[chrikrah]

@csmarchbanks, a small parser fix. _parse_timestamp() splits -0.5 at the dot and calls int("-0"), which gives 0, so a timestamp between -1 and 0 parses as +0.5. The spec allows negative timestamps, and since #1229 the library writes this exact value (tests/test_samples.py:19 expects -0.500000000). Exemplar timestamps take the same path.

The fix builds that case as Timestamp(-0.5, 500000000), the form the samples test already uses. The new test parses one sample at -0.500000000 and one exemplar at -0.250000000, next to a -1.5 control.

$ tox -e py3.12,flake8,isort,mypy   # base e2f2bf9 plus this branch
py3.12: OK (440 passed, 14 skipped)
flake8: OK
isort: OK
mypy: OK
# parser.py from e2f2bf9 with the new test kept: 1 failed (test_negative_subsecond_timestamps)
# not run: py3.10, py3.11, py3.13, py3.14, pypy

No issue exists for this; it follows #1229. I kept the fix in the parser and left Timestamp alone, because its constructor is public and the writer side already works. Does that split work for you?