1fanwang opened a new pull request, #25138:
URL: https://github.com/apache/datafusion/pull/25138
## Which issue does this PR close?
- Closes #25137.
## Rationale for this change
`power(1.0, log(1.0, b))` returns `b` instead of `1.0`, and a filter over it
matches rows it should not:
```sql
SELECT count(*) FROM t WHERE power(1.0, log(1.0, b)) = 2.0;
-- returns 1, should be 0
```
Four rewrites assume the logarithm and power identity holds for their shared
base, which is only true for a base that is positive and not 1. They checked
nullability, which says nothing about the value, so a base of 1, 0, negative,
NaN or infinity folded wrongly. The same query over a nullable column already
returned the correct result, so the answer depended on the schema, not the
data.
## What changes are included in this PR?
The four rewrites now fire only when the base is a literal that is finite,
positive and not 1. A base whose value cannot be proven, such as a column,
keeps the original expression rather than folding.
## What is the testing strategy for this PR?
`test_log_simplify_rejects_degenerate_base` covers all three log rewrites
against bases of 1, 0, -2, NaN and infinity, and
`test_log_simplify_accepts_valid_base`
confirms a valid base still folds. The issue's reproducer was also run
through
`datafusion-cli` on the merge base and on this branch.
<details><summary>Reproducer, before and after</summary>
```sql
CREATE VIEW t AS SELECT * FROM (VALUES (2.0), (3.0)) s(b);
SELECT b, power(1.0, log(1.0, b)) AS p, log(1.0, power(1.0, b)) AS l FROM t;
SELECT count(*) AS c FROM t WHERE power(1.0, log(1.0, b)) = 2.0;
CREATE VIEW nn AS SELECT * FROM (VALUES (1.0), (0.0), (-2.0), (2.0)) s(a);
SELECT a, log(a, 1.0) AS l1, log(a, a) AS la FROM nn;
```
Before, on the merge base:
```
| b | p | l | | c | | a | l1 | la |
| 2.0 | 2.0 | 2.0 | | 1 | | 1.0 | 0.0 | 1.0 |
| 3.0 | 3.0 | 3.0 | | 0.0 | 0.0 | 1.0 |
| -2.0 | 0.0 | 1.0 |
| 2.0 | 0.0 | 1.0 |
```
After, on this branch:
```
| b | p | l | | c | | a | l1 | la |
| 2.0 | 1.0 | NaN | | 0 | | 1.0 | NaN | NaN |
| 3.0 | 1.0 | NaN | | 0.0 | -0.0 | NaN |
| -2.0 | NaN | NaN |
| 2.0 | 0.0 | 1.0 |
```
</details>
## Are there any user-facing changes?
Yes. Queries whose base is a literal 1, 0, negative, NaN or infinity now
return
the correct value instead of a folded one. No API change.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
