opencode-codebase-index 0.24.0 → 0.25.0

package/dist/pi-extension.js

@@ -2043,6 +2043,26 @@ function formatCostEstimate(estimate) {
 \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518
 `;
 }
+function formatDryRunEstimate(estimate) {
+  return `Dry run: parsed the file set to measure the embedding workload. No embedding requests were made and the index was not changed.
+
+  Files to embed:   ${estimate.filesCount.toLocaleString()}
+  Chunks to embed:  ${estimate.chunksCount.toLocaleString()}
+  Tokens to embed:  ${estimate.tokensToEmbed.toLocaleString()}
+
+The "Tokens to embed" value uses the local estimateTokens(text) = ceil(len/4). It
+matches the live "Tokens used" counter only for providers that report usage on the
+same basis (ollama); for providers that report a server tokenizer count (OpenAI,
+Gemini, custom) it is only an estimate.
+
+For a matching provider and a project-scoped force index, the force pass clears its
+own cached embeddings, so the live counter climbs to this number. A force index on a
+shared global index can reuse cached embeddings from other projects, and an
+incremental index counts cached chunks that are not re-embedded; in both cases this
+number is an upper bound on the live counter, so a progress percent against this
+total tops out below 100%.
+`;
+}
 function formatBytes(bytes) {
   if (bytes === 0) return "0 B";
   const k = 1024;
@@ -8881,6 +8901,9 @@ var CALL_GRAPH_SYMBOL_CHUNK_TYPES = /* @__PURE__ */ new Set([
   "enum_declaration",
   "function_definition",
   "class_definition",
+  // Ruby module/class symbols that are declaration-bearing and navigable.
+  "class",
+  "module",
   "class_specifier",
   "struct_specifier",
   "namespace_definition",
@@ -12481,6 +12504,70 @@ var Indexer = class _Indexer {
     );
     return createCostEstimate(files, configuredProviderInfo);
   }
+  // Dry-run counterpart to index()/forceIndex(): parse the real file set and sum
+  // estimateTokens over the embedding text of every indexable chunk, without
+  // calling the embedding provider or writing to the index. Read-only and
+  // lock-free (mirrors estimateCost). The token sum is the exact value "Tokens
+  // used" climbs to for a force index (cache bypassed); for an incremental it is
+  // an upper bound because cached chunks are counted here but not re-embedded.
+  // Used by index_codebase(dryRun:true) to give a stable, monotonic progress
+  // denominator that matches the live "Tokens used" basis.
+  async dryRunCost() {
+    const { configuredProviderInfo } = await this.ensureInitialized();
+    const maxChunkTokens = getSafeEmbeddingChunkTokenLimit(configuredProviderInfo);
+    const includePatterns = [...this.config.include, ...this.config.additionalInclude];
+    const { files } = await collectFiles(
+      this.materializedProjectRoot,
+      includePatterns,
+      this.config.exclude,
+      this.config.indexing.maxFileSize,
+      this.getMaterializedKnowledgeBases(),
+      { maxDepth: this.config.indexing.maxDepth, maxFilesPerDirectory: this.config.indexing.maxFilesPerDirectory }
+    );
+    let filesCount = 0;
+    let chunksCount = 0;
+    let tokensToEmbed = 0;
+    for (const batch of iterateOrderedFileBatches(files, (f) => f.size, this.fileBatchLimits)) {
+      const loadedFiles = await Promise.all(batch.map(async (f) => {
+        try {
+          return {
+            path: this.toStoredFilePath(f.path),
+            content: await fsPromises3.readFile(f.path, "utf-8")
+          };
+        } catch {
+          return null;
+        }
+      }));
+      const readable = loadedFiles.filter(
+        (f) => f !== null
+      );
+      filesCount += readable.length;
+      const contentByPath = new Map(readable.map((f) => [f.path, f.content]));
+      const parsedFiles = parseFiles(readable, this.config.indexing.linesPerChunk);
+      for (const parsed of parsedFiles) {
+        let chunksToProcess = parsed.chunks;
+        if (this.config.indexing.fallbackToTextOnMaxChunks && chunksToProcess.length > this.config.indexing.maxChunksPerFile) {
