fix(pagination): keep has_next_page() consistent with envelope cursors

[feiiiiii5]

• I understand that this repository is auto-generated and my pull request may not be merged as-is (see the note at the bottom about src/openai/pagination.py).

Changes being requested

The bug

For the six envelope-cursor page classes, has_next_page() can return False while next_page_info() on the same object returns a usable PageInfo. A response with data: [], has_more: true and a cursor (last_id / next) therefore ends the iteration even though there is a next page:

page = client.conversations.items.list(conversation_id, limit=10)
page.has_next_page()      # False
page.next_page_info()     # PageInfo(...)  <- a real cursor
page.get_next_page()      # returns the next items
list(client.conversations.auto_paging_iter(conversation_id))  # stops early, silently

iter_pages() and auto_paging_iter() build on has_next_page(), so the remaining pages are dropped without any error. Affected classes: Sync/AsyncConversationCursorPage, Sync/AsyncNextCursorPage, Sync/AsyncTokenPage — i.e. conversations.items.list, videos.list, the admin roles / groups listings and beta.agents.environments.files.list.

Cause

Those six overrides end with return super().has_next_page(), and BasePage.has_next_page() (src/openai/_base_client.py:205-209) short-circuits when self._get_page_items() is empty. That is correct for SyncCursorPage / AsyncCursorPage, whose cursor is literally data[-1].id and thus cannot exist without items — but the envelope classes read their cursor from a response field, so the emptiness of data says nothing about whether the server has more to give.

Change

For exactly those six classes, delegate to the cursor itself:

-        return super().has_next_page()
+        return self.next_page_info() is not None

The has_more is False short-circuit above it is kept, so a server that says it is done still wins, and no-cursor responses still report False. SyncCursorPage / AsyncCursorPage and BasePage are untouched.

The change

For the six envelope-cursor classes, has_next_page() now asks the cursor instead of the row list:

-        return super().has_next_page()
+        return self.next_page_info() is not None

The has_more is False short-circuit above it is kept, so a server that says it is done still wins, and no-cursor responses still report False. SyncCursorPage / AsyncCursorPage and BasePage are untouched.

Additional context & links

Testing

New tests/test_pagination_envelope_cursor.py covers each of the six classes plus SyncCursorPage/AsyncCursorPage as controls, and an iter_pages() case that stops only when the cursor is exhausted:

(cd repo root) PYTEST_DISABLE_PLUGIN_AUTOLOAD=1 PYTHONPATH=src:. python -m pytest tests/test_pagination_envelope_cursor.py -q -o addopts="" -p pytest_asyncio.plugin

• pristine 69a2c1db: 7 failed, 19 passed — the seven failures are the six empty-page assertions and the iter_pages() case, with AssertionError: assert False is True on has_next_page().
• this branch: 26 passed.

Neighbouring streaming/pagination files as a regression control (tests/test_streaming.py, tests/lib/test_streaming_deltas.py, tests/lib/chat/test_completions_streaming.py, tests/lib/chat/test_stream_moderation.py, tests/lib/streaming/agents/test_streams.py), same command: 13 failed, 139 passed, 12 skipped at head and identically 13 failed, 139 passed, 12 skipped on pristine 69a2c1db. Those 13 need live credentials, so they fail the same before and after; no test changed status because of this patch.

ruff check → All checks passed; ruff format --check → 2 files already formatted. All numbers above are local runs — CI on a fork PR still needs a maintainer to approve the workflow runs.

• src/openai/pagination.py carries no # File generated from our OpenAPI spec by Castiron header (compare src/openai/types/chat/chat_completion.py, which does), so I read it as the hand-written pagination layer rather than generator output. If the cursor logic in fact lives in a template I cannot see, the same one-line change belongs there and I would defer to that.
• No .castiron-ratchet.json change: this replaces 6 existing lines and adds no new custom-code surface beyond the test file.
• I have an equivalent fix open against the Anthropic SDK for the same BasePage.has_next_page() shape (fix(pagination): continue past an empty page that carries a cursor anthropics/anthropic-sdk-python#1943). Different repository and different code — mentioning it only so you can see the pattern is not specific to one SDK.
• This PR was prepared with an AI assistant; I reviewed the diff and ran the commands above locally.

[jnohclee-rgb]

Independent offline verification (AI-assisted): I independently checked the six sync/async envelope-cursor page classes across empty/populated data, absent/empty/nonempty cursor, and absent/false/true has_more, plus four row-derived cursor controls. Against main, 100/112 cases passed: all 12 failures had empty rows with a nonempty envelope cursor and has_more absent or true. This head passed 112/112. Explicit has_more=False remained terminal. This checks has_next_page() policy on constructed pages; it does not exercise transport or full iter_pages() traversal.

Compared main 919b6236382f3f9f76d8453efb9f9ad03cff1126 with this PR head 413ee5966bfa48cbd3c65dc0b470490c403cedce using Python 3.14.7 and a shared existing dependency environment (not a historical lockfile matrix). Source trees were isolated; runs denied network access. This is bounded additional evidence, not full SDK or merge approval.

Reproducer, runnable separately against each source tree:

import json,itertools
from pydantic import BaseModel
import openai.pagination as source
class Item(BaseModel):id:str
rows=[]
for name in ['SyncConversationCursorPage','AsyncConversationCursorPage','SyncNextCursorPage','AsyncNextCursorPage','SyncTokenPage','AsyncTokenPage']:
 cls=getattr(source,name)[Item];field='last_id' if 'Conversation' in name else 'next'
 for populated,cursor,has_more in itertools.product([False,True],[None,'','fictional-cursor'],[None,False,True]):
  page=cls(data=[Item(id='fictional')] if populated else [],has_more=has_more,**{field:cursor})
  expected=has_more is not False and bool(cursor);actual=page.has_next_page()
  rows.append({'class':name,'populated':populated,'cursor':cursor,'has_more':has_more,'expected_has_next':expected,'actual_has_next':actual,'pass':actual==expected})
for name in ['SyncCursorPage','AsyncCursorPage']:
 for populated in [False,True]:
  page=getattr(source,name)[Item](data=[Item(id='fictional')] if populated else [])
  rows.append({'class':name,'populated':populated,'expected_has_next':populated,'actual_has_next':page.has_next_page(),'pass':page.has_next_page()==populated})
print(json.dumps({'source_module':source.__file__,'cases':rows,'passed':sum(r['pass'] for r in rows),'failed':sum(not r['pass'] for r in rows),'scope':'Constructed page objects and cursor/has_more policy, not transport, actual iter_pages traversal or provider contract.'}))