This PR uses Lean.RArray in bv_decide's reflection proofs. Giving
speedups on problems with lots of variables.
Implement like #6068, speedup:
```
# before
λ hyperfine "lean +nightly-2024-12-02 tests/lean/run/bv_reflection_stress.lean"
Benchmark 1: lean +nightly-2024-12-02 tests/lean/run/bv_reflection_stress.lean
Time (mean ± σ): 1.939 s ± 0.007 s [User: 1.549 s, System: 0.104 s]
Range (min … max): 1.928 s … 1.947 s 10 runs
# after
λ hyperfine "lean tests/lean/run/bv_reflection_stress.lean"
Benchmark 1: lean tests/lean/run/bv_reflection_stress.lean
Time (mean ± σ): 1.409 s ± 0.006 s [User: 1.058 s, System: 0.073 s]
Range (min … max): 1.401 s … 1.419 s 10 runs
```
This PR runs all linters for a single command (together) on a separate
thread from further elaboration, making a first step towards
parallelizing the elaborator.
This PR ensure `bv_decide` uses definitional equality in its reflection
procedure as much as possible. Previously it would build up explicit
congruence proofs for the kernel to check. This reduces the size of
proof terms passed to kernel speeds up checking of large reflection
proofs.
This PR puts code in terms of syntax quotations now that there has been
a stage0 update. Fixes a lingering bug in StructInst where some
intermediate syntax was malformed, but this had no observable effects
outside of some debug messages.
This PR modifies structure instance notation and `where` notation to use
the same notation for fields. Structure instance notation now admits
binders, type ascriptions, and equations, and `where` notation admits
full structure lvals. Examples of these for structure instance notation:
```lean
structure PosFun where
f : Nat → Nat
pos : ∀ n, 0 < f n
def p : PosFun :=
{ f n := n + 1
pos := by simp }
def p' : PosFun :=
{ f | 0 => 1
| n + 1 => n + 1
pos := by rintro (_|_) <;> simp }
```
Just like for the structure `where` notation, a field `f x y z : ty :=
val` expands to `f := fun x y z => (val : ty)`. The type ascription is
optional.
The PR also is setting things up for future expansion. Pending some
discussion, in the future structure/`where` notation could have have
embedded `where` clauses; rather than `{ a := { x := 1, y := z } }` one
could write `{ a where x := 1; y := z }`.
This PR changes how generalized field notation ("dot notation") resolves
the function. The new resolution rule is that if `x : S`, then `x.f`
resolves the name `S.f` relative to the root namespace (hence it now
affected by `export` and `open`). Breaking change: aliases now resolve
differently. Before, if `x : S`, and if `S.f` is an alias for `S'.f`,
then `x.f` would use `S'.f` and look for an argument of type `S'`. Now,
it looks for an argument of type `S`, which is more generally useful
behavior. Code making use of the old behavior should consider defining
`S` or `S'` in terms of the other, since dot notation can unfold
definitions during resolution.
This also fixes a bug in explicit-mode generalized field notation
(`@x.f`) where `x` could be passed as the wrong argument. This was not a
bug for explicit-mode structure projections.
Closes#3031. Addresses the `Function` namespace issue in #1629.
This PR fixes a bug with the `structure`/`class` command where if there
are parents that are not represented as subobjects but which used other
parents as instances, then there would be a kernel error. Closes#2611.
Note: there is still the limitation that parents that are not
represented as subobjects do not themselves provide instances to other
parents.
This PR uses `Array.findFinIdx?` in preference to `Array.findIdx?` where
it allows converting a runtime bounds check to a compile time bounds
check.
(and some other minor cleanup)
This PR fixes a non-termination bug that occurred when generating the
match-expression equation theorems. The bug was triggered when the proof
automation for the equation theorem repeatedly applied `injection(` to
the same local declaration, as it could not be removed due to forward
dependencies. See issue #6067 for an example that reproduces this issue.
closes#6067
This PR adds core metaprogramming functions for forking off background
tasks from elaboration such that their results are visible to reporting
and the language server
This PR adds support for `structure` in `mutual` blocks, allowing
inductive types defined by `inductive` and `structure` to be mutually
recursive. The limitations are (1) that the parents in the `extends`
clause must be defined before the `mutual` block and (2) mutually
recursive classes are not allowed (a limitation shared by `class
inductive`). There are also improvements to universe level inference for
inductive types and structures. Breaking change: structure parents now
elaborate with the structure in scope (fix: use qualified names or
rename the structure to avoid shadowing), and structure parents no
longer elaborate with autoimplicits enabled.
Internally, this is a large refactor of both the `inductive` and
`structure` commands. Common material is now in
`Lean.Elab.MutualInductive`, and each command plugs into this mutual
inductive elaboration framework with the logic specific to the
respective command. For example, `structure` has code to add projections
after the inductive types are added to the environment.
Closes#4182
This PR modifies the signature of the functions `Nat.fold`,
`Nat.foldRev`, `Nat.any`, `Nat.all`, so that the function is passed the
upper bound. This allows us to change runtime array bounds checks to
compile time checks in many places.
This PR adds a slight performance improvement to reflection of `if`
statements that I noticed by profiling Leanwuzla against SMTCOMP's
`non-incremental/QF_BV/fft/Sz256_6616.smt2`.
In particular:
1. The profile showed about 4 percent of the total run time were spent
constructing these decidable instances in reflection of `if` statements.
We can construct them much quicker by hand as they always have the same
structure
2. This delays construction of these statements until we actually
generate the reflection proof that we wish to submit to the kernel. Thus
if we encounter a SAT instad of an UNSAT problem we will not spend time
generating these expressions anymore.