+          const content = contentByPath.get(parsed.path);
+          if (content !== void 0) {
+            chunksToProcess = parseFileAsText(parsed.path, content, this.config.indexing.linesPerChunk);
+          }
+        }
+        chunksToProcess = selectIndexableChunks(
+          chunksToProcess,
+          this.config.indexing.maxChunksPerFile,
+          this.config.indexing.semanticOnly
+        );
+        for (const chunk of chunksToProcess) {
+          const texts = createEmbeddingTexts(chunk, parsed.path, maxChunkTokens);
+          chunksCount += 1;
+          for (const text3 of texts) {
+            tokensToEmbed += estimateTokens(text3);
+          }
+        }
+      }
+    }
+    return { filesCount, chunksCount, tokensToEmbed };
+  }
   async index(onProgress) {
     return this.withIndexMutationLease("index", async (recoveredOwners) => {
       return this.indexUnlocked(onProgress, recoveredOwners);
@@ -15027,6 +15114,9 @@ async function runIndexCodebase(projectRoot3, host, args, onProgress) {
     if (args.estimateOnly) {
       return { kind: "estimate", estimate: await indexer.estimateCost() };
     }
+    if (args.dryRun) {
+      return { kind: "dryrun", dryrun: await indexer.dryRunCost() };
+    }
     const coordinated = runCoordinatedIndex(root, host, args.force ?? false, (progress) => {
       if (onProgress) {
         void onProgress(formatProgressTitle(progress), {
@@ -19103,12 +19193,14 @@ function codebaseIndexPiExtension(pi) {
     parameters: Type2.Object({
       force: Type2.Optional(Type2.Boolean({ default: false })),
       estimateOnly: Type2.Optional(Type2.Boolean({ default: false })),
+      dryRun: Type2.Optional(Type2.Boolean({ default: false })),
       verbose: Type2.Optional(Type2.Boolean({ default: false }))
     }),
     async execute(_toolCallId, params, _signal, _onUpdate, ctx) {
       try {
         const result = await runIndexCodebase(projectRoot2(ctx), HOST2, params);
         if (result.kind === "estimate") return text2(formatCostEstimate(result.estimate), result.estimate);
+        if (result.kind === "dryrun") return text2(formatDryRunEstimate(result.dryrun), result.dryrun);
         if (result.kind === "busy") return text2(result.text, { code: "INDEX_BUSY" });
         if (result.kind === "message") return text2(result.text);
         return text2(formatIndexStats(result.stats, params.verbose ?? false), result.stats);
@@ -19178,7 +19270,7 @@ function codebaseIndexPiExtension(pi) {
     return {
       systemPrompt: `${event.systemPrompt}
 
-Check index_status first when index readiness is unknown. Use codebase_context only when repository orientation is needed (for layout, key symbols, or cross-file dependency intent), not mechanically for every task. When using codebase_context for orientation, request a compact first pass (for example: tokenBudget: 600, limit: 5) and inspect returned evidence before broad search/grep/bash/read-style reads. Avoid repeating broad reads when the compact evidence already answers the question. Use implementation_lookup for known symbols and call_graph/call_graph_path after endpoints are identified for dependency flow.`
+Check index_status first when index readiness is unknown. Use codebase_context only when repository orientation is needed (for layout, key symbols, or cross-file dependency intent), not mechanically for every task. When using codebase_context for orientation, request a compact first pass (for example: tokenBudget: 600, limit: 5) and inspect returned evidence before broad search/grep/bash/read-style reads. For change requests with a known or strongly suspected target symbol, optionally use codebase_edit_context as a compact, bounded pre-edit context for source plus direct callers and callees. Avoid repeating broad reads when the compact evidence already answers the question. Use implementation_lookup for known symbols and call_graph/call_graph_path after endpoints are identified for dependency flow.`
     };
   });
   pi.on("session_shutdown", async (_event, ctx) => {