest {
void buildParsedShops_emptyInputReturnsEmpty() {
assertTrue(ShopDataDuplicateCheckScanService.buildParsedShops(List.of(), Map.of()).isEmpty());
}
/**
* 明细表的品牌是改口径之前落库的(采集端原样写入亚马逊署名行),扫描直接查这张表:
* 标签前缀要在进聚合这一端剥掉,落库 payload、台账页面与出件拿到的才是同一个干净值。
* 摘掉 buildParsedShops 里的 strip,下面第一条断言读回来的就是 "Brand: NIKE"。
*/
@Test
void buildParsedShops_stripsAmazonBylinePrefixFromDbBrandColumn() {
List<ShopParsed> parsed = ShopDataDuplicateCheckScanService.buildParsedShops(
List.of(new ShopLatestItemRowDto("店A", "UK", "B0001", "2026-09-01 10:00", "9.90", "Brand: NIKE"),
new ShopLatestItemRowDto("店A", "UK", "B0002", "2026-09-01 10:00", "9.90", "品牌:耐克"),
new ShopLatestItemRowDto("店A", "UK", "B0003", "2026-09-01 10:00", "9.90", "Brands: NIKE")),
Map.of());
assertEquals(List.of("NIKE", "耐克", "Brands: NIKE"),
parsed.get(0).rows().stream().map(RawRow::brand).toList(),
"行首标签剥掉,「Brands:」这类真实品牌写法原样保留");
}
}
@@ -75,4 +75,29 @@ class DuplicateCheckWorkbookParserTest {
new String[]{"2026-08-15", "9.99"});
assertEquals(0, DuplicateCheckWorkbookParser.parse(bytes).size(), "无 ASIN 列的 sheet 跳过");
}
/**
* 品牌列读入即剥标签前缀:用户上传的结果文件是亚马逊详情页复制来的,"Brand: XXX" 带着标签
* 进聚合就一路带到 payload、台账与出件。摘掉解析这一端的 strip,这里读到的就是原始署名值。
*/
@Test
void stripsAmazonBylinePrefixFromBrandColumn() throws Exception {
byte[] bytes = workbookWith(false, "英国",
new String[]{"ASIN", "日期", "价格", "品牌"},
new String[]{"B0BRAND01", "2026-08-15", "£34.46", "Brand: NIKE"});
assertEquals("NIKE", DuplicateCheckWorkbookParser.parse(bytes).get(0).brand(),
"行首标签在读入这一端剥掉");
}
/** "Brands:" 是真实品牌写法,不是字段标签;同一份文件里的价格/日期列不受品牌清洗影响。 */
@Test
void keepsLookalikeBrandAndOtherColumnsVerbatim() throws Exception {
byte[] bytes = workbookWith(false, "英国",
new String[]{"ASIN", "日期", "价格", "品牌"},
new String[]{"B0BRAND02", "Brand: 2026-08-15", "£Brand: 34.46", "Brands: NIKE"});
var row = DuplicateCheckWorkbookParser.parse(bytes).get(0);
assertEquals("Brands: NIKE", row.brand(), "近似标签原样保留");
assertEquals("Brand: 2026-08-15", row.date(), "非品牌列不参与清洗");
assertEquals("£Brand: 34.46", row.price(), "非品牌列不参与清洗");
}
}
@@ -209,6 +209,21 @@ class SimilarAsinSheetBuilderTest {
assertEquals("品牌Y", cellString(workbook, 1, 5), "品牌按包含匹配");
}
@Test
void test_build_brand_label_prefix_stripped() {
// 出件是用户的下一份上传物:亚马逊页面复制带上的 "Brand: " 前缀不在此剥掉,二次上传就把脏值送回链路
Workbook workbook = build(List.of(row("1", "B01A", "US", "19.9", "卖家1", "Brand: NIKE")));
assertEquals("NIKE", cellString(workbook, 1, 5), "品牌列剥掉行首标签前缀");
}
@Test
void test_build_brand_lookalike_kept() {
Workbook workbook = build(List.of(row("1", "B01A", "US", "19.9", "卖家1", "Brands: NIKE")));
assertEquals("Brands: NIKE", cellString(workbook, 1, 5), "近似标签是真实品牌写法,不随行首清洗被误剥");
}
@Test
void test_build_null_input() {
Workbook workbook = build(null);
@@ -1,3 +1,3 @@
rows=2
日期=2026-07-25;ASIN=B012345678;商品图片=;库存销量=128;销售排名=#1,245;页面浏览量=3560;售出件数=42;价格=GBP 19.99;推荐报价=GBP 18.99;品牌=Example Brand;
日期=;ASIN=B099999999;商品图片=;库存销量=;销售排名=;页面浏览量=;售出件数=;价格=;推荐报价=;品牌= Brand With Spaces ;
日期=;ASIN=B099999999;商品图片=;库存销量=;销售排名=;页面浏览量=;售出件数=;价格=;推荐报价=;品牌=Brand With Spaces;