[lesomnus]

ClearX() followed by SetX() left the mutation with the field both set and cleared: X() answered with the value, and XCleared() answered true. An update then emitted both assignments, SET x = NULL, x = ?.

	Result
SQLite, MySQL, MariaDB	The later assignment wins, which happens to be the value
PostgreSQL	The statement is refused: multiple assignments to same column
Hooks	See the field set and cleared at once

Every other pair of calls already lets the later call win:

• ClearX drops a value that was set.
• SetX drops what was added or appended.
• The update builder's setter for a numeric field calls ResetX first, which also drops the clear. That call was added in 2019 to drop the Add, not for this case.

This change makes SetX drop the clear of an optional field. The rule then holds for every field, including when a hook sets the mutation directly. The template change is one line; the rest of the diff is regenerated mutation.go files.

Upstream has the same code, and nothing about it had been reported there. UPSTREAM.md gains a Set after clear row, which says to drop it once upstream has the fix.

Tests

ClearFields in entc/integration gains a "set after clear" case. After ClearUser().SetUser(v), the mutation is not cleared, and the row holds v.

• Before this change: the case failed on every engine, at the mutation's cleared state.
• With this change, locally:
  • root and examples tests pass;
  • the whole entc/integration matrix passes with -race -count=2 on SQLite, MySQL 8.0 and 8.4, MariaDB 10.11 and 11.4, and PostgreSQL 14 and 17;
  • the generate gate is clean, run twice to rule out the ci: generate and lint fail now and then, on files the change did not touch #10 flake;
  • .github/lint.sh passes.

🤖 Generated with Claude Code