Closed JosepSampe closed 1 year ago
Unless I missed something special we defined, that looks like a bug on the Ibis side. There are many sum impls in that file and the data type is required to determine which specific one should be used. I suggest opening a bug on Ibis to that effect.
Thanks @jacques-n, I opened the issue in https://github.com/ibis-project/ibis-substrait/issues/703
I'm trying to load into calcite an ibis-generated substrait plan, but it seems it is unable to map the
sum
(and possibly others) function.On the python side I have this code that ends up creating an
ibis.prop
file:On the java side, I'm doing this:
The exception I get is this:
Checking the plans, I can see that the plan generated using ibis contains this:
While if I generate the same plan using substrait-java (using
SqlToSubstrait()
), the plan contains this:Note that in ibis the function is named
sum
while in substrait-java is namedsum:fp32
If i print the
extensions.aggregateFunctions()
on the java side, I can see that all the functions include a data-type suffix:I don't know If I'm missing something or this is actually an issue. I wonder if this different mapping of functions between ibis and substrait-java is the reason why it is unable to map the
sum
function to a substrait java Plan when I call theProtoPlanConverter
. Do you know if there is a way/workaround to bypass this?Note that if in the python side I create a plan that does not include any scalar/aggregate function, the plan is correctly loaded into a java substrait Plan and then to Calcite.