[#44] 위험한 branch 조합의 병합 코드 문맥을 추출한다

.gitignore

@@ -10,3 +10,6 @@ dist/
 
 # macOS
 .DS_Store
+
+# AI
+superpowers/

src/git/gitCleanCodeContextCollector.ts

@@ -0,0 +1,178 @@
+import {
+  collectPreparedEvidence,
+  type PreparedCodeContextEvidence
+} from "./gitCodeContextEvidenceBuilder.js"
+import {
+  changedFiles,
+  diffHunks,
+  pairCodeContext
+} from "./gitCodeContextReader.js"
+import {
+  boundedContextRange,
+  functionRangeFor,
+  newRange,
+  oldRange,
+  overlapRegions,
+  unionRange
+} from "./mergeCodeContextParser.js"
+import type {
+  GitDiffHunk,
+  GitMergeTreePairResult,
+  MergeCodeContextEvidence,
+  MergeCodeContextLineRange
+} from "./types.js"
+
+// clean 조합에서 양쪽 exact hunk가 merge-base 좌표로 겹치는 문맥만 수집
+export async function collectCleanOverlapEvidence(
+  mergeResult: GitMergeTreePairResult,
+  pairIndex: number,
+  repositoryPath: string
+): Promise<MergeCodeContextEvidence[]> {
+  const context = await pairCodeContext(mergeResult, repositoryPath)
+  const [leftFiles, rightFiles] = await Promise.all([
+    changedFiles(repositoryPath, context.mergeBaseOid, context.leftOid),
+    changedFiles(repositoryPath, context.mergeBaseOid, context.rightOid)
+  ])
+  const rightFileSet = new Set(rightFiles)
+  const commonFiles = leftFiles
+    .filter(filePath => rightFileSet.has(filePath))
+    .sort(compareText)
+  const prepared: PreparedCodeContextEvidence[] = []
+
+  for (const [fileIndex, filePath] of commonFiles.entries()) {
+    const [leftHunks, rightHunks] = await Promise.all([
+      diffHunks(
+        repositoryPath,
+        context.mergeBaseOid,
+        context.leftOid,
+        filePath,
+        "exact"
+      ),
+      diffHunks(
+        repositoryPath,
+        context.mergeBaseOid,
+        context.rightOid,
+        filePath,
+        "exact"
+      )
+    ])
+    const regions = overlapRegions(filePath, leftHunks, rightHunks)
+
+    if (!regions.length) {
+      continue
+    }
+
+    const [
+      mergedHunks,
+      leftFunctionHunks,
+      rightFunctionHunks,
+      mergedFunctionHunks
+    ] = await Promise.all([
+      diffHunks(
+        repositoryPath,
+        context.mergeBaseOid,
+        context.mergedTreeOid,
+        filePath,
+        "exact"
+      ),
+      diffHunks(
+        repositoryPath,
+        context.mergeBaseOid,
+        context.leftOid,
+        filePath,
+        "function"
+      ),
+      diffHunks(
+        repositoryPath,
+        context.mergeBaseOid,
+        context.rightOid,
+        filePath,
+        "function"
+      ),
+      diffHunks(
+        repositoryPath,
+        context.mergeBaseOid,
+        context.mergedTreeOid,
+        filePath,
+        "function"
+      )
+    ])
+
+    for (const [regionIndex, region] of regions.entries()) {
+      const mergedTarget = mappedNewRangeFor(
+        mergedHunks,
+        region.baseRange
+      ) ?? unionRange(region.leftRange, region.rightRange)
+      const key = `${pairIndex}:${fileIndex}:${regionIndex}`
+
+      prepared.push({
+        filePath,
+        baseRange: boundedContextRange(
+          region.baseRange,
+          functionRangeFor(
+            [...leftFunctionHunks, ...rightFunctionHunks],
+            "old",
+            region.baseRange
+          )
+        ),
+        leftRange: boundedContextRange(
+          region.leftRange,
+          functionRangeFor(leftFunctionHunks, "new", region.leftRange)
+        ),
+        rightRange: boundedContextRange(
+          region.rightRange,
+          functionRangeFor(rightFunctionHunks, "new", region.rightRange)
+        ),
+        mergedRange: boundedContextRange(
+          mergedTarget,
+          functionRangeFor(mergedFunctionHunks, "new", mergedTarget)
+        ),
+        baseKey: `${key}:base`,
+        leftKey: `${key}:left`,
+        rightKey: `${key}:right`,
+        mergedKey: `${key}:merged`
+      })
+    }
+  }
+
+  return collectPreparedEvidence(
+    mergeResult,
+    "clean_hunk_overlap",
+    context,
+    prepared,
+    repositoryPath
+  )
+}
+
+// base 좌표와 겹치는 merged hunk의 변경 후 좌표를 선택
+function mappedNewRangeFor(
+  hunks: GitDiffHunk[],
+  baseRange: MergeCodeContextLineRange
+): MergeCodeContextLineRange | undefined {
+  return hunks
+    .map(hunk => ({
+      oldRange: oldRange(hunk),
+      newRange: newRange(hunk)
+    }))
+    .filter(item =>
+      item.oldRange !== undefined &&
+      item.newRange !== undefined &&
+      Math.max(item.oldRange.startLine, baseRange.startLine) <=
+        Math.min(item.oldRange.endLine, baseRange.endLine)
+    )
+    .sort((item, other) => {
+      const range = item.oldRange!
+      const otherRange = other.oldRange!
+
+      return range.endLine - range.startLine -
+        (otherRange.endLine - otherRange.startLine)
+    })[0]?.newRange
+}
+
+function compareText(text: string, other: string): number {
+  if (text === other) {
+    return 0
+  }
+
+  return text < other ? -1 : 1
+}