```
baseline
Time (mean ± σ): 31.236 s ± 0.258 s
Range (min … max): 30.899 s … 31.661 s 10 runs
after
Time (mean ± σ): 30.671 s ± 0.288 s
Range (min … max): 30.350 s … 31.156 s 10 runs
```
This PR avoids runtime array bounds checks in places where it can
trivially be done at compile time.
None of these changes are of particular consequence: I mostly wanted to
learn how much we do this, and what the obstacles are to doing it less.
This PR adds support for displaying multiple threads in the trace
profiler output.
`TraceState.tid` needs to be adjusted for this purpose, which is not
done yet by the Lean elaborator as it is still single-threaded.
This PR replaces `Array.feraseIdx` and `Array.insertAt` with
`Array.eraseIdx` and `Array.insertIdx`, both of which take a `Nat`
argument and a tactic-provided proof that it is in bounds. We also have
`eraseIdxIfInBounds` and `insertIdxIfInBounds` which are noops if the
index is out of bounds. We also provide a `Fin` valued version of
`Array.findIdx?`. Together, these quite ergonomically improve the array
indexing safety at a number of places in the compiler/elaborator.
This PR improves the `#print` command for structures to show all fields
and which parents the fields were inherited from, hiding internal
details such as which parents are represented as subobjects. This
information is still present in the constructor if needed. The pretty
printer for private constants is also improved, and it now handles
private names from the current module like any other name; private names
from other modules are made hygienic.
Example output for `#print Monad`:
```
class Monad.{u, v} (m : Type u → Type v) : Type (max (u + 1) v)
number of parameters: 1
parents:
Monad.toApplicative : Applicative m
Monad.toBind : Bind m
fields:
Functor.map : {α β : Type u} → (α → β) → m α → m β
Functor.mapConst : {α β : Type u} → α → m β → m α
Pure.pure : {α : Type u} → α → m α
Seq.seq : {α β : Type u} → m (α → β) → (Unit → m α) → m β
SeqLeft.seqLeft : {α β : Type u} → m α → (Unit → m β) → m α
SeqRight.seqRight : {α β : Type u} → m α → (Unit → m β) → m β
Bind.bind : {α β : Type u} → m α → (α → m β) → m β
constructor:
Monad.mk.{u, v} {m : Type u → Type v} [toApplicative : Applicative m] [toBind : Bind m] : Monad m
resolution order:
Monad, Applicative, Bind, Functor, Pure, Seq, SeqLeft, SeqRight
```
Suggested by Floris van Doorn [on
Zulip](https://leanprover.zulipchat.com/#narrow/channel/270676-lean4/topic/.23print.20command.20for.20structures/near/482503637).
This PR fixes a bug where structural recursion did not work when indices
of the recursive argument appeared as function parameters in a different
order than in the argument's type's definition.
Fixes#6015.
This PR liberalizes atom rules by allowing `''` to be a prefix of an
atom, after #6012 only added an exception for `''` alone, and also adds
some unit tests for atom validation.
This PR fixes the caching infrastructure for `whnf` and `isDefEq`,
ensuring the cache accounts for all relevant configuration flags. It
also cleans up the `WHNF.lean` module and improves the configuration of
`whnf`.
This PR fixes a stack overflow caused by a cyclic assignment in the
metavariable context. The cycle is unintentionally introduced by the
structure instance elaborator.
closes#3150
This PR makes the `change` tactic and conv tactic use the same
elaboration strategy. It works uniformly for both the target and local
hypotheses. Now `change` can assign metavariables, for example:
```lean
example (x y z : Nat) : x + y = z := by
change ?a = _
let w := ?a
-- now `w : Nat := x + y`
```
This PR improves the validation of new syntactic tokens. Previously, the
validation code had inconsistencies: some atoms would be accepted only
if they had a leading space as a pretty printer hint. Additionally,
atoms with internal whitespace are no longer allowed.
Closes#6011
This PR adds a new definition `Message.kind` which returns the top-level
tag of a message. This is serialized as the new field `kind` in
`SerialMessaege` so that i can be used by external consumers (e.g.,
Lake) to identify messages via `lean --json`.
The tag of trace messages has also been changed from `_traceMsg` to the
more friendly `trace`.
This PR simplifies the implementation of `omega`.
When constructing the proof, `omega` is using MVars only for the purpose
of doing case analysis on `Or`. We can simplify the implementation a
fair bit if we just produce the proof directly using `Or.elim`.
While it didn’t yield the performance benefits I was hoping for, this
still seems a worthwhile simplification, now that we already have it.
This PR changes the signature of `Array.get` to take a Nat and a proof,
rather than a `Fin`, for consistency with the rest of the (planned)
Array API. Note that because of bootstrapping issues we can't provide
`get_elem_tactic` as an autoparameter for the proof. As users will
mostly use the `xs[i]` notation provided by `GetElem`, this hopefully
isn't a problem.
We may restore `Fin` based versions, either here or downstream, as
needed, but they won't be the "main" functions.
---------
Co-authored-by: David Thrane Christiansen <david@davidchristiansen.dk>
This PR changes the rule for which projections become instances. Before,
all parents along with all indirect ancestors that were represented as
subobject fields would have their projections become instances. Now only
projections for direct parents become instances.
Features:
- Only parents that are not ancestors of other parents get instances.
This allows "discretionary" indirect parents to be inserted for the
purpose of computing strict resolution orders when
`structure.strictResolutionOrder` is enabled, without having an impact
on typeclass synthesis.
- Non-subobject projections are now theorems if the parent is a
proposition. These are also no longer `@[reducible]`.
Closes#2905
This PR fixes `bv_decide`'s embedded constraint substitution to generate
correct counter examples in the corner case where duplicate theorems are
in the local context.
