Add column type tests

[isVoid]

Addresses column requests for #8357

This PR adds nested type checks for cudf::column.

[davidwendt]

I think this would return true if either lhs or rhs were empty but the other was not.

Suggested change

	return lhs.num_children() > 0 and rhs.num_children() > 0
	? lhs.child(kidx).type() == rhs.child(kidx).type()
	: true;
	return lhs.num_children() > 0 and rhs.num_children() > 0
	? lhs.child(kidx).type() == rhs.child(kidx).type()
	: lhs.is_empty() == rhs.is_empty();

I see there is a test for this case, so was that intended?

[isVoid]

Thanks for pointing this out. From arrow:

In [16]: empty = pa.array(pd.Series([], dtype="category"), from_pandas=True)

In [18]: arr = pa.array(pd.Series([1, 1, 2], dtype="category"), from_pandas=True)

In [19]: empty.type == empty.type
Out[19]: True

In [20]: arr.type == empty.type
Out[20]: False

So what you suggested above is correct.

[jrhemstad]

I'm not sure about the plural "s" here and with is_lists. All the other traits are singular like is_fixed_width or is_chrono.

[isVoid]

Right, it probably make more sense align with other namings in traits.hpp.

I got the plural form idea from lists_column_view and structs_column_view and was under the impression they were named that way since.

[harrism]

Why notis_list_view

[isVoid]

The below reveals the fact that we use list_view as an alias for LIST type during type-dispatching, which is too detailed IMO.

template <typename T, CUDF_ENABLE_IF(is_list_view<T>())>

while

template <typename T, CUDF_ENABLE_IF(is_list<T>())>

seems just clear enough.

[harrism]

But LIST is not a C++ type, and is_X() is a type trait. After all, is_list<T>() is just syntactic sugar for std::is_same<T, list_view>, so it's barely needed at all.

[davidwendt]

Previously the code was doing rhs.type().id()==type_id::LIST and rhs.type().id()==type_id::STRUCT so the thought was to make is_list(data_type) and is_struct(data_type) functions. Since the code is now using the type-dispatcher, these are no longer required since we can just do std::is_same.

[isVoid]

We might be able to get rid of is_same in all if we use explicit specialization:

template <>
bool columns_equal_fn::operator()<list_view>(column_view const& lhs, column_view const& rhs)

[codecov]

Codecov Report

❗ No coverage uploaded for pull request base (branch-21.08@cefc266). Click here to learn what that means.
The diff coverage is n/a.

❗ Current head 2de2006 differs from pull request most recent head 1777355. Consider uploading reports for the commit 1777355 to get more accurate results

@@               Coverage Diff               @@
##             branch-21.08    #8505   +/-   ##
===============================================
  Coverage                ?   83.01%           
===============================================
  Files                   ?      109           
  Lines                   ?    18225           
  Branches                ?        0           
===============================================
  Hits                    ?    15129           
  Misses                  ?     3096           
  Partials                ?        0           

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update cefc266...1777355. Read the comment docs.

[isVoid]

rerun tests

[davidwendt]

Just wondering if this function is something that will be used outside of libcudf? Like from Python cudf or Spark Java? the #8357 issue only mentions internal libcudf.

[davidwendt]

Nice work.

[ajschmidt8]

Approving ops-codeowner file changes

[robertmaynard]

CMake changes LGTM

[isVoid]

@gpucibot merge