feat(data-exploration): countIf and other aggregations

[mariusandra]

Changes

Adds support for the following aggregations (with number of allowed arguments)

# Permitted HogQL aggregations
HOGQL_AGGREGATIONS = {
    "count": 0,
    "countIf": 1,
    "countDistinct": 1,
    "countDistinctIf": 2,
    "min": 1,
    "minIf": 2,
    "max": 1,
    "maxIf": 2,
    "sum": 1,
    "sumIf": 2,
    "avg": 1,
    "avgIf": 2,
    "any": 1,
    "anyIf": 2,
}

These two translate as follows

• countDistinct(x) -> count(distinct x) (in hogql) -> uniqExact(x) (internally in clickhouse)
• countDistinctIf(x) -> countIf(distinct x) (in hogql) -> uniqExactIf(x) (internally in clickhouse)

I'm softly biased towards not exposing uniq and uniqExact and uniqIf and uniqExactIf to the users.

• uniq and uniqExact are non-standard SQL terms (compared to "count distinct" which I've seen in probably every SQL dialect)
• The default uniq returns non-precise results (uniqExtract is precise but slower), so we'd need to explain this to users at every turn. We need to talk about the difference between the two often.
• The uniqExact aggregation just rolls off your tongue (uniq? unique? uniquerque?)
• Thus I propose we abstract all of this away into our own countDistinct, which is the natural thing you'd try if you have used any SQL before. This way we can later optimize slow queries to use uniq if needed, while keeping everyone using the "data integrity is important" uniqExact instead.

How did you test this code?

Updated the test

[Twixes]

What happens if one does count(properties.value == 2)?

[mariusandra]

Error because it expects just 0 arguments. I'd add the test, but the tests pass right now so it'll come in another PR 😅

[Twixes]

Oh yeah, sorry, I meant countDistinct

[mariusandra]

Well, give it a try :)

I guess the answer is you'll get a 0, 1 or 2 depending on the data.

For me it crashes, but adding toInt around the property gives 0. Doing coalesce(toInt(properties.value), 0) == 2 would likely work.

[Twixes]

Looks good to me. I completely agree with the thinking behind only exposing intuitive count functions – anyone with just a bit of SQL experience should be able to pick this up immediately