fix(expr): fix a critical bug in case expression

[wangrunji0408]

I hereby agree to the terms of the RisingWave Labs, Inc. Contributor License Agreement.

What's changed and what's your intention?

Sometimes the CASE expression fails to short circuit, causing the last matching value to be returned instead of the first one.

 dev=> create table t(x boolean, a int, y boolean, b int);
 CREATE_TABLE
 dev=> insert into t values (true, 1, true, 2);
 INSERT 0 1
 dev=> select *, case when x then a when y then b end from t;
  x | a | y | b | case 
 ---+---+---+---+------
- t | 1 | t | 2 |    2
+ t | 1 | t | 2 |    1

The reason is that some expressions ignore the visibility and return non-null values for invisible rows. Examples include literal expression and simd-optimized expressions such as =. The CASE expression doesn't mask invisible rows in the condition result, so later values may override earlier ones.

This bug seems to have been there from day 1 v0.1.15 (Jan 2023). It sounds very horrible. 🤯

Checklist

• I have written necessary rustdoc comments
• I have added necessary unit tests and integration tests
• I have added test labels as necessary. See details.
• I have added fuzzing tests or opened an issue to track them. (Optional, recommended for new SQL features Sqlsmith: Sql feature generation #7934).
• My PR contains breaking changes. (If it deprecates some features, please create a tracking issue to remove them in the future).
• All checks passed in ./risedev check (or alias, ./risedev c)
• My PR changes performance-critical code. (Please run macro/micro-benchmarks and show the results.)

• My PR contains critical fixes that are necessary to be merged into the latest release. (Please check out the details)

Documentation

• My PR needs documentation updates. (Please use the Release note section below to summarize the impact on users)

[wangrunji0408]

The core fix.

[stdrc]

LGTM! Thx for the quick fix!

[codecov]

Codecov Report

Attention: 28 lines in your changes are missing coverage. Please review.

Comparison is base (b87bcc3) 68.33% compared to head (9c64760) 68.27%.
Report is 2 commits behind head on main.

Files	Patch %	Lines
src/storage/hummock_sdk/src/compact.rs	0.00%	14 Missing ⚠️
src/meta/src/model/stream.rs	0.00%	10 Missing ⚠️
src/expr/impl/src/scalar/case.rs	89.47%	4 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main   #13890      +/-   ##
==========================================
- Coverage   68.33%   68.27%   -0.06%     
==========================================
  Files        1528     1528              
  Lines      263343   263322      -21     
==========================================
- Hits       179958   179793     -165     
- Misses      83385    83529     +144     

Flag	Coverage Δ
rust	68.27% <57.57%> (-0.06%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[wangrunji0408]

I confirmed that this bug was introduced during v0.1.14 to v0.1.15, by #6581.
And it seems that we have never tested cases with more than 1 when clause. 🥶

[xxchan]

Is this considered a breaking change? 🥵 risingwavelabs/rfcs#24 (comment) @BugenZhao

[BugenZhao]

Is this considered a breaking change? 🥵 risingwavelabs/rfcs#24 (comment) @BugenZhao

Strictly speaking, yes. 😕 We need to introduce a new function type with the fixed implementation, while keeping the buggy one as "backend only". But it's also okay for me to simply assume that no one persisted the result prior to this PR. 🤣