feat(rag): 增加文档切片工厂基础层

src/main/java/com/bruce/rag/parse/Chunker.java

@@ -0,0 +1,26 @@
+package com.bruce.rag.parse;
+
+import com.bruce.rag.entity.RagChunk;
+import com.bruce.rag.enums.RagChunkStrategyEnum;
+
+import java.util.List;
+
+/**
+ * 切片策略统一接口。
+ * <p>
+ * 这里的职责只有两个：
+ * 1. 告诉工厂自己支持哪一种切片策略
+ * 2. 根据切片命令生成切片结果
+ */
+public interface Chunker {
+
+    /**
+     * 判断当前实现是否支持指定的切片策略。
+     */
+    boolean supports(RagChunkStrategyEnum strategy);
+
+    /**
+     * 执行切片，返回内存中的切片对象列表。
+     */
+    List<RagChunk> chunk(RagChunkCommand command);
+}

src/main/java/com/bruce/rag/parse/ChunkerFactory.java

@@ -0,0 +1,32 @@
+package com.bruce.rag.parse;
+
+import com.bruce.rag.enums.RagChunkStrategyEnum;
+import org.springframework.stereotype.Component;
+
+import java.util.List;
+
+@Component
+/**
+ * 切片策略工厂。
+ * <p>
+ * Spring 会把所有实现了 {@link Chunker} 的 Bean 注入进来，
+ * 工厂再根据切片策略挑出对应实现，避免业务层自己写 if-else 或 switch。
+ */
+public class ChunkerFactory {
+
+    private final List<Chunker> chunkers;
+
+    public ChunkerFactory(List<Chunker> chunkers) {
+        this.chunkers = chunkers;
+    }
+
+    /**
+     * 根据切片策略解析出具体的切片器实现。
+     */
+    public Chunker resolve(RagChunkStrategyEnum strategy) {
+        return chunkers.stream()
+                .filter(chunker -> chunker.supports(strategy))
+                .findFirst()
+                .orElseThrow(() -> new IllegalArgumentException("不支持的切片方式: " + strategy));
+    }
+}

src/main/java/com/bruce/rag/parse/RagChunkCommand.java

@@ -0,0 +1,45 @@
+package com.bruce.rag.parse;
+
+import com.bruce.common.document.parse.DocumentParseResult;
+import com.bruce.rag.entity.RagDocument;
+import lombok.Data;
+
+@Data
+/**
+ * 一次切片请求的上下文参数。
+ * <p>
+ * 这里把文档信息、解析结果和切片配置收拢到一个对象里，
+ * 这样切片器接口不会因为参数越来越多而变得难维护。
+ */
+public class RagChunkCommand {
+
+    /**
+     * 当前要切片的文档实体。
+     */
+    private RagDocument document;
+
+    /**
+     * 文档解析后的文本结果。
+     */
+    private DocumentParseResult parseResult;
+
+    /**
+     * 切片策略枚举值，通常来自前端请求。
+     */
+    private Integer chunkStrategy;
+
+    /**
+     * 单个切片的目标长度，主要给定长切片使用。
+     */
+    private Integer chunkSize;
+
+    /**
+     * 相邻切片之间的重叠长度，主要给定长切片使用。
+     */
+    private Integer chunkOverlap;
+
+    /**
+     * 自定义分隔符，主要给分隔符切片使用。
+     */
+    private String delimiter;
+}

