bug(frontend): pgwire binary format for interval overflows

[kwannoel]

Describe the bug

No response

To Reproduce

Execute a query in extended mode, to force binary format to be returned:

SELECT ((-2147483648) * (INTERVAL '86400'));

Expected behavior

No panic, return interval.

Interim workaround can be to use checked_mul to avoid panic.
But proper fix needs to avoid error.

Additional context

risingwave/src/common/src/types/interval.rs

Line 723 in f5b7fd7

out.put_i64(self.ms * 1000);

https://bulidkite-artifacts-bucket.s3.amazonaws.com/9eed51d0-eaaf-4ea3-95a2-2e9e557607f6/01865e72-d4b3-425e-b291-836a1daf317c/01865e73-2536-4397-af1c-c0f6b20b9033/risedev-logs/fuzzing-4.log

[jon-chuang]

It's supposed to be a decimal type, is that correct?

[kwannoel]

It's supposed to be a decimal type, is that correct?

Not sure, see: https://github.com/postgres/postgres/blob/517bf2d9107f0d45c5fea2e3904e8d3b10ce6bb2/src/backend/utils/adt/timestamp.c#L1014

[xiangjinwu]

#4514
It should be an overflow error during *, rather than during to_sql. I think we can checked_mul here first.

[xiangjinwu]

This has been fixed "automatically" by #8501, as binary formatting no longer requires any calculation, and the multiplication already handles the error reporting properly.

Before closing this issue as resolved, are there disabled test cases that we want to re-enable?

[kwannoel]

This has been fixed "automatically" by #8501, as binary formatting no longer requires any calculation, and the multiplication already handles the error reporting properly.

Before closing this issue as resolved, are there disabled test cases that we want to re-enable?

• I have tracked testing binary_format via sqlsmith here: Tracking: Sqlsmith extras #7329.
• We should have e2e_test for binary format too, just opened an issue to track it here: E2e testing for binary format #8560.

[kwannoel]

Currently sqlsmith does not test binary format at all, since I have changed it to use simple_query to avoid binary_format errors.

Will add it back eventually.