This PR introduces stricter inference for the `@[defeq]` attribute and a
companion `@[backward_defeq]` attribute that preserves the pre-PR
behavior
as an opt-in.
### What changed
* `@[defeq]` is now inferred only when the equation holds at
`.instances` transparency (the transparency `dsimp` operates at).
* `@[backward_defeq]` is the old set: every theorem whose `rfl` proof
the legacy inference would have accepted is tagged `@[backward_defeq]`,
so `defeq ⊆ backward_defeq` holds by construction.
* The option `backward.defeqAttrib.useBackward` (default `false`) makes
`dsimp` also use `@[backward_defeq]` theorems, restoring the pre-PR
behavior for a specific proof or file.
* The option is eqn-affecting: its value at the point of a function's
definition is recorded so that the equation lemmas later generated for
that function use the same value, regardless of the ambient option at
the use site.
### Mathlib adaption
A companion adaption branch (`lean-pr-testing-backward-defeq-attrib` on
mathlib4) builds cleanly against this PR and passes `lake test` without
warnings. Most adaption changes are scoped
`set_option backward.defeqAttrib.useBackward true in` additions on the
failing declarations; a small number of files needed proof-level edits
where the stored form of a `dsimp%`/`@[reassoc]`/`@[elementwise]`
/`@[simps]`/`@[to_app]`-generated lemma had drifted under the stricter
regime.
---------
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This PR sets up the new integrated test/bench suite. It then migrates
all benchmarks and some related tests to the new suite. There's also
some documentation and some linting.
For now, a lot of the old tests are left alone so this PR doesn't become
even larger than it already is. Eventually, all tests should be migrated
to the new suite though so there isn't a confusing mix of two systems.
