feat: test case file format for aggregate functions

[scgkiran]

max([20, -3, 1, -10, 0, 5]::i8) = 20::i8

corr([20, -3, 1, 10, 5]::fp32, [20, -3, 1, 10, 5]::fp32) = 1::fp64

some_agg_func([30, 3, 1, 10]::i32, ['app', 'bat', 'cat', 'dog']::string) = 30::i8

[jacques-n]

It doesn't work well with multi argument aggregates. Maybe we need to declare a table of data and then refer to in test cases? Other thoughts on multi args

[scgkiran]

It doesn't work well with multi argument aggregates. Maybe we need to declare a table of data and then refer to in test cases? Other thoughts on multi args

Option1:
corr([20, -3, 1, 10, 5]::fp32, [20, -3, 1, 10, 5]::fp32) = 1::fp64

In the parsing we could check both columns have same number of elements.

Option2:
corr([(20, 20), (-3, -3), (1, 1), (10,10), (5,5)]::Table(fp32, fp32)) = 1::fp64

[scgkiran]

Option3:

DEFINE TABLE test_table WITH SCHEMA (x fp32, y fp32) AS 
     VALUES [(20, 20), (-3, -3), (1, 1), (10, 10), (5, 5)];

# basic tests of corr
corr(test_table.x, test_table.y) = 1::fp64

# basic tests of mean
mean(test_table.x) = 6.6::fp64
mean(test_table.y) = 6.6::fp64

[jacques-n]

How about we start with supporting these and reexamine how we're enjoying them in 3-6 months?

multi-reference
DEFINE t1(fp32, fp32) = ((20, 20), (-3, -3), (1, 1), (10,10), (5,5))
corr(t1.col0, t1.col1) = 1::fp64

compact
((20, 20), (-3, -3), (1, 1), (10,10), (5,5)) corr(col0::fp32, col1::fp32) = 1::fp64

supercompact (single argument only)
avg((1,2,3)::fp32) = 1::fp64

[jacques-n]

How we might support a function that takes a aggregate reference and a constant value.

((20), (-3), (1), (10)) LIST_AGG(col0::fp32, ','::string) = 1::fp64

Single-valued arguments are passed in a direct literals. Multi-valued are passed as references.