Remove invalid column_view usage in string-scalar-to-column function

[davidwendt]

Description

Fixes the make_column_from_scalar function for string_scalar internal usage of a temporary column_view with non-zero size but no data or children to call cudf::strings::detail::fill. This relied too much on fragile internal logic which has cause several headaches including the recent work adding prefetch logic to libcudf.

Checklist

• I am familiar with the Contributing Guidelines.
• New or existing tests cover these changes.
• The documentation is up to date with these changes.

[vyasr]

Trying to understand this. Are the checks for an invalid or empty value sufficient to make calling strings::detail::fill safe, or is it still fragile even in that scenario? If it is safe, are you choosing the generate->make_strings_column approach so that you can use the same code path in all cases?

[davidwendt]

Trying to understand this. Are the checks for an invalid or empty value sufficient to make calling strings::detail::fill safe, or is it still fragile even in that scenario? If it is safe, are you choosing the generate->make_strings_column approach so that you can use the same code path in all cases?

It is not safe to call strings::detail::fill since strings::detail::fill expects a column of rows and I was using a trick where I created a strings column_view with non-zero size and a null-mask but no children (or data). This is fragile and now removed by performing the scalar-to-column logic without using strings::detail::fill. The code here is simpler (and probably faster) than the strings::detail::fill logic since every row is created with the same value.

[davidwendt]

/merge