src/main/java/com/bruce/rag/parse/impl/DelimiterChunker.java

@@ -0,0 +1,68 @@
+package com.bruce.rag.parse.impl;
+
+import com.bruce.common.document.parse.DocumentParseResult;
+import com.bruce.rag.entity.RagChunk;
+import com.bruce.rag.entity.RagDocument;
+import com.bruce.rag.enums.RagChunkStrategyEnum;
+import com.bruce.rag.parse.Chunker;
+import com.bruce.rag.parse.RagChunkCommand;
+import org.springframework.stereotype.Component;
+import org.springframework.util.StringUtils;
+
+import java.util.ArrayList;
+import java.util.List;
+import java.util.regex.Pattern;
+
+@Component
+/**
+ * 分隔符切片实现。
+ * <p>
+ * 先按外部传入的 delimiter 拆分文本，再过滤空片段，生成顺序切片。
+ */
+public class DelimiterChunker implements Chunker {
+
+    @Override
+    public boolean supports(RagChunkStrategyEnum strategy) {
+        return RagChunkStrategyEnum.DELIMITER == strategy;
+    }
+
+    @Override
+    public List<RagChunk> chunk(RagChunkCommand command) {
+        String text = extractText(command);
+        String delimiter = command == null ? null : command.getDelimiter();
+        if (!StringUtils.hasText(text) || !StringUtils.hasText(delimiter)) {
+            return List.of();
+        }
+
+        // 使用 Pattern.quote 处理正则特殊字符，确保分隔符按字面值切分。
+        String[] parts = text.split(Pattern.quote(delimiter));
+        List<RagChunk> chunks = new ArrayList<>();
+        for (String part : parts) {
+            if (!StringUtils.hasText(part)) {
+                continue;
+            }
+            chunks.add(buildChunk(command.getDocument(), chunks.size(), part.trim()));
+        }
+        return chunks;
+    }
+
+    private String extractText(RagChunkCommand command) {
+        DocumentParseResult parseResult = command == null ? null : command.getParseResult();
+        return parseResult == null ? null : parseResult.getText();
+    }
+
+    /**
+     * 分隔符切片同样只负责生成基础切片结构，不处理持久化和向量化。
+     */
+    private RagChunk buildChunk(RagDocument document, int index, String content) {
+        RagChunk chunk = new RagChunk();
+        if (document != null) {
+            chunk.setStoreId(document.getStoreId());
+            chunk.setDocumentId(document.getId());
+        }
+        chunk.setChunkIndex(index);
+        chunk.setChunkContent(content);
+        chunk.setEnabled(Boolean.TRUE);
+        return chunk;
+    }
+}

src/main/java/com/bruce/rag/parse/impl/FixedLengthChunker.java

@@ -0,0 +1,91 @@
+package com.bruce.rag.parse.impl;
+
+import com.bruce.common.document.parse.DocumentParseResult;
+import com.bruce.rag.entity.RagChunk;
+import com.bruce.rag.entity.RagDocument;
+import com.bruce.rag.enums.RagChunkStrategyEnum;
+import com.bruce.rag.parse.Chunker;
+import com.bruce.rag.parse.RagChunkCommand;
+import org.springframework.stereotype.Component;
+import org.springframework.util.StringUtils;
+
+import java.util.ArrayList;
+import java.util.List;
+
+@Component
+/**
+ * 定长切片实现。
+ * <p>
+ * 按 chunkSize 顺序截取文本，并结合 chunkOverlap 控制相邻切片的重叠部分。
+ */
+public class FixedLengthChunker implements Chunker {
+
+    @Override
+    public boolean supports(RagChunkStrategyEnum strategy) {
+        return RagChunkStrategyEnum.FIXED_LENGTH == strategy;
+    }
+
+    @Override
+    public List<RagChunk> chunk(RagChunkCommand command) {
+        String text = extractText(command);
+        if (!StringUtils.hasText(text)) {
+            return List.of();
+        }
+
+        int chunkSize = resolveChunkSize(command, text.length());
+        int overlap = resolveChunkOverlap(command, chunkSize);
+        // 实际步长等于切片长度减去重叠长度，最小保证为 1，避免死循环。
+        int step = Math.max(1, chunkSize - overlap);
+        List<RagChunk> chunks = new ArrayList<>();
+        for (int start = 0; start < text.length(); start += step) {
+            int end = Math.min(text.length(), start + chunkSize);
+            chunks.add(buildChunk(command.getDocument(), chunks.size(), text.substring(start, end)));
+            if (end >= text.length()) {
+                break;
+            }
+        }
+        return chunks;
+    }
+
+    private String extractText(RagChunkCommand command) {
+        DocumentParseResult parseResult = command == null ? null : command.getParseResult();
+        return parseResult == null ? null : parseResult.getText();
+    }
+
+    /**
+     * 当未传 chunkSize 或传入非法值时，退化为整段文本一个切片。
+     */
+    private int resolveChunkSize(RagChunkCommand command, int textLength) {
+        Integer chunkSize = command == null ? null : command.getChunkSize();
+        if (chunkSize == null || chunkSize <= 0) {
+            return textLength;
+        }
+        return chunkSize;
+    }
+
+    /**
+     * overlap 不能为负，也不能大于等于 chunkSize，否则步长会变成 0 或负数。
+     */
+    private int resolveChunkOverlap(RagChunkCommand command, int chunkSize) {
+        Integer overlap = command == null ? null : command.getChunkOverlap();
+        if (overlap == null || overlap < 0) {
+            return 0;
+        }
+        return Math.min(overlap, Math.max(0, chunkSize - 1));
+    }
+
+    /**
+     * 这里只构造最基础的切片对象，后续落库时再补充摘要、向量等扩展字段。
+     */
+    private RagChunk buildChunk(RagDocument document, int index, String content) {
+        RagChunk chunk = new RagChunk();
+        if (document != null) {
+            chunk.setStoreId(document.getStoreId());
+            chunk.setDocumentId(document.getId());
+        }
+        chunk.setChunkIndex(index);
+        chunk.setChunkContent(content);
+        chunk.setEnabled(Boolean.TRUE);
+        return chunk;
+    }
+}

