[ybai08]

Ecto.Type.cast/2 for :naive_datetime and :utc_datetime (and the _usec variants) raises FunctionClauseError on any malformed string shaped like YYYY-MM-DD HH:MM. It should return :error. Strings like "2015-01-23 ab:50" or "abcd-01-23 23:50" match the clause for datetimes without seconds. to_i/1 turns the bad parts into nil, and those nils go straight into NaiveDateTime.new/6, which fails its guards in Calendar.ISO.valid_date?/3 / valid_time?/4.

So a form param can crash Ecto.Changeset.cast/4 where it should add a validation error:

Ecto.Changeset.cast({%{}, %{at: :naive_datetime}}, %{"at" => "2015-01-23 xx:50"}, [:at])
# before: ** (FunctionClauseError) no function clause matching in Calendar.ISO.valid_time?/4
# after:  %Ecto.Changeset{valid?: false, errors: [at: {"is invalid", ...}]}

The fix builds the date and time with the existing guarded cast_date/3 and cast_time/4 helpers, the same way the map clause of cast_naive_datetime/1 already does. Valid inputs give the same results as before.

Checks run:

• mix test test/ecto/type_test.exs: before the fix, 2 failures (naive_datetime casting naive datetime, utc_datetime cast), both FunctionClauseError. After the fix, 99 passed.
• mix test: 1611 passed (97 doctests, 1514 tests)
• mix compile --force --warnings-as-errors: clean
• mix format --check-formatted lib/ecto/type.ex test/ecto/type_test.exs: clean
• A script that casts 33 datetime strings (valid, invalid, negative years, offsets, the no-seconds form) to all four datetime types, run before and after the change. The only differences are the "abcd-01-23 23:50" rows, which go from raising to :error.

Not run: the earthly integration matrix, and Elixir versions other than 1.20.4 / OTP 29.