feat: generalize elab_as_elim to allow arbitrary motive applications (#5510)
Now the elab-as-elim procedure allows eliminators whose result is an
arbitrary application of the motive. For example, the following is now
accepted. It will generalize `Int.natAbs _` from the expected type.
```lean
@[elab_as_elim]
theorem natAbs_elim {motive : Nat → Prop} (i : Int)
(hpos : ∀ (n : Nat), i = n → motive n)
(hneg : ∀ (n : Nat), i = -↑n → motive n) :
motive (Int.natAbs i) := by sorry
```
This change simplifies the elaborator, since it no longer needs to keep
track of discriminants (which can easily be read off from the return
type of the eliminator) or the difference between "targets" and "extra
arguments" (which are now both "major arguments" that should be eagerly
elaborated).
Closes #4086
This commit is contained in:
parent
ccdf07b6a1
commit
130b465aaf
2 changed files with 153 additions and 89 deletions
|
|
@ -721,6 +721,104 @@ end
|
|||
|
||||
end ElabAppArgs
|
||||
|
||||
|
||||
/-! # Eliminator-like function application elaborator -/
|
||||
|
||||
/--
|
||||
Information about an eliminator used by the elab-as-elim elaborator.
|
||||
This is not to be confused with `Lean.Meta.ElimInfo`, which is for `induction` and `cases`.
|
||||
The elab-as-elim routine is less restrictive in what counts as an eliminator, and it doesn't need
|
||||
to have a strict notion of what is a "target" — all it cares about are
|
||||
1. that the return type of a function is of the form `m ...` where `m` is a parameter
|
||||
(unlike `induction` and `cases` eliminators, the arguments to `m`, known as "discriminants",
|
||||
can be any expressions, not just parameters), and
|
||||
2. which arguments should be eagerly elaborated, to make discriminants be as elaborated as
|
||||
possible for the expected type generalization procedure,
|
||||
and which should be postponed (since they are the "minor premises").
|
||||
|
||||
Note that the routine isn't doing induction/cases *on* particular expressions.
|
||||
The purpose of elab-as-elim is to successfully solve the higher-order unification problem
|
||||
between the return type of the function and the expected type.
|
||||
-/
|
||||
structure ElabElimInfo where
|
||||
/-- The eliminator. -/
|
||||
elimExpr : Expr
|
||||
/-- The type of the eliminator. -/
|
||||
elimType : Expr
|
||||
/-- The position of the motive parameter. -/
|
||||
motivePos : Nat
|
||||
/--
|
||||
Positions of "major" parameters (those that should be eagerly elaborated
|
||||
because they can contribute to the motive inference procedure).
|
||||
All parameters that are neither the motive nor a major parameter are "minor" parameters.
|
||||
The major parameters include all of the parameters that transitively appear in the motive's arguments,
|
||||
as well as "first-order" arguments that include such parameters,
|
||||
since they too can help with elaborating discriminants.
|
||||
|
||||
For example, in the following theorem the argument `h : a = b`
|
||||
should be elaborated eagerly because it contains `b`, which occurs in `motive b`.
|
||||
```
|
||||
theorem Eq.subst' {α} {motive : α → Prop} {a b : α} (h : a = b) : motive a → motive b
|
||||
```
|
||||
For another example, the term `isEmptyElim (α := α)` is an underapplied eliminator, and it needs
|
||||
argument `α` to be elaborated eagerly to create a type-correct motive.
|
||||
```
|
||||
def isEmptyElim [IsEmpty α] {p : α → Sort _} (a : α) : p a := ...
|
||||
example {α : Type _} [IsEmpty α] : id (α → False) := isEmptyElim (α := α)
|
||||
```
|
||||
-/
|
||||
majorsPos : Array Nat := #[]
|
||||
deriving Repr, Inhabited
|
||||
|
||||
def getElabElimExprInfo (elimExpr : Expr) : MetaM ElabElimInfo := do
|
||||
let elimType ← inferType elimExpr
|
||||
trace[Elab.app.elab_as_elim] "eliminator {indentExpr elimExpr}\nhas type{indentExpr elimType}"
|
||||
forallTelescopeReducing elimType fun xs type => do
|
||||
let motive := type.getAppFn
|
||||
let motiveArgs := type.getAppArgs
|
||||
unless motive.isFVar do
|
||||
throwError "unexpected eliminator resulting type{indentExpr type}"
|
||||
let motiveType ← inferType motive
|
||||
forallTelescopeReducing motiveType fun motiveParams motiveResultType => do
|
||||
unless motiveParams.size == motiveArgs.size do
|
||||
throwError "unexpected number of arguments at motive type{indentExpr motiveType}"
|
||||
unless motiveResultType.isSort do
|
||||
throwError "motive result type must be a sort{indentExpr motiveType}"
|
||||
let some motivePos ← pure (xs.indexOf? motive) |
|
||||
throwError "unexpected eliminator type{indentExpr elimType}"
|
||||
/-
|
||||
Compute transitive closure of fvars appearing in arguments to the motive.