src/test/java/com/bruce/rag/parse/ChunkerFactoryTests.java

@@ -0,0 +1,50 @@
+package com.bruce.rag.parse;
+
+import com.bruce.rag.entity.RagChunk;
+import com.bruce.rag.enums.RagChunkStrategyEnum;
+import org.junit.jupiter.api.Test;
+
+import java.util.List;
+
+import static org.junit.jupiter.api.Assertions.assertSame;
+import static org.junit.jupiter.api.Assertions.assertThrows;
+
+class ChunkerFactoryTests {
+
+    @Test
+    void resolveShouldReturnMatchingChunker() {
+        Chunker supported = new StubChunker(RagChunkStrategyEnum.FIXED_LENGTH);
+        Chunker unsupported = new StubChunker(RagChunkStrategyEnum.DELIMITER);
+        ChunkerFactory factory = new ChunkerFactory(List.of(supported, unsupported));
+
+        Chunker resolved = factory.resolve(RagChunkStrategyEnum.FIXED_LENGTH);
+
+        assertSame(supported, resolved);
+    }
+
+    @Test
+    void resolveShouldRejectUnsupportedStrategy() {
+        ChunkerFactory factory = new ChunkerFactory(List.of(new StubChunker(RagChunkStrategyEnum.FIXED_LENGTH)));
+
+        assertThrows(IllegalArgumentException.class, () -> factory.resolve(RagChunkStrategyEnum.SEMANTIC));
+    }
+
+    private static class StubChunker implements Chunker {
+
+        private final RagChunkStrategyEnum strategy;
+
+        private StubChunker(RagChunkStrategyEnum strategy) {
+            this.strategy = strategy;
+        }
+
+        @Override
+        public boolean supports(RagChunkStrategyEnum strategy) {
+            return this.strategy == strategy;
+        }
+
+        @Override
+        public List<RagChunk> chunk(RagChunkCommand command) {
+            return List.of();
+        }
+    }
+}

