v1.2.0

Install with pip install -U async-limiter (PyPI).

Fixed

• Calling release() more often than acquire() raised the concurrent limit
  for the rest of the limiter's life. The extra release is now ignored and
  logged as a warning.
• Setting the system clock back kept callers waiting until the old timestamps
  expired. The limiter now measures time with a monotonic clock.
• A limiter used from a second event loop, such as a module-level limiter and
  two asyncio.run() calls or a test suite with one loop per test, raised
  RuntimeError as soon as a caller had to wait. This also affected
  module-level limiters on Python 3.9.
• On Python 3.9, creating a limiter after an earlier asyncio.run() had
  finished raised RuntimeError.

Added

• wait_margin argument: the extra time a waiting caller sleeps past the moment
  a slot opens. It defaults to 0.1 seconds, the value that was hardcoded before;
  wait_margin=0 gives the full allowed throughput.
• py.typed marker, so type checkers use the package's annotations.
• Python 3.12 and 3.13 are tested and declared as supported.
• dev extra with the tools CI runs.

Changed

• An unexpected error inside acquire() is no longer re-raised as
  RuntimeError; the original exception propagates. No such error is known to
  occur, and the errors metric stays in get_metrics() with a value of 0.
• The limiter no longer logs a warning when a waiting caller is cancelled, or a
  debug line each time old requests leave the window.
• Using one limiter from two event loops that run at the same time (two
  threads) raises RuntimeError on acquire().
• The test extra no longer installs aiohttp, which only the example uses.
• The package is classified as Beta, not Production/Stable.
• request_times holds time.monotonic() values, not time.time() values.

Full changelog: v1.1.0...v1.2.0