----------------
New Features
^^^^^^^^^^^^
- Adding environment variables to build and benchmark commands. (809, 833)
- Added ``--strict`` option to ``asv run`` to set exit code on failure. (865)
- Added ``--no-stats`` option to ``asv compare`` and ``asv continuous``. (879)
- Added ``--durations`` option to ``asv run`` and ``asv show`` for displaying
benchmark run durations. (838)
- Added ``--date-period`` option to ``asv run`` for running benchmarks for
commits separated by a constant time interval. (835)
- Web UI button to group regressions by benchmark. (869)
- Space-saving v2 file format for storing results. (847)
- ``timeraw_*`` benchmarks for measuring e.g. import times. (832)
- Support for using conda environment files for env setup. (793)
API Changes
^^^^^^^^^^^
- Results file format change requires ``asv update`` to update old data to v2
format.
- The configuration syntax for "matrix", "exclude", and "include" in
``asv.conf.json`` has changed. The old syntax is still supported, unless you
are installing packages named ``req``, ``env``, ``env_nobuild``.
Bug Fixes
^^^^^^^^^
- When an ``asv find`` step fails due to timeout, assume runtime equal to
timeout to allow bisection to proceed (768)
- Minor fixes and improvements (897, 896, 888, 881, 877, 876, 875, 861,
870, 868, 867, 866, 864, 863, 857, 786, 854, 855, 852, 850, 844,
843, 842, 839, 841, 840, 837, 836, 834, 831, 830, 829, 828, 826,
825, 824)
Other Changes and Additions
^^^^^^^^^^^^^^^^^^^^^^^^^^^
- Uniqueness of ``repr()`` for ``param`` objects is now guaranteed by suffixing
unique identifier corresponding to order of appearance. (771)
- Memory addresses are now stripped from the ``repr()`` of ``param`` elements,
allowing comparison across multiple runs. (771)
- ``asv dev`` is now equivalent to ``asv run`` with ``--python=same`` default.
(874)
- ``asv continuous`` by default now records measurement samples, for better
comparison statistics. (878)
- ASV now uses PEP 518 ``pyproject.toml`` in packaging. (853)