select_merge regression

[JamesKentDemcon]

Elixir version

1.15.4

Database and Version

MySQL 8.0.40

Ecto Versions

ecto 3.10.3, ecto_sql 3.10.2

Database Adapter and Versions (postgrex, myxql, etc)

myxql 0.6.4

Current behavior

Merging with maps which are nil throws an exception. The example below is taken from #3218, where this problem was initially mentioned. As described, if the translation t is missing, an error is thrown.

def product_with_translations(sku_code, locale, translated_fields \\ ~w(name)a) do
  Product
  |> from(as: :product)
  |> where(sku_code: ^sku_code)
  |> join(
    :left, [product: p],
    t in Product.Translation,
    on: t.product_id == p.id and t.locale == ^locale,
    as: :translation
  )
  |> select([product: p], p)
  |> select_merge([translation: t], map(t, ^translated_fields))
end

results in the error:

 ** (ArgumentError) cannot merge because the left side is not a map, got: nil

This issue was supposedly solved with #3219. Looking at git blame, the relevant lines were merged.

However, when I look at a git blame of the most recent version of Ecto, the relevant lines are missing and only a subset of the fix is merged.

Expected behavior

I would expect no exception to be thrown as is intended by the fix in #3219.

[JamesKentDemcon]

I was doing some experimenting, and interestingly enough, I am able to get this to work if I call the functions in IEx. It seems just not to work in my testing environment. I'll have to do some more digging.

[josevalim]

Can you try removing MIX_ENV=test mix clean and, if that doesn't fix it, rm -rf _build/test, to rule out it being a build issue?

[JamesKentDemcon]

If you meant running MIX_ENV=test mix clean, then neither of those steps worked.

[greg-rychlewski]

it might be your test database is empty but whatever is being called in iex is not

[JamesKentDemcon]

The test database is the one controlled by ExUnit so its cleaned up after every test.

[v0idpwn]

Part of it was removed right after the merge: d1910c4

I'll try to write a test that reproduces the behaviour, and potentially restore these clauses.

[v0idpwn]

Not able to reproduce here either :)

[greg-rychlewski]

From the description it seems it was happening when the left side of the join is nil. So if it's a left join I think we won't even get to the part of the code where this error can raise because the result is empty.

Maybe we improved that after Ecto 3.10. @JamesKentDemcon does using the latest Ecto solve the issue?