cudf
cudf copied to clipboard
[FEA] Remove legacy JSON reader
The legacy JSON reader has been replaced by an "experimental" JSON reader. The experimental reader has been in production use for over a year, so we'd like to remove the legacy code.
Python's options for using the legacy reader in cudf.read_json have been deprecated since 23.02 and we use the "experimental" (non-legacy) reader by default in C++ (#11982). I'll file an issue describing how we can remove this in parts:
- [x] Remove legacy reader support from Python in 24.06: https://github.com/rapidsai/cudf/pull/15538
- [x] Deprecate C++ code used to enable the legacy reader in 24.06 https://github.com/rapidsai/cudf/pull/15558
- [x] Fix deprecation warnings (oops, forgot this step): https://github.com/rapidsai/cudf/pull/15563
- [x] Remove the name "experimental" from JSON test files:
- https://github.com/rapidsai/cudf/pull/15563/files#r1570973860
- https://github.com/rapidsai/cudf/pull/15568
- [ ] Remove
host_parse_nested_json:- https://github.com/rapidsai/cudf/pull/15568#issuecomment-2065369856
- https://github.com/rapidsai/cudf/pull/15674
- [ ] Remove legacy reader from C++ code in 24.08
- [ ] Attempt to remove
concurrent_unordered_map. libcudf's copy of this code is used in legacy JSON reader, which might be the last remaining usage. All others have been migrated to use cuCollections. See https://github.com/rapidsai/cudf/issues/12261
I feel confident about (1) and will open a PR. @vuule -- can you weigh in on the rest of the plan here?
sounds good. Only one thing - we can move the new reader out of experimental earlier. IIRC this is something we missed when we switched the default reader.
~Great. I edited the task list to move out of the experimental namespace in 24.06.~
Apparently this was already done. Task removed.