up

src/PyArrowTable.jl

@@ -10,7 +10,12 @@ Supports:
 - the Tables.jl columnar interface, so e.g. `DataFrame(table)` should work. The columns are `PyArray`'s (i.e. `AbstractVector`'s pointing to python objects), or `ChainedVector`'s with `PyArray` subcomponents (if there is more than 1 batch).
 - DataAPI's `ncol` and `nrow` accessors.
 
-Currently does not support DataAPI's metadata-related functions, but support may be added later.
+Does not yet support:
+
+- DataAPI's metadata-related functions
+- `Tables.partitions` to iterate record batches
+
+This functionality may be added in future non-breaking releases.
 """
 struct PyArrowTable <: PyTable
     py::Py

test/runtests.jl

@@ -4,24 +4,35 @@ using Tables
 using DataAPI
 using DataFrames
 using CondaPkg
+using SentinelArrays
+
+const pa = pyarrow
 
 # Print packages for CI logs
 CondaPkg.withenv() do
-    run(`micromamba list`)
+    return run(`micromamba list`)
 end
 
-
 @testset "PyArrow.jl" begin
     jl_table = (; a=[rand(2); NaN], b=["a", "b", "c"])
     py_table = PyArrow.table(jl_table)
     @test py_table isa Py
 
     t = PyArrowTable(py_table)
-    # Test indexing
+
+    # PythonCall interface
+    @test PythonCall.ispy(t)
+    @test pyconvert(Bool, Py(t) == t.py)
+    @test pyconvert(Bool, Py(t) == py_table)
+
+    # Properties
     @test isequal(pyconvert(Vector, t["a"]), jl_table.a)
     @test isequal(pyconvert(Vector, t[0]), jl_table.a)
 
-    # Test property-access
+    # Test properties
+    @test propertynames(t) isa Vector{Symbol}
+    @test :py in propertynames(t)
+    @test :num_columns in propertynames(t)
     @test pyconvert(Int, t.num_columns) == 2
     @test pyconvert(Int, t.num_rows) == 3
 
@@ -58,6 +69,18 @@ end
     jl_table.b[3] = "no"
     @test_broken isequal(Tables.getcolumn(t, :a)[3], 5.0)
     @test_broken isequal(Tables.getcolumn(t, :b)[3], "no")
+
+    # Multiple record batches
+    data = (; f0 = @py([1, 2, 3, 4]),
+        f1 = @py(["foo", "bar", "baz", nothing]),
+        f2 = @py([true, nothing, false, true]))
+    batch = pa.RecordBatch.from_arrays(pylist(data), @py(["f0", "f1", "f2"]))
+    table = pa.Table.from_batches(pylist([batch for _ in 1:5]))
+    @test pyconvert(Bool, table[0].num_chunks == 5)
+    jl_table = PyArrowTable(table)
+    @test Tables.getcolumn(jl_table, :f0) isa ChainedVector
+    @test eltype(Tables.getcolumn(jl_table, :f0)) == Int
+    @test collect(Tables.getcolumn(jl_table, :f0)) == repeat([1,2,3,4], 5)
 end
 
 @testset "README examples" begin