Merge pull request #1030 from getappmap/feat/token-limit-setting

feat: Token limit setting

.gitignore

@@ -33,4 +33,5 @@ resources/*.jar
 
 # AppMap
 tmp/appmap
+.navie/**/diff.txt
 

.navie/prompts/pr_description.md

@@ -0,0 +1,9 @@
+Describe the changes in the manner of a PR description.
+
+In the initial section, describe the changes that directly address the issue description and/or
+plan, if provided.
+
+In a subsequent section, describe any changes that were not part of the original plan or issue
+description, if provided.
+
+Do not emit any code snippets; just a description of the changes made.

.navie/prompts/review_vs_plan.md

@@ -0,0 +1,28 @@
+/nocontext
+
+Compare a set of code changes with an implementation plan.
+
+If no code changes are provided, emit a message indicating that the code changes are missing.
+
+If no implementation plan is provided, emit a message indicating that the implementation plan is
+missing.
+
+If both code changes and an implementation plan are provided, compare the code changes with the
+implementation plan and emit a list of differences between the two.
+
+Output your analysis in the following sections:
+
+## As planned
+
+- A change that is present in the implementation plan and also present in the code changes.
+- Another change that is present in the implementation plan and also present in the code changes.
+
+## Unplanned changes
+
+- A change that is present in the code changes but not in the implementation plan.
+- Another change that is present in the code changes but not in the implementation plan.
+
+## Unimplemented changes
+
+- A change that is present in the implementation plan but not in the code changes.
+- Another change that is present in the implementation plan but not in the code changes.

package.json

@@ -315,6 +315,11 @@
           "type": "number",
           "description": "Port number to pass to the Navie UI (used for extension development and debugging only)"
         },
+        "appMap.navie.contextTokenLimit": {
+          "type": "number",
+          "default": 8000,
+          "description": "Default size of the context to send to Navie AI"
+        },
         "appMap.scannerEnabled": {
           "type": "boolean",
           "default": false,

src/configuration/extensionSettings.ts

@@ -52,6 +52,11 @@ export default class ExtensionSettings {
     if (port && typeof port === 'number' && port > 0) return port;
   }
 
+  public static get navieContextTokenLimit(): number | undefined {
+    const limit = vscode.workspace.getConfiguration('appMap').get('navie.contextTokenLimit');
+    if (limit && typeof limit === 'number' && limit > 0) return limit;
+  }
+
   public static get useVsCodeLM(): boolean {
     return [true, 'true'].includes(
       vscode.workspace.getConfiguration('appMap').get('navie.useVSCodeLM') || false

src/services/chatCompletion.ts

@@ -76,12 +76,24 @@ export default class ChatCompletion implements Disposable {
 
   get env(): Record<string, string> {
     const pref = ChatCompletion.preferredModel;
-    return {
+
+    const modelTokenLimit = pref?.maxInputTokens ?? 3926;
+    const tokenLimitSetting = ExtensionSettings.navieContextTokenLimit;
+    const tokenLimits = [modelTokenLimit, tokenLimitSetting].filter(
+      (limit) => limit && limit > 0
+    ) as number[];
+
+    const env: Record<string, string> = {
       OPENAI_API_KEY: this.key,
       OPENAI_BASE_URL: this.url,
-      APPMAP_NAVIE_TOKEN_LIMIT: String(pref?.maxInputTokens ?? 3926),
       APPMAP_NAVIE_MODEL: pref?.family ?? 'gpt-4o',
     };
+
+    if (tokenLimits.length) {
+      env['APPMAP_NAVIE_TOKEN_LIMIT'] = Math.min(...tokenLimits).toString();
+    }
+
+    return env;
   }
 
   private static models: vscode.LanguageModelChat[] = [];

test/unit/services/chatCompletion.test.ts

@@ -9,6 +9,7 @@ import type {
   LanguageModelChatMessage,
   LanguageModelChatResponse,
 } from 'vscode';
+import { workspace } from 'vscode';
 
 import ChatCompletion from '../../../src/services/chatCompletion';
 import { addMockChatModel, resetModelMocks } from '../mock/vscode/lm';
@@ -33,9 +34,19 @@ describe('ChatCompletion', () => {
   beforeEach(async () => {
     resetModelMocks();
     addMockChatModel(mockModel);
+
     await ChatCompletion.refreshModels();
   });
 
+  function mockTokenLimitSetting(limit: number | undefined) {
+    sinon.stub(workspace, 'getConfiguration').returns({
+      get: sinon.stub().callsFake((key) => {
+        if (key === 'navie.contextTokenLimit') return limit;
+        return undefined;
+      }),
+    } as any); // eslint-disable-line @typescript-eslint/no-explicit-any
+  }
+
   before(async () => {
     mockModel.sendRequest = sendRequestEcho;
     chatCompletion = new ChatCompletion(0, 'test-key');
@@ -57,6 +68,24 @@ describe('ChatCompletion', () => {
     });
   });
 
+  describe('when token limit is configured below the LLM default', () => {
+    beforeEach(() => mockTokenLimitSetting(100));
+
+    it('applies the configuration setting', () => {
+      const env = chatCompletion.env;
+      expect(env['APPMAP_NAVIE_TOKEN_LIMIT']).to.equal('100');
+    });
+  });
+
+  describe('when token limit is configured above the LLM default', () => {
+    beforeEach(() => mockTokenLimitSetting(100_000));
+
+    it('uses the LLM limit', () => {
+      const env = chatCompletion.env;
+      expect(env['APPMAP_NAVIE_TOKEN_LIMIT']).to.equal('325');
+    });
+  });
+
   it('should refresh models and set the preferred model', async () => {
     resetModelMocks();
     const mockModel1 = {