[bassilkhilo-ag2]

Summary

Adds tests/py39-compat.test.py — an AST-based lint that catches the #960-class bug (telegram-bridge outage from missing __future__ import) before it merges.

What it checks: Every src/*.py file that uses PEP-604 X | Y union syntax in annotation positions (function args, return types, variable annotations) must have from __future__ import annotations at module level. If not, the file crashes on import under Python 3.9 (macOS stock CLT interpreter).

Scope: Only src/*.py — skills and tests are typically invoked under Homebrew Python where 3.10+ is available.

Why Option A (lint) instead of Option B (baseline bump)

Option B (changing startup.sh to resolve a 3.10+ interpreter) is the cleaner long-term answer per issue #961, but it's a launch-path change across startup.sh and all bare-python3 call sites — a deliberate migration. Option A is a 1-file addition that catches regressions regardless of which option eventually wins.

This lint unblocks the decision: if owner picks A, this is the implementation. If owner picks B, this lint is a belt-and-suspenders safety net during the transition.

Current state

27 src/*.py files checked — all pass (all current files that use | unions already have the guard).

Test plan

python3 tests/py39-compat.test.py  # → PASS: 27 src/*.py files checked

Verify regression detection works:

# Temporarily add to a src/*.py without the __future__ import:
def foo(x: str | None) -> None: pass
# → FAIL: flagged by lint

🤖 Generated with Claude Code

[VasiliyRad]

Lint rule is correct: AST walk covers all annotation positions (function args, posonlyargs, kwonlyargs, vararg, kwarg, returns, AnnAssign). Logic is sound — only flags when future guard is absent AND PEP-604 BinOp|BitOr appears in annotation nodes. Single-file addition, no unrelated changes. Passes clean against all 27 current src/*.py files.

[VasiliyRad]

Functionally correct, but this overlaps almost entirely with #1199's scripts/check-py39-union-annotations.py + CI wiring — same goal (Option A of #961), same logic (AST walk for BitOr in annotation positions, gated on from __future__ import annotations), same allowed/disallowed set.

If #1199 lands, this test file is redundant — the script is the gate, and CI invokes it. The only marginal value of keeping both would be npm run test:py (or similar local runner) picking it up. If that's the intent, fine — but I'd rather see one canonical implementation. Suggest closing this PR and letting #1199 carry the gate, or vice versa.

Not blocking on the redundancy alone if you want both for belt-and-suspenders, but please confirm the intent before landing.

[bassilkhilo-ag2]

Closing — PR #1199 is the more complete solution for #961: dedicated scripts/check-py39-union-annotations.py + explicit CI step + companion deal-finder fix. VasiliyRad flagged the overlap in the review here. #1199 covers the same AST-walk logic and has two approvals.