Resolve path parsing issues in get_json_object

[SurajAralihalli]

This PR addresses a parsing issue related to JSONPath by implementing distinct parsing rules for values inside and outside brackets. For instance, in { "A.B": 2, "'A": { "B'": 3 } }, $.'A.B' differs from $['A.B']. (See Assertible JSON Path Documentation)

The fix ensures accurate parsing of JSONPath values containing quotes. For example in { "A.B": 2, "'A": { "B'": 3 } }

JSONPath	Before Fix	Spark	After Fix
$.'A.B'	2	3	3
$.'A	CUDF_FAIL("Encountered invalid JSONPath input string")	{"B'": 3}	{"B'": 3}

Resolves 12483.

Checklist

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

[SurajAralihalli]

However, when encountering an invalid JSONPath, CUDF throws an error, whereas Spark silently returns null. For instance, consider the example { "A" : 1, "B" : 2 }

JSONPath	Cudf	Spark
.	CUDF_FAIL: invalid_argument	null
${A}	CUDF_FAIL: invalid_argument	null
][	CUDF_FAIL: invalid_argument	null

We can update the get_json_object to return null instead. If direct modification is not feasible, in order to maintain consistency with the current behavior in CUDF, we can introduce a new parameter named return_null_if_invalid_argument to the get_json_object_options function. This new parameter can have a default value of false.

[davidwendt]

However, when encountering an invalid JSONPath, CUDF throws an error, whereas Spark silently returns null.

If this is an invalid argument, could the Spark logic just catch the exception and then return a null on its own?

[SurajAralihalli]

However, when encountering an invalid JSONPath, CUDF throws an error, whereas Spark silently returns null.

If this is an invalid argument, could the Spark logic just catch the exception and then return a null on its own?

Yes, I think that can be done. We can have the Spark catch CudfException and return a column of nulls.

[ttnghia]

Please clarify: In such situations, Spark will return a column of all nulls, or only nulls corresponding to the invalid input rows (JSON objects)?

[SurajAralihalli]

Please clarify: In such situations, Spark will return a column of all nulls, or only nulls corresponding to the invalid input rows (JSON objects)?

I believe the column would consist entirely of null values since an invalid JSONPath (argument) would affect all the input rows uniformly.

[ttnghia]

If this is an invalid argument, could the Spark logic just catch the exception and then return a null on its own?

This is dangerous. We cannot know if such exception is due to invalid JSON path or something else unless we have a new exception type dedicated to it. So we should better to handle this case explicitly in the code.

[davidwendt]

If this is an invalid argument, could the Spark logic just catch the exception and then return a null on its own?

This is dangerous. We cannot know if such exception is due to invalid JSON path or something else unless we have a new exception type dedicated to it. So we should better to handle this case explicitly in the code.

I believe the exception occurs because of an unexpected parameter value which is checked in host code.
The exception type is std::invalid_argument and hope this would be specific enough to check by the caller.

[revans2]

But I am just looking at the results, not the code as I am no C++ expert

[bdice]

Let's refactor the bool inside_brackets to an enum class. It would help eliminate the comments that are currently needed to clarify the intent of the argument.

[ttnghia]

/ok to test

[revans2]

This looks fine for the java side of things.

[SurajAralihalli]

/ok to test

[ttnghia]

/ok to test

[SurajAralihalli]

/merge