fix(community): resolve document_ids MIME types via files().get (#1867)

[Luis Martinez (lcmartinezdev)]

Description

GoogleDriveLoader.load() fails with HTTP 400 for any document_ids. _load_documents_from_ids resolves MIME types with files().list(q="id = '...'"), but the Drive API q parameter has no id field, so the request is rejected with "Invalid Value". Loading by document_ids is broken against the real API on every release since #1644 (4.0.0, 5.0.0, and main). 3.x is not affected. Mocked unit tests did not catch it because the Drive service is stubbed.

Fix

Resolve each id with files().get(fileId=..., fields="mimeType") — the correct way to fetch a file by id. The per-type dispatch added in #1644 (Sheets → _load_sheet_from_id, PDF → _load_file_from_id, else → _load_document_from_id) is unchanged.

Tests

• Updated the mock service to stub files().get() (the call the loader now makes).
• Added a regression test asserting MIME resolution uses files().get(fileId=...) and never files().list.
• All unit tests pass; ruff and mypy are clean.

Note

This makes one files().get call per id instead of a single files().list. That is required, since files().list cannot query by id, and document_ids lists are typically small.

Fixes #1867

[Luis Martinez (lcmartinezdev)]

Why files.get per id (and not a single request)

Per the Drive API docs, id is not a valid files.list query term. The supported q terms are:

name, fullText, mimeType, modifiedTime, viewedByMeTime, trashed, starred, parents, owners, writers, readers, sharedWithMe, createdTime, properties, appProperties, visibility, shortcutDetails.targetId

So there is no "fixed query" version of files.list(q="id = '...'") — you fetch a file by id with files.get, not with search. That leaves two shapes:

1. files.get per id (this fix) — N calls, simple and standard.
2. Batch HTTP request — one round-trip for up to 100 ids, but more error-handling code, and it does not reduce quota: the docs state "n requests batched together count as n requests, not one."

Batching would only save round-trip latency, not API usage, so it adds complexity for a benefit most callers don't need (id lists are usually short). I went with files.get per id: it is the documented way to fetch by id and keeps the change minimal. A batched variant can be added later if a caller needs to load many ids at once.

Docs: file-specific query terms, batching.