src/test/java/com/bruce/rag/parse/DelimiterChunkerTests.java

@@ -0,0 +1,64 @@
+package com.bruce.rag.parse;
+
+import com.bruce.common.document.parse.DocumentParseResult;
+import com.bruce.rag.entity.RagChunk;
+import com.bruce.rag.entity.RagDocument;
+import com.bruce.rag.enums.RagChunkStrategyEnum;
+import com.bruce.rag.parse.impl.DelimiterChunker;
+import org.junit.jupiter.api.Test;
+
+import java.util.List;
+
+import static org.junit.jupiter.api.Assertions.assertEquals;
+import static org.junit.jupiter.api.Assertions.assertTrue;
+
+class DelimiterChunkerTests {
+
+    @Test
+    void chunkShouldSplitByDelimiterAndIgnoreBlankSegments() {
+        DelimiterChunker chunker = new DelimiterChunker();
+
+        RagChunkCommand command = new RagChunkCommand();
+        command.setDocument(buildDocument());
+        command.setParseResult(buildParseResult("第一段。第二段。。第三段"));
+        command.setChunkStrategy(RagChunkStrategyEnum.DELIMITER.getValue());
+        command.setDelimiter("。");
+
+        List<RagChunk> chunks = chunker.chunk(command);
+
+        assertEquals(3, chunks.size());
+        assertEquals("第一段", chunks.get(0).getChunkContent());
+        assertEquals("第二段", chunks.get(1).getChunkContent());
+        assertEquals("第三段", chunks.get(2).getChunkContent());
+        assertEquals(0, chunks.get(0).getChunkIndex());
+        assertEquals(1, chunks.get(1).getChunkIndex());
+        assertEquals(2, chunks.get(2).getChunkIndex());
+    }
+
+    @Test
+    void chunkShouldReturnEmptyListForBlankText() {
+        DelimiterChunker chunker = new DelimiterChunker();
+
+        RagChunkCommand command = new RagChunkCommand();
+        command.setDocument(buildDocument());
+        command.setParseResult(buildParseResult(" "));
+        command.setChunkStrategy(RagChunkStrategyEnum.DELIMITER.getValue());
+        command.setDelimiter("。");
+
+        assertTrue(chunker.chunk(command).isEmpty());
+    }
+
+    private static RagDocument buildDocument() {
+        RagDocument document = new RagDocument();
+        document.setId(66L);
+        document.setStoreId(55L);
+        return document;
+    }
+
+    private static DocumentParseResult buildParseResult(String text) {
+        DocumentParseResult result = new DocumentParseResult();
+        result.setText(text);
+        result.setTextLength(text.length());
+        return result;
+    }
+}

src/test/java/com/bruce/rag/parse/FixedLengthChunkerTests.java

@@ -0,0 +1,69 @@
+package com.bruce.rag.parse;
+
+import com.bruce.common.document.parse.DocumentParseResult;
+import com.bruce.rag.entity.RagChunk;
+import com.bruce.rag.entity.RagDocument;
+import com.bruce.rag.enums.RagChunkStrategyEnum;
+import com.bruce.rag.parse.impl.FixedLengthChunker;
+import org.junit.jupiter.api.Test;
+
+import java.util.List;
+
+import static org.junit.jupiter.api.Assertions.assertEquals;
+import static org.junit.jupiter.api.Assertions.assertTrue;
+
+class FixedLengthChunkerTests {
+
+    @Test
+    void chunkShouldSplitTextByChunkSizeAndOverlap() {
+        FixedLengthChunker chunker = new FixedLengthChunker();
+
+        RagChunkCommand command = new RagChunkCommand();
+        command.setDocument(buildDocument());
+        command.setParseResult(buildParseResult("abcdefghij"));
+        command.setChunkStrategy(RagChunkStrategyEnum.FIXED_LENGTH.getValue());
+        command.setChunkSize(4);
+        command.setChunkOverlap(1);
+
+        List<RagChunk> chunks = chunker.chunk(command);
+
+        assertEquals(3, chunks.size());
+        assertEquals("abcd", chunks.get(0).getChunkContent());
+        assertEquals("defg", chunks.get(1).getChunkContent());
+        assertEquals("ghij", chunks.get(2).getChunkContent());
+        assertEquals(0, chunks.get(0).getChunkIndex());
+        assertEquals(1, chunks.get(1).getChunkIndex());
+        assertEquals(2, chunks.get(2).getChunkIndex());
+        assertEquals(99L, chunks.get(0).getDocumentId());
+        assertEquals(88L, chunks.get(0).getStoreId());
+        assertTrue(Boolean.TRUE.equals(chunks.get(0).getEnabled()));
+    }
+
+    @Test
+    void chunkShouldReturnEmptyListForBlankText() {
+        FixedLengthChunker chunker = new FixedLengthChunker();
+
+        RagChunkCommand command = new RagChunkCommand();
+        command.setDocument(buildDocument());
+        command.setParseResult(buildParseResult("   "));
+        command.setChunkStrategy(RagChunkStrategyEnum.FIXED_LENGTH.getValue());
+        command.setChunkSize(4);
+        command.setChunkOverlap(1);
+
+        assertTrue(chunker.chunk(command).isEmpty());
+    }
+
+    private static RagDocument buildDocument() {
+        RagDocument document = new RagDocument();
+        document.setId(99L);
+        document.setStoreId(88L);
+        return document;
+    }
+
+    private static DocumentParseResult buildParseResult(String text) {
+        DocumentParseResult result = new DocumentParseResult();
+        result.setText(text);
+        result.setTextLength(text.length());
+        return result;
+    }
+}