[FEA] cuio: use datasource as the exclusive reader input argument

[cwharris]

Internally, readers are instantiating datasources (from filepaths or otherwise), presumably out of convenience. This complicates the reader interfaces and implementation. Let's factor the readers filepath arguments in to explicit datasource initialization outside of the reader and accept std::unique_ptr<datasource> or std::vector<std::unique_ptr<datasource>> (in the case of parquet, for example) as file/data input exclusively.

cudf/cpp/src/io/csv/reader_impl.hpp

Lines 78 to 81 in f4735c7

	explicit impl(std::unique_ptr<datasource> source,
	std::string filepath,
	reader_options const &options,
	rmm::mr::device_memory_resource *mr);

cudf/cpp/src/io/json/reader_impl.hpp

Lines 177 to 180 in f4735c7

	explicit impl(std::unique_ptr<datasource> source,
	std::string filepath,
	reader_options const &args,
	rmm::mr::device_memory_resource *mr);

cudf/cpp/src/io/orc/reader_impl.cu

Lines 817 to 823 in f4735c7

	reader::reader(std::vector<std::string> const &filepaths,
	reader_options const &options,
	rmm::mr::device_memory_resource *mr)
	{
	CUDF_EXPECTS(filepaths.size() == 1, "Only a single source is currently supported.");
	_impl = std::make_unique<impl>(datasource::create(filepaths[0]), options, mr);
	}

cudf/cpp/src/io/parquet/reader_impl.cu

Lines 1248 to 1253 in f4735c7

	reader::reader(std::vector<std::string> const &filepaths,
	reader_options const &options,
	rmm::mr::device_memory_resource *mr)
	: _impl(std::make_unique<impl>(datasource::create(filepaths), options, mr))
	{
	}

[vuule]

The reason why we have a separate path for file names is because JSON and CSV derive compression type from the file extension. If we pass this information in a different way, we can remove the filepath overloads.

[cwharris]

Sounds like that makes this issue dependent on #6188

[github-actions]

This issue has been marked rotten due to no recent activity in the past 90d. Please close this issue if no further response or action is needed. Otherwise, please respond with a comment indicating any updates or changes to the original issue and/or confirm this issue still needs to be addressed.

[vuule]

This will be closed by #9088 and #9089

[vuule]

@cwharris can we close this one now? :)