[BUG] AST fails with zero-row inputs

[aocsa]

Describe the bug

AST can't handle inputs with zero rows because it can't get the right numbers of kernel's blocks for this case. This is an edge case, but it should be covered as pandas do.

C++ exception with description "CUDF failure at: /opt/mambaforge/conda-bld/libcudf-ext_1681746453367/work/cpp/include/cudf/detail/utilities/cuda.cuh:66: num_blocks must be > 0" thrown in the test body.

Steps/Code to reproduce bug

import cudf
df = cudf.DataFrame({'A': [], 'B': []})
df.eval('A + B')

RuntimeError: CUDF failure at: /opt/conda/conda-bld/work/cpp/include/cudf/detail/utilities/cuda.cuh:66: num_blocks must be > 0

Expected behavior

For the example above it should return a column with zero rows.

import pandas as pd
df = pd.DataFrame({'A': [], 'B': []})
df.eval('A + B')
Out[10]: Series([], dtype: float64)

Environment overview (please complete the following information)

• Environment location: conda
• Method of cuDF install: conda
• branch-23.04, origin/branch-23.04

[wence-]

Thanks, agreed. I think that #13245 fixes this correctly. I still run through the parser to infer types but just don't launch the kernel if the input is empty.