@purista/harness-azure-foundry 1.2.6 → 1.5.1

package/dist/index.js

@@ -1,4 +1,4 @@
-import { BaseModelProvider, ModelError } from '@purista/harness';
+import { BaseModelProvider, accumulateStreamToolCallDeltas, createStreamToolCallState, finalizeStreamToolCalls, parseProviderJson, safePartialJson, toTokenUsage } from '@purista/harness';
 import ModelClient, {} from '@azure-rest/ai-inference';
 import { AzureKeyCredential } from '@azure/core-auth';
 import { createSseStream } from '@azure/core-sse';
@@ -41,8 +41,9 @@ class AzureFoundryModelProvider extends BaseModelProvider {
         return {
             content: choice?.message?.content ?? '',
             ...(toolCalls ? { toolCalls } : {}),
-            usage: toUsage(body.usage?.prompt_tokens, body.usage?.completion_tokens, body.usage?.total_tokens),
+            usage: toTokenUsage(body.usage?.prompt_tokens, body.usage?.completion_tokens, body.usage?.total_tokens),
             finishReason: toFinishReason(choice?.finish_reason),
+            outcome: toOutcome(choice?.finish_reason),
             raw: response
         };
     }
@@ -50,8 +51,9 @@ class AzureFoundryModelProvider extends BaseModelProvider {
         req.signal.throwIfAborted();
         let usage = { inputTokens: 0, outputTokens: 0, totalTokens: 0 };
         let finishReason = 'stop';
-        const toolState = new Map();
-        for await (const event of streamChat(this.client, req, false)) {
+        let providerFinishReason;
+        const toolState = createStreamToolCallState();
+        for await (const event of streamChat(this.client, req)) {
             req.signal.throwIfAborted();
             const data = parseStreamData(event, req, 'textStream');
             if (!data)
@@ -61,20 +63,21 @@ class AzureFoundryModelProvider extends BaseModelProvider {
                     yield { kind: 'delta', text: choice.delta.content };
                 }
                 if (choice.delta?.tool_calls) {
-                    accumulateToolCallDeltas(toolState, choice.delta.tool_calls);
+                    accumulateStreamToolCallDeltas(toolState, choice.delta.tool_calls);
                 }
                 if (choice.finish_reason) {
-                    finishReason = toFinishReason(choice.finish_reason);
+                    providerFinishReason = choice.finish_reason;
+                    finishReason = toFinishReason(providerFinishReason);
                 }
             }
             if (data.usage) {
-                usage = toUsage(data.usage.prompt_tokens, data.usage.completion_tokens, data.usage.total_tokens);
+                usage = toTokenUsage(data.usage.prompt_tokens, data.usage.completion_tokens, data.usage.total_tokens);
             }
         }
-        for (const call of finalizeStreamToolCalls(toolState, req, 'textStream')) {
+        for (const call of finalizeStreamToolCalls(toolState, callContext(req, 'textStream'), MALFORMED_JSON_MESSAGE)) {
             yield { kind: 'tool_call', call };
         }
-        yield { kind: 'finish', usage, finishReason };
+        yield { kind: 'finish', usage, finishReason, outcome: streamOutcome(finishReason, providerFinishReason) };
     }
     async doObject(req) {
         req.signal.throwIfAborted();
@@ -86,8 +89,9 @@ class AzureFoundryModelProvider extends BaseModelProvider {
         return {
             object: parseJson(text, req, 'object'),
             ...(toolCalls ? { toolCalls } : {}),
