@ai-ecoverse/py-pandas

Pandas 3.0.6 built as WASIX side modules for @ai-ecoverse/wasix-python CPython 3.15 (cpython-315-wasm32-wasix). Since the Python 3.15 flip (homescoop#761) this is latest; the cp314 builds stay on npm by exact version (release notes).

• Build: in homescoop's CI from the PyPI sdist, pandas-3.0.6.tar.gz, with meson and Cython, against wasix-python 3.15.0-1 and py-numpy 2.5.4-1 (wasix-sysroot 2025.9.30-38; C++ headers from the sysroot). No patches.
• Ranges: wasix-python and py-numpy are taken as ranges from those revisions on.
• pandas 3 changes behaviour from 2.3:
  • copy-on-write is always on;
  • the default string dtype is str, a PyArrow-less Python-backed string array here;
  • pytz isn't a dependency any more;
  • several deprecated APIs are removed. See pandas' 3.0 release notes.

What changes for your code (pandas 3 after 2.3)

• Copy-on-write is always on. Chained assignment (df["a"][0] = 1) no longer writes through to df; use df.loc[0, "a"] = 1.

• Text columns default to the str dtype, not object.

• pytz is no longer a dependency: use zoneinfo (or install @ai-ecoverse/py-pytz yourself).

• Datetimes default to microseconds: parsed strings become datetime64[us], not [ns].

• Removed:

  • DataFrame.applymap (use map);
  • swapaxes;
  • Series.view and ravel;
  • fillna(method=) (use ffill/bfill);
  • the axis= keywords of groupby/rolling/resample;
  • read_csv's delim_whitespace, date_parser, infer_datetime_format and keep_date_col;
  • errors="ignore" in to_datetime/to_numeric;
  • mode.use_inf_as_na;
  • the Index.is_* attributes.

  groupby(observed=) defaults to True, and Series[int] is always by label.

Details: homescoop's release notes.