[FEA] Cover struct columns in fuzz testing

[vuule]

The cuIO fuzz tests currently cannot generate input with struct columns. cudf now support this type of data for Parquet IO, so coverage should be expanded to include columns of struct data type.

Ideally, the generated columns should cover the following:

• Variable levels of nesting;
• Combination of nested structs, lists and non-nested data types;

[vuule]

CC @devavret and @nvdbaranec to add any corner cases that need to be covered.

[devavret]

1. One of the children is strings
2. Columns have an offset (sliced column)
3. Many permutations/combinations of nullability of levels. (every level nullable, no level nullable etc)

[github-actions]

This issue has been labeled inactive-30d due to no recent activity in the past 30 days. 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. This issue will be labeled inactive-90d if there is no activity in the next 60 days.

[vuule]

what is missing before we can close this one?

[nvdbaranec]

I would add: Many permutations of empty things. Empty strings, empty lists, empty structs. These tend to expose slightly different problems than just null values (particularly with strings, where various child columns can be nullptr, etc).

[galipremsagar]

#9180 has already started paying dividends for ORC struct support fuzz-testing: #9395, #9179

I would like to verify the nested empty types cases that @nvdbaranec pointed out and there is some cleanup to be done hence moving this to 22.02.