Python: fix(redis): type-check the history provider across the supported redis range

[Chinmay V (chinmayv095)]

Motivation & Context

The weekly dependency-range validator reported in #7340 that agent-framework-redis cannot move past its current redis bound because pyright fails at redis==8.0.1 with five reportUnnecessaryTypeIgnoreComment errors in _history_provider.py.

Removing those five comments does not fix it. I reproduced the validator locally and checked each version in the supported range:

redis version	original code	with the ignores deleted
6.4.0 (declared lower bound)	clean	6 errors
7.1.1 (currently locked)	clean	6 errors
8.0.1 (the failing candidate)	5 errors	clean

The cause is redis-py itself. On the older releases the asyncio client annotates lrange, rpush, llen and ltrim as returning Awaitable[T] | T, so await-ing them directly does not type-check ("int" is not awaitable) and the ignore is required. Newer releases narrow those annotations to the awaitable alone, which makes the very same comment unnecessary. No single # type: ignore comment satisfies the whole supported range, which is why this needs a code change rather than a comment sweep.

This matters beyond the bound bump: reportUnnecessaryTypeIgnoreComment = "error" is enabled precisely so these comments do not accumulate, and today five of them are load-bearing on one redis version and dead weight on another.

Description & Review Guide

• What are the major changes?

  • A module-level _redis_result helper accepts the Awaitable[T] | T union and returns T. It uses isawaitable, matching the existing pattern in agent_framework/_types.py. rpush, llen and ltrim now go through it and carry no ignore comment.
  • lrange is a slightly different case: it is partially unknown on the older annotations rather than merely union-typed, so even the helper leaves a strict-mode complaint about the argument expression. It now reads through an explicitly Any-typed local, which makes the call site independent of how precisely redis-py annotates it, with a cast pinning the list[str] that decode_responses=True guarantees. That also removes the two union-attr ignores on the loop below it.
  • The if redis_messages: guard was dropped as dead — iterating an empty list is already a no-op.

• What is the impact of these changes?

  • No behaviour change. _redis_result awaits exactly what the previous await awaited; the sync arm only exists to satisfy the union that redis-py declares.
  • This unblocks the pyright half of Dependency validation failed: redis (agent-framework-redis) #7340. It does not by itself let the bound move to redis>=8, and I want to flag why: the package also pins redisvl>=0.11.0,<0.16, and redisvl 0.15.0 requires redis<7.2,>=5.0. So redis 8.x is not co-installable with the current redisvl pin regardless of typing, and the validator's proposed redis<8.0.0 is already in tension with it. Raising redisvl is a separate call I have deliberately left to you — I have not touched any bound in this PR.

• What do you want reviewers to focus on?

  • Whether the client: Any escape hatch for lrange is acceptable, or whether you would rather keep an ignore there and accept one error on newer redis. I went this way because it is the only form I found that is clean on all three versions, but it is a real trade of local type precision and it is your call.
  • Whether _redis_result belongs in this module or in a shared location, since agent_framework_redis is not the only place that awaits redis-py commands.

Related Issue

Fixes #7340

Verification

• Pyright, run the way the validator does (--project pyproject.toml, strict, against each installed redis): 0 errors at 6.4.0, 7.1.1 and 8.0.1 on this branch. The table above is the before/after.
• Unit tests: packages/redis on main 45 passed / 0 failed / 0 errors, branch 48 passed / 0 failed / 0 errors. The FAILED/ERROR sets are identical (both empty) and the delta is exactly the three new tests.
• ruff format --diff and ruff check clean on both touched files.

Three tests added, all offline:

Test	Covers
test_returns_messages_when_lrange_is_synchronous	the non-awaitable arm end to end through get_messages, which the existing AsyncMock-based tests never exercise
TestRedisResultHelper::test_awaits_an_awaitable_result	the awaitable arm
TestRedisResultHelper::test_passes_through_a_plain_result	the plain-value arm

Overlap note: #7470, also mine, touches save_messages in this same file. The two changes are independent and either merge order works, but whichever lands second will want a trivial rebase — happy to do that whenever you tell me which you prefer first.

Contribution Checklist

• The code builds clean without any errors or warnings
• All unit tests pass, and I have added new tests where possible
• The PR follows the Contribution Guidelines
• This PR is linked to an issue and there is no other open PR for this issue (see Related Issue above).
• This is not a breaking change. If it is a breaking change, add the breaking change label (or add "[BREAKING]" to the title prefix, before or after any language prefix) — a workflow keeps the label and title prefix in sync automatically.

[Copilot]

Pull request overview

Fixes Redis history-provider type checking across supported redis-py versions without changing runtime behavior.

Changes:

• Adds a helper that normalizes awaitable and synchronous Redis results.
• Removes version-dependent type-ignore comments and adds focused tests.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
python/packages/redis/agent_framework_redis/_history_provider.py	Normalizes Redis command results across annotation versions.
python/packages/redis/tests/test_providers.py	Tests both helper branches and synchronous lrange results.

---

💡 Add a code-review agent skill for context-aware, tailored reviews. Learn more in the docs.

[Chinmay V (chinmayv095)]

Rebased onto current main to clear a conflict and re-verified, since this file has moved since the PR was opened.

The conflict was only the import block: main added import json for the new _serialize_json / _deserialize_json helpers while this branch added Awaitable, isawaitable, TypeVar and cast for the _redis_result helper. Both sets are kept. The two changes main landed in the meantime, dedup in save_messages (#7242) and the zero-retention early return (#7470), are untouched, and their redis calls go through _redis_result along with the original four.

Re-ran the version matrix from the PR body against the rebased branch:

redis	main	this branch
6.4.0 (declared lower bound)	clean	clean
7.1.1 (currently locked)	clean	clean
8.0.1 (the failing candidate)	5 errors	clean

So the report in #7340 still reproduces on main as five reportUnnecessaryTypeIgnoreComment errors at 8.0.1, and this branch is still clean across the whole supported range. packages/redis tests are 46/46 and ruff check is clean.

[Eduard van Valkenburg (eavanvalkenburg)]

small nit on the code (I hate unnecessary reassignments), but overall looks good.

[Eduard van Valkenburg (eavanvalkenburg)]

Suggested change

	client: Any = self._redis_client
	redis_messages = cast("list[str]", await _redis_result(client.lrange(key, 0, -1)))
	redis_messages = cast("list[str]", await _redis_result(self._redis_client.lrange(key, 0, -1)))

[github-actions]

Python Test Coverage Report •

File	Stmts	Miss	Cover	Missing
packages/redis/agent_framework_redis
_history_provider.py	76	1	98%	236
TOTAL	48077	4490	90%

Python Unit Test Overview

Tests	Skipped	Failures	Errors	Time
9715	36 💤	0 ❌	0 🔥	2m 34s ⏱️