casacoretables 0.0.3

Standalone, minimal-dependency casacore table system with a pybind11 interface

casacoretables

A standalone, minimal-dependency build of casacore's table system, with a pybind11 Python interface that mirrors python-casacore's casacore.tables.

It bundles only casacore's casa (base) and tables modules — nothing else — so it builds and installs quickly with very few dependencies.

What was removed relative to casacore / python-casacore

• All other casacore modules: scimath, measures, ms, images, coordinates, fits, lattices, derivedmscal, ...
• Heavy third-party libraries: HDF5, ADIOS2, Dysco, BLAS/LAPACK, FFTW, CFITSIO, WCSLIB, MPI, OpenMP, and Boost (python-casacore uses Boost.Python; this package uses pybind11 instead).

Remaining dependencies

• Runtime: the C++ standard library, libdl, libm, pthreads, and numpy. (Nothing else.)
• Build: a C++17 compiler, CMake ≥ 3.18, and bison ≥ 3 + flex (build-time only, to generate the TaQL/Json parsers).

The MeasurementSet helpers (default_ms, msconcat, addImagingColumns, ...) are intentionally absent: they belong to casacore's ms module.

Install

conda install -n <env> bison flex      # build tools (>=3 for bison)
pip install .

Use

import numpy as np
from casacoretables.tables import table, maketabdesc, makescacoldesc, makearrcoldesc, taql

desc = maketabdesc([makescacoldesc("ANT", 0),
                    makearrcoldesc("DATA", 0.0, ndim=2, shape=[2, 4])])
t = table("my.tab", desc, nrow=5)
t.putcol("DATA", np.zeros((5, 2, 4)))
data = t.getcol("DATA")            # zero-copy: numpy view over casacore storage
sub  = taql("select from $t where ANT > 2")
t.close()

Documentation

A runnable notebook tours the entire Python API — table creation, columns, cells and slices, keywords, tablecolumn/tablerow, TaQL queries, iteration, indexed lookups, copy/rename, and ASCII import/export: docs/casacoretables_api_tour.ipynb.

Using alongside CASA

casacoretables statically links its own copy of casacore and hides all of those symbols (only the module init symbol is exported), so it can be imported into the same Python process as CASA's casatools — which embeds a different, ABI-incompatible casacore — without symbol clashes or crashes.

Zero-copy

Large column reads (getcol) return a numpy array that shares casacore's storage (no copy) via a capsule that keeps the buffer alive. Writes (putcol) and the in-place getcolnp/getcellnp fill paths wrap the numpy buffer directly when its dtype already matches the column, avoiding an intermediate copy. casacore's column-major arrays are exposed in numpy's row-major convention by reversing axes — exactly as python-casacore does.

Tests

• C++ unit tests (vendored from casacore) — configure with -DBUILD_CPP_TESTS=ON and run with ctest.
• Python tests (ported from python-casacore) — pytest tests/python.