|
||||
These are the primary set of major parameters.
|
||||
-/
|
||||
let initMotiveFVars : CollectFVars.State := motiveArgs.foldl (init := {}) collectFVars
|
||||
let motiveFVars ← xs.size.foldRevM (init := initMotiveFVars) fun i s => do
|
||||
let x := xs[i]!
|
||||
if s.fvarSet.contains x.fvarId! then
|
||||
return collectFVars s (← inferType x)
|
||||
else
|
||||
return s
|
||||
/- Collect the major parameter positions -/
|
||||
let mut majorsPos := #[]
|
||||
for i in [:xs.size] do
|
||||
let x := xs[i]!
|
||||
unless motivePos == i do
|
||||
let xType ← x.fvarId!.getType
|
||||
/-
|
||||
We also consider "first-order" types because we can reliably "extract" information from them.
|
||||
We say a term is "first-order" if all applications are of the form `f ...` where `f` is a constant.
|
||||
-/
|
||||
let isFirstOrder (e : Expr) : Bool := Option.isNone <| e.find? fun e => e.isApp && !e.getAppFn.isConst
|
||||
if motiveFVars.fvarSet.contains x.fvarId!
|
||||
|| (isFirstOrder xType
|
||||
&& Option.isSome (xType.find? fun e => e.isFVar && motiveFVars.fvarSet.contains e.fvarId!)) then
|
||||
majorsPos := majorsPos.push i
|
||||
trace[Elab.app.elab_as_elim] "motivePos: {motivePos}"
|
||||
trace[Elab.app.elab_as_elim] "majorsPos: {majorsPos}"
|
||||
return { elimExpr, elimType, motivePos, majorsPos }
|
||||
|
||||
def getElabElimInfo (elimName : Name) : MetaM ElabElimInfo := do
|
||||
getElabElimExprInfo (← mkConstWithFreshMVarLevels elimName)
|
||||
|
||||
builtin_initialize elabAsElim : TagAttribute ←
|
||||
registerTagAttribute `elab_as_elim
|
||||
"instructs elaborator that the arguments of the function application should be elaborated as were an eliminator"
|
||||
|
|
@ -735,33 +833,15 @@ builtin_initialize elabAsElim : TagAttribute ←
|
|||
let info ← getConstInfo declName
|
||||
if (← hasOptAutoParams info.type) then
|
||||
throwError "[elab_as_elim] attribute cannot be used in declarations containing optional and auto parameters"
|
||||
discard <| getElimInfo declName
|
||||
discard <| getElabElimInfo declName
|
||||
go.run' {} {}
|
||||
|
||||
/-! # Eliminator-like function application elaborator -/
|
||||
namespace ElabElim
|
||||
|
||||
/-- Context of the `elab_as_elim` elaboration procedure. -/
|
||||
structure Context where
|
||||
elimInfo : ElimInfo
|
||||
elimInfo : ElabElimInfo
|
||||
expectedType : Expr
|
||||
/--
|
||||
Position of additional arguments that should be elaborated eagerly
|
||||
because they can contribute to the motive inference procedure.
|
||||
For example, in the following theorem the argument `h : a = b`
|
||||
should be elaborated eagerly because it contains `b` which occurs
|
||||
in `motive b`.
|
||||
```
|
||||
theorem Eq.subst' {α} {motive : α → Prop} {a b : α} (h : a = b) : motive a → motive b
|
||||
```
|
||||
For another example, the term `isEmptyElim (α := α)` is an underapplied eliminator, and it needs
|
||||
argument `α` to be elaborated eagerly to create a type-correct motive.
|
||||
```
|
||||
def isEmptyElim [IsEmpty α] {p : α → Sort _} (a : α) : p a := ...