-            usage: toUsage(body.usage?.prompt_tokens, body.usage?.completion_tokens, body.usage?.total_tokens),
+            usage: toTokenUsage(body.usage?.prompt_tokens, body.usage?.completion_tokens, body.usage?.total_tokens),
             finishReason: toFinishReason(choice?.finish_reason),
+            outcome: toOutcome(choice?.finish_reason),
             raw: response
         };
     }
@@ -96,8 +100,9 @@ class AzureFoundryModelProvider extends BaseModelProvider {
         let partial = '';
         let usage = { inputTokens: 0, outputTokens: 0, totalTokens: 0 };
         let finishReason = 'stop';
-        const toolState = new Map();
-        for await (const event of streamChat(this.client, req, true)) {
+        let providerFinishReason;
+        const toolState = createStreamToolCallState();
+        for await (const event of streamChat(this.client, req)) {
             req.signal.throwIfAborted();
             const data = parseStreamData(event, req, 'objectStream');
             if (!data)
@@ -108,21 +113,22 @@ class AzureFoundryModelProvider extends BaseModelProvider {
                     yield { kind: 'partial', partial: safePartialJson(partial) };
                 }
                 if (choice.delta?.tool_calls) {
-                    accumulateToolCallDeltas(toolState, choice.delta.tool_calls);
+                    accumulateStreamToolCallDeltas(toolState, choice.delta.tool_calls);
                 }
                 if (choice.finish_reason) {
-                    finishReason = toFinishReason(choice.finish_reason);
+                    providerFinishReason = choice.finish_reason;
+                    finishReason = toFinishReason(providerFinishReason);
                 }
             }
             if (data.usage) {
-                usage = toUsage(data.usage.prompt_tokens, data.usage.completion_tokens, data.usage.total_tokens);
+                usage = toTokenUsage(data.usage.prompt_tokens, data.usage.completion_tokens, data.usage.total_tokens);
             }
         }
-        for (const call of finalizeStreamToolCalls(toolState, req, 'objectStream')) {
+        for (const call of finalizeStreamToolCalls(toolState, callContext(req, 'objectStream'), MALFORMED_JSON_MESSAGE)) {
             yield { kind: 'tool_call', call };
         }
         const object = parseJson(partial || '{}', req, 'objectStream');
-        yield { kind: 'finish', object, usage, finishReason };
+        yield { kind: 'finish', object, usage, finishReason, outcome: streamOutcome(finishReason, providerFinishReason) };
     }
     async doEmbed(req) {
         req.signal.throwIfAborted();
@@ -146,7 +152,7 @@ class AzureFoundryModelProvider extends BaseModelProvider {
                 index: item.index,
                 vector: Array.isArray(item.embedding) ? item.embedding : []
             })),
-            usage: toUsage(body.usage?.prompt_tokens, 0, body.usage?.total_tokens),
+            usage: toTokenUsage(body.usage?.prompt_tokens, 0, body.usage?.total_tokens),
             raw: response
         };
     }
@@ -160,7 +166,10 @@ function createClient(options) {
     if (!auth) {
         throw new Error('Azure AI Foundry apiKey or credential is required when no client is injected.');
     }
-    return ModelClient(endpoint, auth, clientOptions);
+    // The harness owns retry budgets (spec 23): disable the SDK pipeline's
+    // default retries while preserving explicit user retryOptions as an escape
+    // hatch via the spread below.