src/git/gitCodeContextEvidenceBuilder.ts

@@ -0,0 +1,156 @@
+import { readGitObjectSnippets } from "./gitObjectBatchReader.js"
+import type { PairCodeContext } from "./gitCodeContextReader.js"
+import type {
+  GitMergeTreePairResult,
+  GitObjectSnippetRequest,
+  MergeCodeContextEvidence,
+  MergeCodeContextKind,
+  MergeCodeContextLineRange,
+  MergeCodeContextSnippet
+} from "./types.js"
+
+export type PreparedCodeContextEvidence = {
+  filePath: string
+  baseRange: MergeCodeContextLineRange
+  leftRange: MergeCodeContextLineRange
+  rightRange: MergeCodeContextLineRange
+  mergedRange: MergeCodeContextLineRange
+  baseKey: string
+  leftKey: string
+  rightKey: string
+  mergedKey: string
+}
+
+// 준비된 네 version 요청을 evidence model로 변환
+export async function collectPreparedEvidence(
+  mergeResult: GitMergeTreePairResult,
+  kind: MergeCodeContextKind,
+  context: PairCodeContext,
+  prepared: PreparedCodeContextEvidence[],
+  repositoryPath: string
+): Promise<MergeCodeContextEvidence[]> {
+  const snippets = await readGitObjectSnippets(
+    snippetRequests(
+      prepared,
+      context.mergeBaseOid,
+      context.leftOid,
+      context.rightOid,
+      context.mergedTreeOid
+    ),
+    { repositoryPath }
+  )
+
+  return prepared.map(item => {
+    const baseSnippet = requiredSnippet(snippets, item.baseKey)
+    const leftSnippet = deletedVersionSnippet(
+      baseSnippet,
+      requiredSnippet(snippets, item.leftKey)
+    )
+    const rightSnippet = deletedVersionSnippet(
+      baseSnippet,
+      requiredSnippet(snippets, item.rightKey)
+    )
+    const mergedSnippet = deletedMergedSnippet(
+      baseSnippet,
+      leftSnippet,
+      rightSnippet,
+      requiredSnippet(snippets, item.mergedKey)
+    )
+
+    return {
+      pair: mergeResult.pair,
+      kind,
+      filePath: item.filePath,
+      mergeBaseOid: context.mergeBaseOid,
+      mergedTreeOid: context.mergedTreeOid,
+      baseCommit: context.baseCommit,
+      leftCommit: context.leftCommit,
+      rightCommit: context.rightCommit,
+      baseSnippet,
+      leftSnippet,
+      rightSnippet,
+      mergedSnippet
+    }
+  })
+}
+
+// 준비된 evidence를 Git object batch 요청 네 개로 변환
+function snippetRequests(
+  prepared: PreparedCodeContextEvidence[],
+  mergeBaseOid: string,
+  leftOid: string,
+  rightOid: string,
+  mergedTreeOid: string
+): GitObjectSnippetRequest[] {
+  return prepared.flatMap(item => [{
+    key: item.baseKey,
+    objectOid: mergeBaseOid,
+    filePath: item.filePath,
+    range: item.baseRange
+  }, {
+    key: item.leftKey,
+    objectOid: leftOid,
+    filePath: item.filePath,
+    range: item.leftRange
+  }, {
+    key: item.rightKey,
+    objectOid: rightOid,
+    filePath: item.filePath,
+    range: item.rightRange
+  }, {
+    key: item.mergedKey,
+    objectOid: mergedTreeOid,
+    filePath: item.filePath,
+    range: item.mergedRange
+  }])
+}
+
+function requiredSnippet(
+  snippets: ReadonlyMap<string, MergeCodeContextSnippet>,
+  key: string
+): MergeCodeContextSnippet {
+  const snippet = snippets.get(key)
+
+  if (!snippet) {
+    throw new Error(`Missing Git object snippet ${key}`)
+  }
+
+  return snippet
+}
+
+// base에 존재하던 file이 side object에 없으면 deleted로 분류
+function deletedVersionSnippet(
+  baseSnippet: MergeCodeContextSnippet,
+  snippet: MergeCodeContextSnippet
+): MergeCodeContextSnippet {
+  if (snippet.status !== "missing" || baseSnippet.status === "missing") {
+    return snippet
+  }
+
+  return {
+    ...snippet,
+    status: "deleted"
+  }
+}
+
+// 어느 version에든 존재한 file이 merge tree에서 없으면 deleted로 분류
+function deletedMergedSnippet(
+  baseSnippet: MergeCodeContextSnippet,
+  leftSnippet: MergeCodeContextSnippet,
+  rightSnippet: MergeCodeContextSnippet,
+  mergedSnippet: MergeCodeContextSnippet
+): MergeCodeContextSnippet {
+  if (
+    mergedSnippet.status !== "missing" ||
+    [baseSnippet, leftSnippet, rightSnippet].every(
+      snippet => snippet.status === "missing"
+    )
+  ) {
+    return mergedSnippet
+  }
+
+  return {
+    ...mergedSnippet,
+    status: "deleted"
+  }
+}