Add array index number to output when casting from json to array.

[dnwpark]

Given the query SELECT <array<int64>>to_json('[1, 2, "asdf"]'), produces the error:

InvalidValueError: while casting 'std::json' to 'array<std::int64>', in array elements, at index 2, expected JSON number or null; got JSON string

[dnwpark]

Adding error_message_context in fini_expression is super dubious imo

[msullivan]

Adding error_message_context in fini_expression is super dubious imo

Why is that necessary?

[dnwpark]

In _compile_call_args there is a check to ensure that all args have a known multiplicity. The multiplicity is added in fini_expression. I suppose the other alternative is to add an override to infer_multiplicity on TypeCast.

[msullivan]

Yeah, adding it to multiplicity and cardinality checks is the correct thing

…

On Thu, May 30, 2024, 07:21 dnwpark ***@***.***> wrote: In _compile_call_args there is a check to ensure that all args have a known multiplicity. The multiplicity is added in fini_expression. I suppose the other alternative is to add an override to infer_multiplicity on TypeCast. — Reply to this email directly, view it on GitHub <#7397 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AACTC7NSVWCINFOACRD3RTDZE4YUZAVCNFSM6AAAAABIIIVZQ6VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDCMZZGY3DSMBZGI> . You are receiving this because your review was requested.Message ID: ***@***.***>

[dnwpark]

Benchmarking against three queries:

Simple null - 10000 times

client.query(
    "SELECT <array<int64>>to_json(<str>$input)",
    input='[null]'
)

• master: 8.927s
• With array indexes: 13.453s

Long array - 1000 times

client.query(
    "SELECT <array<int64>>to_json(<str>$input)",
    input='['+','.join(str(i) for i in range(1000))+',null'+']'
)

• master: 4.647s
• With array indexes: 10.134s

Long and deep array - 1000 times

client.query(
    "SELECT <array<tuple<array<tuple<array<int64>>>>>>to_json(<str>$input)",
    input='['+','.join('[[[[1,2,3],[4,5,6],[7,8,9]]]]' for i in range(1000))+',[[[[1,2,null]]]]'+']'
)

• master: 9.406s
• With array indexes: 90.841s

[dnwpark]

@msullivan non-trivial cases seem convincingly much slower.

[msullivan]

Great work experimenting with this, though!

[msullivan]

Also, we support json arguments directly too