Fix : signum function bug when 0.0 input

[getChan]

Which issue does this PR close?

Closes #11557

What changes are included in this PR?

1. add signum function unit test
2. move signum_order function from monotonicity.rs to signum.rs
3. signum function implementation by new code. not make_udf_function! macro

Are these changes tested?

1. Add some unit tests.
2. datafusion-cli test

> select signum(-1), signum(1), signum(0), signum(-0.0), signum(0.0), signum(1.0), signum(-1.0)
;
+-------------------+------------------+------------------+---------------------+--------------------+--------------------+---------------------+
| signum(Int64(-1)) | signum(Int64(1)) | signum(Int64(0)) | signum(Float64(-0)) | signum(Float64(0)) | signum(Float64(1)) | signum(Float64(-1)) |
+-------------------+------------------+------------------+---------------------+--------------------+--------------------+---------------------+
| -1.0              | 1.0              | 0.0              | 0.0                 | 0.0                | 1.0                | -1.0                |
+-------------------+------------------+------------------+---------------------+--------------------+--------------------+---------------------+

Are there any user-facing changes?

no

[andygrove]

LGTM for a Postgres-compatible implementation of signum.

For spark compatibility we would need a small change to handle the -0.0 case to return -1. We could either implement a copy of this code in the Comet project, or add a flag to the constructor of this function to control behavior for the -0.0 special case.

[Throne3d]

Here are the values I get for Spark 3.5.1:

data = ["-1", "1", "0", "-0.0", "0.0", "1.0", "-1.0"]
df = spark.createDataFrame([(datum,) for datum in data], "col1 string")
df.collect()
# [Row(col1='-1'), Row(col1='1'), Row(col1='0'), Row(col1='-0.0'), Row(col1='0.0'), Row(col1='1.0'), Row(col1='-1.0')]

pprint.pprint(df.selectExpr("col1", "signum(cast(col1 as float)) as result").collect())
# [Row(col1='-1', result=-1.0),
#  Row(col1='1', result=1.0),
#  Row(col1='0', result=0.0),
#  Row(col1='-0.0', result=-0.0),
#  Row(col1='0.0', result=0.0),
#  Row(col1='1.0', result=1.0),
#  Row(col1='-1.0', result=-1.0)]

So it seems like -0.0 in Spark should specifically return -0.0 in the signum function, not -1, right?

Literals seem to be treated as Decimal instead, which doesn't make the distinction between negative and positive zero, so those don't see the same behavior - I'd guess that's what caused the specific results listed for Spark in #11557 (comment):

spark.sql("""SELECT -1, 1, 0, -0.0, 0.0, 1.0, -1.0""").collect()
# [Row(-1=-1, 1=1, 0=0, 0.0=Decimal('0.0'), 0.0=Decimal('0.0'), 1.0=Decimal('1.0'), -1.0=Decimal('-1.0'))]
spark.sql("""SELECT signum(-1), signum(1), signum(0), signum(-0.0), signum(0.0), signum(1.0), signum(-1.0)""").collect()
# [Row(SIGNUM(-1)=-1.0, SIGNUM(1)=1.0, SIGNUM(0)=0.0, SIGNUM(0.0)=0.0, SIGNUM(0.0)=0.0, SIGNUM(1.0)=1.0, SIGNUM(-1.0)=-1.0)]

For Postgres 16.3, this behavior looks like what I see for both decimal and float though!

postgres=# WITH tab(datum) AS (VALUES ('-1'), ('1'), ('0'), ('-0.0'), ('0.0'), ('1.0'), ('-1.0'))
SELECT *, cast(datum as decimal) decimal, cast(datum as float) float, sign(cast(datum as decimal)) sign_decimal, sign(cast(datum as float)) sign_float FROM tab;
 datum | decimal | float | sign_decimal | sign_float
-------+---------+-------+--------------+------------
 -1    |      -1 |    -1 |           -1 |         -1
 1     |       1 |     1 |            1 |          1
 0     |       0 |     0 |            0 |          0
 -0.0  |     0.0 |    -0 |            0 |          0
 0.0   |     0.0 |     0 |            0 |          0
 1.0   |     1.0 |     1 |            1 |          1
 -1.0  |    -1.0 |    -1 |           -1 |         -1
(7 rows)

[alamb]

Thanks @getChan @andygrove and @Throne3d