+    return ModelClient(endpoint, auth, { retryOptions: { maxRetries: 0 }, ...clientOptions });
 }
 async function postChat(client, req, stream) {
     const providerOptions = {
@@ -188,7 +197,7 @@ async function postChat(client, req, stream) {
         abortSignal: req.signal
     });
 }
-async function* streamChat(client, req, objectMode) {
+async function* streamChat(client, req) {
     const response = await postChat(client, req, true);
     const nodeResponse = typeof response.asNodeStream === 'function' ? await response.asNodeStream() : response;
     if (nodeResponse.status && nodeResponse.status !== '200' && nodeResponse.status !== 200) {
@@ -299,29 +308,6 @@ function extractToolCalls(toolCalls, req, method) {
         arguments: parseJson(call.function.arguments ?? '{}', req, method)
     }));
 }
-function accumulateToolCallDeltas(state, deltas) {
-    for (const delta of deltas) {
-        const index = typeof delta?.index === 'number' ? delta.index : 0;
-        const existing = state.get(index) ?? { args: '' };
-        if (delta?.id)
-            existing.id = String(delta.id);
-        if (delta?.function?.name)
-            existing.name = String(delta.function.name);
-        if (typeof delta?.function?.arguments === 'string')
-            existing.args += delta.function.arguments;
-        state.set(index, existing);
-    }
-}
-function finalizeStreamToolCalls(state, req, method) {
-    return [...state.entries()]
-        .sort((a, b) => a[0] - b[0])
-        .filter(([, call]) => call.id && call.name)
-        .map(([, call]) => ({
-        id: call.id,
-        name: call.name,
-        arguments: parseJson(call.args || '{}', req, method)
-    }));
-}
 function parseStreamData(event, req, method) {
     if (event === '[DONE]')
         return undefined;
@@ -329,35 +315,12 @@ function parseStreamData(event, req, method) {
         return parseJson(event, req, method);
     return event;
 }
-function parseJson(content, req, method) {
-    try {
-        return JSON.parse(content);
-    }
-    catch (error) {
-        throw malformedResponseError(req, method, 'Azure AI Foundry returned malformed JSON.', content, error);
-    }
-}
-function malformedResponseError(req, method, message, body, cause) {
-    return new ModelError(message, {
-        provider: 'azure-foundry',
-        model: req.model,
-        method,
-        reason: 'malformed_response',
-        providerBody: body
-    }, cause);
-}
-function safePartialJson(content) {
-    try {
-        return JSON.parse(content);
-    }
-    catch {
-        return { _partial: content };
-    }
+const MALFORMED_JSON_MESSAGE = 'Azure AI Foundry returned malformed JSON.';
+function callContext(req, method) {
+    return { provider: 'azure-foundry', model: req.model, method };
 }
-function toUsage(inputTokens, outputTokens, totalTokens) {
-    const input = inputTokens ?? 0;
-    const output = outputTokens ?? 0;
-    return { inputTokens: input, outputTokens: output, totalTokens: totalTokens ?? input + output };
+function parseJson(content, req, method) {
+    return parseProviderJson(content, callContext(req, method), MALFORMED_JSON_MESSAGE);
 }
 function toFinishReason(value) {
     switch (value) {
@@ -369,7 +332,26 @@ function toFinishReason(value) {
             return 'tool_calls';
         case 'content_filter':
             return 'content_filter';
+        case 'function_call':
+            return 'tool_calls';
         default:
             return 'error';
     }
 }
+function toOutcome(value) {
+    const finishReason = toFinishReason(value);
+    return {
+        finishReason,
+        ...(typeof value === 'string' ? { providerFinishReason: value } : {})
+    };
+}
+/**
+ * Stream outcome built from the tracked finish reason; `providerFinishReason`
+ * is omitted entirely when the provider never sent a finish reason.
+ */
+function streamOutcome(finishReason, providerFinishReason) {
+    return {
+        finishReason,
+        ...(typeof providerFinishReason === 'string' ? { providerFinishReason } : {})
+    };
+}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@purista/harness-azure-foundry",
-  "version": "1.2.6",
+  "version": "1.5.1",
   "description": "Azure AI Foundry model provider adapter for @purista/harness.",
   "type": "module",
   "main": "./dist/index.js",
@@ -45,12 +45,12 @@
     "@azure/core-sse": "^2.3.0"
   },
   "devDependencies": {
-    "@vitest/coverage-v8": "^4.1.8",
+    "@vitest/coverage-v8": "^4.1.9",
     "typescript": "^6.0.3",
-    "vitest": "^4.1.8"
+    "vitest": "^4.1.9"
   },
   "peerDependencies": {
-    "@purista/harness": "*"
+    "@purista/harness": "^1.5.1"
   },
   "engines": {
     "node": ">=24.15.0"