random behaviour in condense of structs

[Dimitar Misev]

condense on structs behaves a bit randomly. Some examples below.

• works as expected:

  $ rasql -q 'select condense min over i in [0:9] using {100,100} from mr as c' --out string
  
    Result element 1: { 100, 100 }
  

• works as expected:

  $ rasql -q 'select condense min over i in [0:9] using {100, c[100,100]} from mr as c' --out string
    Result element 1: { 100, 163 }
  

• should output { 163, 100 }:

  $ rasql -q 'select condense min over i in [0:9] using {c[100,100], 100} from mr as c' --out string
    Result element 1: { 163, 1677983744 }
  

• sometimes correct ({0}), sometimes outputs random value, like:

  $ rasql -q 'select condense min over i in [0:9] using {i[0]} from mr as c' --out string
    Result element 1: { -67108744 }
  

• always correct:

  $ rasql -q 'select condense min over i in [0:9] using {i[0],100} from mr as c' --out string
    Result element 1: { 0, 100 }
  
  $ rasql -q 'select condense min over i in [0:9] using {100,i[0]} from mr as c' --out string
    Result element 1: { 100, 0 }
  

• random output:

  $ rasql -q 'select condense min over i in [0:9] using {c[100,100],i[0]} from mr as c' --out string
    Result element 1: { 163, 16515072 }
  

[Dimitar Misev]

@bbell which case is still an issue here?

[bbell]

The issue here is when there is a struct {X, Y} where Y is generated in the query but not assigned the "char" type. Most likely, the memory is being allocated and initialized, but it is not being handled correctly. My justification for this is that performing the operation:

valgrind directql -q 'select condense min over i in [0:9] using {c[100,100], 100} from mr as c' —out string -d ~/../RASBASE

Result element 1: { 163, 1677983744 }

describes no unexpected and uninitialized values that do not already appear in every directql query (which we have narrowed down to 1 or 2 so far, I believe).

Also, as far as I know, the result element here is going to be the same on every system. It is not random.

[bbell]

Was going through some old tickets, and this one seems to work for me now (probably resolved by the struct patch before the summer). Can anyone confirm?