|
||||
example {α : Type _} [IsEmpty α] : id (α → False) := isEmptyElim (α := α)
|
||||
```
|
||||
-/
|
||||
extraArgsPos : Array Nat
|
||||
|
||||
/-- State of the `elab_as_elim` elaboration procedure. -/
|
||||
structure State where
|
||||
|
|
@ -773,8 +853,6 @@ structure State where
|
|||
namedArgs : List NamedArg
|
||||
/-- User-provided arguments that still have to be processed. -/
|
||||
args : List Arg
|
||||
/-- Discriminants (targets) processed so far. -/
|
||||
discrs : Array (Option Expr)
|
||||
/-- Instance implicit arguments collected so far. -/
|
||||
instMVars : Array MVarId := #[]
|
||||
/-- Position of the next argument to be processed. We use it to decide whether the argument is the motive or a discriminant. -/
|
||||
|
|
@ -820,7 +898,7 @@ def finalize : M Expr := do
|
|||
let some motive := (← get).motive?
|
||||
| throwError "failed to elaborate eliminator, insufficient number of arguments"
|
||||
trace[Elab.app.elab_as_elim] "motive: {motive}"
|
||||
forallTelescope (← get).fType fun xs _ => do
|
||||
forallTelescope (← get).fType fun xs fType => do
|
||||
trace[Elab.app.elab_as_elim] "xs: {xs}"
|
||||
let mut expectedType := (← read).expectedType
|
||||
trace[Elab.app.elab_as_elim] "expectedType:{indentD expectedType}"
|
||||
|
|
@ -829,6 +907,7 @@ def finalize : M Expr := do
|
|||
let mut f := (← get).f
|
||||
if xs.size > 0 then
|
||||
-- under-application, specialize the expected type using `xs`
|
||||
-- Note: if we ever wanted to support optParams and autoParams, this is where it could be.
|
||||
assert! (← get).args.isEmpty
|
||||
for x in xs do
|
||||
let .forallE _ t b _ ← whnf expectedType | throwInsufficient
|
||||
|
|
@ -845,18 +924,11 @@ def finalize : M Expr := do
|
|||
trace[Elab.app.elab_as_elim] "expectedType after processing:{indentD expectedType}"
|
||||
let result := mkAppN f xs
|
||||
trace[Elab.app.elab_as_elim] "result:{indentD result}"
|
||||
let mut discrs := (← get).discrs
|
||||
let idx := (← get).idx
|
||||
if discrs.any Option.isNone then
|
||||
for i in [idx:idx + xs.size], x in xs do
|
||||
if let some tidx := (← read).elimInfo.targetsPos.indexOf? i then
|
||||
discrs := discrs.set! tidx x
|
||||
if let some idx := discrs.findIdx? Option.isNone then
|
||||
-- This should not happen.
|
||||
trace[Elab.app.elab_as_elim] "Internal error, missing target with index {idx}"
|
||||
throwError "failed to elaborate eliminator, insufficient number of arguments"
|
||||
trace[Elab.app.elab_as_elim] "discrs: {discrs.map Option.get!}"
|
||||
let motiveVal ← mkMotive (discrs.map Option.get!) expectedType
|
||||
unless fType.getAppFn == (← get).motive? do
|
||||
throwError "Internal error, eliminator target type isn't an application of the motive"
|
||||
let discrs := fType.getAppArgs
|
||||
trace[Elab.app.elab_as_elim] "discrs: {discrs}"
|
||||
let motiveVal ← mkMotive discrs expectedType
|
||||
unless (← isTypeCorrect motiveVal) do
|
||||
throwError "failed to elaborate eliminator, motive is not type correct:{indentD motiveVal}"
|
||||
unless (← isDefEq motive motiveVal) do
|
||||
|
|
@ -890,10 +962,6 @@ def getNextArg? (binderName : Name) (binderInfo : BinderInfo) : M (LOption Arg)
|
|||
def setMotive (motive : Expr) : M Unit :=
|
||||
modify fun s => { s with motive? := motive }
|
||||
|
||||
/-- Push the given expression into the `discrs` field in the state, where `i` is which target it is for. -/
|
||||
def addDiscr (i : Nat) (discr : Expr) : M Unit :=
|
||||
modify fun s => { s with discrs := s.discrs.set! i discr }
|
||||
|
||||
