bug in array_add

[chochkov]

@roa

see this https://github.com/adjust/pg-numhstore/compare/array_add_fix?expand=1 for an illustration of a bug that results in us having cohorts with 0 valued keys for apps that make no revenue.

Could you please fix that

[chochkov]

that thing fails to build

[roa]

what is the expected result of

SELECT array_add(ARRAY[1,2,3]::text[], ARRAY[1,NULL,3]);

?

[chochkov]

the expected result would be:

  "1"=>"1", "3"=>"3"

[chochkov]

strangely, there's already a spec that seems to cover NULL args and that one passes, but the one I added doesnt

[roa]

is it possible, that a value can be "0"?

[chochkov]

From my perspective, 0 should be respected, but NULLs ignored. However, I'd like @rapimo's view on this

[roa]

that is super hard and expensive as we would have to evaluate the array as a c and postgres datatype array. what is the difference between "0" and "NULL"?

[roa]

it is fixed here, if "0" and "NULL" is the same:

#19

[chochkov]

I think @rapimo should have the final say here.

[rapimo]

mmmh yeah not sure for me 0 is a valid value for a numhstore where NULL is a non existent key
so if it doesn't take too much effort I'd love to keep 0

[rapimo]

so @roa can you go ahead ? I added a test case in 8fe486f

[roa]

i just tried:

in an postgres integer array "0" and "NULL" is the same. if you explicitly set it to "0" it is set to NULL in the low level array. so i see no possibility to make that happen as i can not differ between both. any ideas?

[roa]

oh wait, i found something! hold on.

[roa]

@rapimo fixed. can you test if it works on mac os as well?