SplitPdfHook ignores custom HTTPClient configuration

[trey-trimble-posh]

SplitPdfHook ignores custom HTTPClient configuration

Description

When providing a custom httpClient to UnstructuredClient, the SplitPdfHook ignores it and creates its own HTTPClientExtension with default settings. This makes it impossible to configure socket-level timeouts for split PDF operations.

Current Behavior

In SplitPdfHook.ts, the sdkInit method discards the user-provided client:

sdkInit(opts: SDKInitOptions): SDKInitOptions {
  const { baseURL } = opts;  // client is ignored
  this.client = new HTTPClientExtension();  // creates new client with defaults
  // ...
  return { baseURL: baseURL, client: this.client };
}

This causes split PDF requests to use Node.js's default undici headersTimeout of 5 minutes, even when users configure extended timeouts via custom HTTPClient.

Expected Behavior

The SplitPdfHook should respect the user-provided httpClient configuration, either by:

1. Passing the custom client through to HTTPClientExtension
2. Using the custom client's fetcher for split PDF requests
3. Exposing timeout configuration options that apply to split PDF operations

Reproduction

import { Agent, fetch as undiciFetch } from 'undici';
import { UnstructuredClient } from 'unstructured-client';
import { HTTPClient } from 'unstructured-client/lib/http';

const agent = new Agent({
  headersTimeout: 20 * 60 * 1000, // 20 minutes
  bodyTimeout: 20 * 60 * 1000,
});

const customFetch = (input, init) => undiciFetch(input, { ...init, dispatcher: agent });
const httpClient = new HTTPClient({ fetcher: customFetch });

const client = new UnstructuredClient({
  security: { apiKeyAuth: 'xxx' },
  httpClient,  // This is ignored when splitPdfPage=true
  timeoutMs: 20 * 60 * 1000,
});

// Large PDF with splitPdfPage=true will timeout at 5 minutes despite config

Workaround

Currently requires setting a global undici dispatcher via setGlobalDispatcher(), which affects all fetch calls in the process.

Environment

• unstructured-client: 0.25.1
• Node.js: 20.x