lode/on_conflict

Upsert policy for inserts — Ecto’s :on_conflict / :conflict_target options as plain data, rendered to Postgres INSERT ... ON CONFLICT.

Column references are typed Field(row) values (not bare strings), so the target/action columns are owner-checked against the schema being upserted — repo.upsert(schema: person_schema(), on_conflict: ...) unifies row, so a column of another schema won’t type-check. repo erases the policy to its runtime form (just names) before handing it to the adapter, which is why the adapter sees OnConflict(Nil).

repo.upsert(r, s, cs, on_conflict: on_conflict.Update( target: on_conflict.Columns([person_email()]), action: on_conflict.Replace([person_name()]), ))

Types

ConflictTarget

</>

Which uniqueness violation triggers the conflict branch (Ecto’s :conflict_target). Column references name the database columns (a field’s source, where they differ).

pub type ConflictTarget(row) {
  AnyConstraint
  Columns(List(field.Field(row)))
  Constraint(String)
}

Constructors

```
AnyConstraint
```
No explicit target — any unique constraint counts. Postgres allows this only for DO NOTHING; DO UPDATE must name its target.
```
Columns(List(field.Field(row)))
```
ON CONFLICT (col, ...) — the columns of a unique index.
```
Constraint(String)
```
ON CONFLICT ON CONSTRAINT name — a named unique constraint.

OnConflict

</>

The upsert policy (Ecto’s :on_conflict option).

pub type OnConflict(row) {
  Raise
  Nothing(target: ConflictTarget(row))
  Update(target: ConflictTarget(row), action: UpdateAction(row))
}

Constructors

```
Raise
```
No ON CONFLICT clause: a uniqueness violation surfaces as a ConstraintError (Ecto’s default, :raise — minus the raising).
```
Nothing(target: ConflictTarget(row))
```
ON CONFLICT [target] DO NOTHING: silently skip conflicting rows.
```
Update(target: ConflictTarget(row), action: UpdateAction(row))
```
ON CONFLICT target DO UPDATE SET ...: update the existing row instead.

UpdateAction

</>

What DO UPDATE sets on the existing row.

pub type UpdateAction(row) {
  ReplaceAll
  ReplaceAllExcept(List(field.Field(row)))
  Replace(List(field.Field(row)))
  Set(List(#(field.Field(row), value.Value)))
}

Constructors

```
ReplaceAll
```
Set every inserted column to the incoming (EXCLUDED) value. The insert payload omits autogenerated and virtual columns, so those are never replaced — but a natural (non-autogenerated) primary key in the payload is, matching Ecto’s :replace_all; use ReplaceAllExcept to keep it.

ReplaceAllExcept(List(field.Field(row)))

ReplaceAll minus the listed columns.

```
Replace(List(field.Field(row)))
```
Set only the listed columns to their incoming (EXCLUDED) values (Ecto’s {:replace, fields}).
```
Set(List(#(field.Field(row), value.Value)))
```
Set the listed columns to explicit values (Ecto’s on_conflict: [set: ...]).

Values

erase

</>

pub fn erase(policy: OnConflict(a)) -> OnConflict(b)

Re-type a policy’s phantom row (a no-op at runtime — the phantom has no runtime representation). repo calls this to hand the schema-checked policy to the adapter as OnConflict(Nil).

set_columns

</>

pub fn set_columns(
  action action: UpdateAction(row),
  insert_columns columns: List(String),
) -> List(String)

Resolve an action to the concrete column names it sets, given the columns the INSERT provides. Used by adapters to expand ReplaceAll/ReplaceAllExcept and by repo to reject actions that would set nothing.