Reject negative ln inputs

[akmhatey-ai]

Which issue does this PR close?

• Closes PostgreSQL compatibility: ln(-1.0::float8) should error, not return NaN #22271.

Rationale for this change

DataFusion currently returns NaN for negative ln inputs. PostgreSQL raises an error for the same case, and issue #22271 tracks aligning ln((-1.0)::float8) with that behavior.

This keeps the existing ln(0) behavior unchanged.

What changes are included in this PR?

• Move ln from the shared unary math macro to a dedicated LnFunc implementation.
• Return a compute error when ln receives a negative Float32 or Float64 value.
• Update scalar.slt to cover positive column inputs plus negative scalar and column error cases.

Are these changes tested?

Yes. I ran:

• cargo fmt --check -- datafusion/functions/src/math/mod.rs datafusion/functions/src/math/ln.rs
• cargo check -p datafusion-functions
• cargo test -p datafusion-functions math::ln --lib
• cargo test -p datafusion-sqllogictest --test sqllogictests -- scalar:586
• cargo test -p datafusion-sqllogictest --test sqllogictests -- scalar:593
• cargo test -p datafusion-sqllogictest --test sqllogictests -- scalar:597
• git diff --check

I also ran cargo test -p datafusion-sqllogictest --test sqllogictests -- scalar. The changed ln records passed, but the full file failed later at scalar.slt:1546+ because local aggregate_test_100 data was empty in this checkout; those failures were unrelated to this change.

Are there any user-facing changes?

Yes. ln now errors on negative inputs instead of returning NaN.

[kumarUjjawal]

I think we also need to look into other such log functions which also return NaN.

DataFusion CLI v53.1.0
> >   SELECT
    ln(2.0)    AS ln_pos,
    log10(100.0) AS log10_pos,
    log(-1.0)    AS log_neg,
    log2(-1.0)   AS log2_neg,
    log10(-1.0)  AS log10_neg;
+--------------------+-----------+---------+----------+-----------+
| ln_pos             | log10_pos | log_neg | log2_neg | log10_neg |
+--------------------+-----------+---------+----------+-----------+
| 0.6931471805599453 | 2.0       | NaN     | NaN      | NaN       |
+--------------------+-----------+---------+----------+-----------+
1 row(s) fetched.
Elapsed 0.004 seconds.

[kumarUjjawal]

We don't need a new file for the ln we can extend the already available macro for this.

[akmhatey-ai]

Updated in 48152f0.

Covered the other negative-input cases from your example:

• log now rejects negative values for float and decimal inputs.
• log2 and log10 now use checked unary handling for negative Float32/Float64 inputs.
• Invalid bases are unchanged and still follow the existing NaN behavior.

Validation run:

• cargo test -p datafusion-functions test_log -- --nocapture
• cargo test -p datafusion-sqllogictest --test sqllogictests -- scalar:640
• cargo test -p datafusion-sqllogictest --test sqllogictests -- scalar:662
• cargo test -p datafusion-sqllogictest --test sqllogictests -- scalar:711
• cargo test -p datafusion-sqllogictest --test sqllogictests -- scalar:722
• cargo test -p datafusion-sqllogictest --test sqllogictests -- scalar:750
• cargo test -p datafusion-sqllogictest --test sqllogictests -- scalar:761
• rustfmt --check datafusion/functions/src/math/log.rs datafusion/functions/src/macros.rs datafusion/functions/src/math/mod.rs
• git diff --check
• gitleaks detect --no-git on the changed files

One limitation: full cargo test -p datafusion-sqllogictest --test sqllogictests -- scalar still fails later on existing aggregate_test_100 expectations around lines 1551-1600, outside this log change.

[kumarUjjawal]

Can you also look into the power usage of log?

[kumarUjjawal]

This rewrite can skip the new error. log(-2.0, power(-2.0, 3.0)) should error because the log value is negative, but this can simplify to 3.0. Please only rewrite when the log domain is proven valid.

[kumarUjjawal]

This rewrite can skip the new error. log(-2.0, -2.0) should error, but this can simplify to 1. Please keep the original expression unless the value and base are proven valid