Update async features examples for Python 3.14

[lpozo]

Refreshes the sample code for understanding-asynchronous-programming, which backs Getting Started With Async Features in Python. Paired with the light update on that tutorial (CMS clone: post 2283).

The repo copy had drifted further than the article itself — it still pinned aiohttp==3.5.4 and requests==2.21.0 from 2019.

Correctness and modernization

• while tasks: instead of the done flag (examples 2, 3, 5). This came from a reader suggestion, and it also fixes a latent bug: the old loop removed items from tasks while iterating over it, which can skip a task. The loop now iterates over tasks.copy().
• asyncio.gather() → async with asyncio.TaskGroup() (examples 4 and 6). Task groups are the recommended default since Python 3.11 and cancel remaining tasks when one fails.
• asyncio.timeout(10) around the HTTP request in example 6, so an unresponsive server can't stall a task indefinitely.
• Renamed the queue parameter to work_queue in examples 2 and 3, where it was shadowing the queue module.
• Dropped the stray from builtins import range in example 2 (an IDE artifact — it's a no-op on Python 3).
• URL lists moved to https://, with the dead entries replaced. twitter.com now redirects to x.com, and yahoo.com returns 404 over https, so it was dropped.
• requirements.txt refreshed to current releases, keeping only the three direct dependencies rather than the full frozen transitive set.

Readability

A second pass after review of the article's code blocks:

• Early return instead of an else branch in example 1, so the main loop body isn't indented behind a conditional.
• Descriptive loop variables in example 1's runner — task_func, task_name, tasks_queue rather than t, n, q.
• t → current_task in the generator-driving loops of examples 2, 3, and 5.
• _ for the unused counter in examples 1 and 2.
• URL lists extracted into a urls variable in examples 5 and 6, instead of being inlined in the for statement header.
• Boilerplate docstrings and step comments dropped — they restated what the code already says.

Example 1 keeps its list of (function, name, queue) tuples on purpose. Deferring the call is exactly the thing example 2 changes when task() becomes a generator, so the two files stay minimally different.

Verification

All six examples run on Python 3.14.6 with the pinned requirements:

file	result
example_1.py	ok — output matches the article byte for byte
example_2.py	ok — output matches the article byte for byte
example_3.py	ok — total 32.0s
example_4.py	ok — total 17.0s
example_5.py	ok — total ~3.0s
example_6.py	ok — total ~1.2s

Examples 5 and 6 hit the live network, so their timings vary between runs. The async version stays roughly twice as fast as the sync one, which is the point the tutorial makes.

black -l 79 and flake8 --max-line-length 79 are both clean.

The code here is semantically identical to the code blocks in the updated article, verified line by line. This repo's PEP 8 blank-line formatting is preserved, including the blank line between the standard-library and third-party import groups — the article renders the same code with single blank lines, so the tutorial's python -m asyncio pstree output cites the line numbers from these files (example_4.py:12 and :22) and says so in a note.

🤖 Generated with Claude Code