/-- Elaborate the given argument with the given expected type. -/
|
||||
private def elabArg (arg : Arg) (argExpectedType : Expr) : M Expr := do
|
||||
match arg with
|
||||
|
|
@ -936,18 +1004,13 @@ partial def main : M Expr := do
|
|||
mkImplicitArg binderType binderInfo
|
||||
setMotive motive
|
||||
addArgAndContinue motive
|
||||
else if let some tidx := (← read).elimInfo.targetsPos.indexOf? idx then
|
||||
else if (← read).elimInfo.majorsPos.contains idx then
|
||||
match (← getNextArg? binderName binderInfo) with
|
||||
| .some arg => let discr ← elabArg arg binderType; addDiscr tidx discr; addArgAndContinue discr
|
||||
| .some arg => let discr ← elabArg arg binderType; addArgAndContinue discr
|
||||
| .undef => finalize
|
||||
| .none => let discr ← mkImplicitArg binderType binderInfo; addDiscr tidx discr; addArgAndContinue discr
|
||||
| .none => let discr ← mkImplicitArg binderType binderInfo; addArgAndContinue discr
|
||||
else match (← getNextArg? binderName binderInfo) with
|
||||
| .some (.stx stx) =>
|
||||
if (← read).extraArgsPos.contains idx then
|
||||
let arg ← elabArg (.stx stx) binderType
|
||||
addArgAndContinue arg
|
||||
else
|
||||
addArgAndContinue (← postponeElabTerm stx binderType)
|
||||
| .some (.stx stx) => addArgAndContinue (← postponeElabTerm stx binderType)
|
||||
| .some (.expr val) => addArgAndContinue (← ensureArgType (← get).f val binderType)
|
||||
| .undef => finalize
|
||||
| .none => addArgAndContinue (← mkImplicitArg binderType binderInfo)
|
||||
|
|
@ -1001,13 +1064,10 @@ def elabAppArgs (f : Expr) (namedArgs : Array NamedArg) (args : Array Arg)
|
|||
let some expectedType := expectedType? | throwError "failed to elaborate eliminator, expected type is not available"
|
||||
let expectedType ← instantiateMVars expectedType
|
||||
if expectedType.getAppFn.isMVar then throwError "failed to elaborate eliminator, expected type is not available"
|
||||
let extraArgsPos ← getElabAsElimExtraArgsPos elimInfo
|
||||
trace[Elab.app.elab_as_elim] "extraArgsPos: {extraArgsPos}"
|
||||
ElabElim.main.run { elimInfo, expectedType, extraArgsPos } |>.run' {
|
||||
ElabElim.main.run { elimInfo, expectedType } |>.run' {
|
||||
f, fType
|
||||
args := args.toList
|
||||
namedArgs := namedArgs.toList
|
||||
discrs := mkArray elimInfo.targetsPos.size none
|
||||
}
|
||||
else
|
||||
ElabAppArgs.main.run { explicit, ellipsis, resultIsOutParamSupport } |>.run' {
|
||||
|
|
@ -1018,12 +1078,12 @@ def elabAppArgs (f : Expr) (namedArgs : Array NamedArg) (args : Array Arg)
|
|||
}
|
||||
where
|
||||
/-- Return `some info` if we should elaborate as an eliminator. -/
|
||||
elabAsElim? : TermElabM (Option ElimInfo) := do
|
||||
elabAsElim? : TermElabM (Option ElabElimInfo) := do
|
||||
unless (← read).heedElabAsElim do return none
|
||||
if explicit || ellipsis then return none
|
||||
let .const declName _ := f | return none
|
||||
unless (← shouldElabAsElim declName) do return none
|
||||
let elimInfo ← getElimInfo declName
|
||||
let elimInfo ← getElabElimInfo declName
|
||||
forallTelescopeReducing (← inferType f) fun xs _ => do
|
||||
/- Process arguments similar to `Lean.Elab.Term.ElabElim.main` to see if the motive has been
|
||||
provided, in which case we use the standard app elaborator.
|
||||
|
|
@ -1054,41 +1114,6 @@ where
|
|||
return none
|
||||
| _, _ => return some elimInfo
|
||||
|
||||
/--
|
||||
Collect extra argument positions that must be elaborated eagerly when using `elab_as_elim`.
|
||||
The idea is that they contribute to motive inference. See comment at `ElamElim.Context.extraArgsPos`.
