Inconsistency in cudf directory and file structures

[ttnghia]

Currently, there has been a number of inconsistency issues in cudf directory and file structures. Below are some examples:

• The current directory structures of cpp/include, cpp/src, cpp/tests and cpp/benchmarks are inconsistent. For instance, there are tests in tests/grouped_rolling which test grouped_rolling_window corresponding to src/rolling; or unit tests in tests/collect_list and tests/lead_lag that are all tests for cpp/aggregation.
• There are tests/utilities and tests/utilities_test which have confusing names.
• There are tests in tests/column/colum_view_test.cpp which are actually testing logical_cast (or bit_cast), so they should be put together with the tests in tests/unary/cast_tests.cpp.
• File name/extensions are inconsistent. For example, many unit tests in the tests/<some test name> are written in both .cu and .cpp extensions. Of course, .cu file may be required here, but having the tests written in two file extensions at the same time make things look messy.

And many other similar issues that are not mentioned here. I would recommend to refactor and enforce more consistency.

[jrhemstad]

It looks like the main issue is that tests/grouped_rolling and tests/collect_list should be moved under tests/rolling?

There are tests in tests/column/colum_view_test.cpp which are actually testing logical_cast (or bit_cast), so they should be put together with the tests in tests/unary/cast_tests.cpp.

Technically, tests/unary are testing the cudf::unary_op functionality. I'm not sure if bit_cast belongs there.

There are tests/utilities and tests/utilities_test which have confusing names.

Can you elaborate? The former are test utilities and the latter are tests of the utilities.

File name/extensions are inconsistent. For example, many unit tests in the tests/ are written in both .cu and .cpp extensions. Of course, .cu file may be required here, but having the tests written in two file extensions at the same time make things look messy.

This one doesn't sound like an issue to me. We prefer to write tests as .cpp files when we can as they usually compile faster, but sometimes they need to be .cu if they have device code.