|
||||
-/
|
||||
getElabAsElimExtraArgsPos (elimInfo : ElimInfo) : MetaM (Array Nat) := do
|
||||
forallTelescope elimInfo.elimType fun xs type => do
|
||||
let targets := type.getAppArgs
|
||||
/- Compute transitive closure of fvars appearing in the motive and the targets. -/
|
||||
let initMotiveFVars : CollectFVars.State := targets.foldl (init := {}) collectFVars
|
||||
let motiveFVars ← xs.size.foldRevM (init := initMotiveFVars) fun i s => do
|
||||
let x := xs[i]!
|
||||
if elimInfo.motivePos == i || elimInfo.targetsPos.contains i || s.fvarSet.contains x.fvarId! then
|
||||
return collectFVars s (← inferType x)
|
||||
else
|
||||
return s
|
||||
/- Collect the extra argument positions -/
|
||||
let mut extraArgsPos := #[]
|
||||
for i in [:xs.size] do
|
||||
let x := xs[i]!
|
||||
unless elimInfo.motivePos == i || elimInfo.targetsPos.contains i do
|
||||
let xType ← x.fvarId!.getType
|
||||
/- We only consider "first-order" types because we can reliably "extract" information from them. -/
|
||||
if motiveFVars.fvarSet.contains x.fvarId!
|
||||
|| (isFirstOrder xType
|
||||
&& Option.isSome (xType.find? fun e => e.isFVar && motiveFVars.fvarSet.contains e.fvarId!)) then
|
||||
extraArgsPos := extraArgsPos.push i
|
||||
return extraArgsPos
|
||||
|
||||
/-
|
||||
Helper function for implementing `elab_as_elim`.
|
||||
We say a term is "first-order" if all applications are of the form `f ...` where `f` is a constant.
|
||||
-/
|
||||
isFirstOrder (e : Expr) : Bool :=
|
||||
Option.isNone <| e.find? fun e =>
|
||||
e.isApp && !e.getAppFn.isConst
|
||||
|
||||
/-- Auxiliary inductive datatype that represents the resolution of an `LVal`. -/
|
||||
inductive LValResolution where
|
||||
|
|
|
|||
39
tests/lean/run/4086.lean
Normal file
39
tests/lean/run/4086.lean
Normal file
|
|
@ -0,0 +1,39 @@
|
|||
/-!
|
||||
# Generalize `elab_as_elim` to allow arbitrary motive applications
|
||||
-/
|
||||
|
||||
/-!
|
||||
The following eliminator isn't valid for `induction`/`cases` since the return type
|
||||
has `motive` applied to `Int.natAbs i`, which isn't a parameter,
|
||||
but this is not a problem for `elab_as_elim`.
|
||||
-/
|
||||
|
||||
@[elab_as_elim]
|
||||
theorem natAbs_elim {motive : Nat → Prop} (i : Int)
|
||||
(hpos : ∀ (n : Nat), i = n → motive n)
|
||||
(hneg : ∀ (n : Nat), i = -↑n → motive n) :
|
||||
motive (Int.natAbs i) := by sorry
|
||||
|
||||
example (x : Int) (y : Nat) : x.natAbs < y := by
|
||||
refine natAbs_elim x ?_ ?_
|
||||
· guard_target = ∀ (n : Nat), x = ↑n → n < y
|
||||
sorry
|
||||
· guard_target = ∀ (n : Nat), x = -↑n → n < y
|
||||
sorry
|
||||
|
||||
example (x : Int) (y : Nat) : (x + 1).natAbs + 1 < y := by
|
||||
refine natAbs_elim (x + 1) ?_ ?_
|
||||
· guard_target = ∀ (n : Nat), x + 1 = ↑n → n + 1 < y
|
||||
sorry
|
||||
· guard_target = ∀ (n : Nat), x + 1 = -↑n → n + 1 < y
|
||||
sorry
|
||||
|
||||
/-!
|
||||
The target can be inferred from the expected type.
|
||||
-/
|
||||
example (x : Int) (y : Nat) : (x + 1).natAbs < y := by
|
||||
refine natAbs_elim _ ?_ ?_
|
||||
· guard_target = ∀ (n : Nat), x + 1 = ↑n → n < y
|
||||
sorry
|
||||
· guard_target = ∀ (n : Nat), x + 1 = -↑n → n < y
|
||||
sorry
|
||||
Loading…
Add table
Reference in a new issue