An Alternative functor is an Applicative functor that can "fail" or be "empty" and a binary operation <|> that “collects values” or finds the “left-most success”.

Important instances include

Option, where failure := none and <|> returns the left-most some.
Parser combinators typically provide an Applicative instance for error-handling and backtracking.

Error recovery and state can interact subtly. For example, the implementation of Alternative for OptionT (StateT σ Id) keeps modifications made to the state while recovering from failure, while StateT σ (OptionT Id) discards them.

map : {α β : Type u} → (α → β) → f α → f β
mapConst : {α β : Type u} → α → f β → f α
pure : {α : Type u} → α → f α
seq : {α β : Type u} → f (α → β) → (Unit → f α) → f β
seqLeft : {α β : Type u} → f α → (Unit → f β) → f α
seqRight : {α β : Type u} → f α → (Unit → f β) → f β
failure : {α : Type u} → f α
Produces an empty collection or recoverable failure. The <|> operator collects values or recovers from failures. See Alternative for more details.
orElse : {α : Type u} → f α → (Unit → f α) → f α
Depending on the Alternative instance, collects values or recovers from failures by returning the leftmost success. Can be written using the <|> operator syntax.

Instances

source

instance instOrElseOfAlternative (f : Type u → Type v) (α : Type u) [Alternative f] :

OrElse (f α)

Equations

instOrElseOfAlternative f α = { orElse := Alternative.orElse }

source

@[inline]

def guard {f : Type → Type v} [Alternative f] (p : Prop) [Decidable p] :

f Unit

If the proposition p is true, does nothing, else fails (using failure).

Equations

guard p = if p then pure () else failure

Instances For

source

@[inline]

def optional {f : Type u → Type v} [Alternative f] {α : Type u} (x : f α) :

f (Option α)

Returns some x if f succeeds with value x, else returns none.

Equations

optional x = HOrElse.hOrElse (some <$> x) fun (x : Unit) => pure none

Instances For

source

class ToBool (α : Type u) :

Type u

toBool : α → Bool

Instances

source

instance instToBoolBool :

ToBool Bool

Equations

instToBoolBool = { toBool := fun (b : Bool) => b }

source

@[macro_inline]

def bool {β : Type u} {α : Type v} [ToBool β] (f : α) (t : α) (b : β) :

Equations

bool f t b = match toBool b with | true => t | false => f

Instances For

source

@[macro_inline]

def orM {m : Type u → Type v} {β : Type u} [Monad m] [ToBool β] (x : m β) (y : m β) :

m β

Equations

(x <||> y) = do let b ← x match toBool b with | true => pure b | false => y

Instances For

source

def «term_<||>_» :

Lean.TrailingParserDescr

Equations

«term_<||>_» = Lean.ParserDescr.trailingNode `term_<||>_ 30 31 (Lean.ParserDescr.binary `andthen (Lean.ParserDescr.symbol " <||> ") (Lean.ParserDescr.cat `term 30))

Instances For

source

@[macro_inline]

def andM {m : Type u → Type v} {β : Type u} [Monad m] [ToBool β] (x : m β) (y : m β) :

m β

Equations

(x <&&> y) = do let b ← x match toBool b with | true => y | false => pure b

Instances For

source

def «term_<&&>_» :

Lean.TrailingParserDescr

Equations

One or more equations did not get rendered due to their size.

Instances For

source

@[macro_inline]

def notM {m : Type → Type v} [Applicative m] (x : m Bool) :

m Bool

Equations

notM x = not <$> x

Instances For

How `MonadControl` works #

There is a tutorial by Alexis King that this docstring is based on.

Suppose we have foo : ∀ α, IO α → IO α and bar : StateT σ IO β (ie, bar : σ → IO (σ × β)). We might want to 'map' bar by foo. Concretely we would write this as:

opaque foo : ∀ {α}, IO α → IO α
opaque bar : StateT σ IO β

def mapped_foo : StateT σ IO β := do
  let s ← get
  let (b, s') ← liftM <| foo <| StateT.run bar s
  set s'
  return b

This is fine but it's not going to generalise, what if we replace StateT Nat IO with a large tower of monad transformers? We would have to rewrite the above to handle each of the run functions for each transformer in the stack.

Is there a way to generalise run as a kind of inverse of lift? We have lift : m α → StateT σ m α for all m, but we also need to 'unlift' the state. But unlift : StateT σ IO α → IO α can't be implemented. So we need something else.

If we look at the definition of mapped_foo, we see that lift <| foo <| StateT.run bar s has the type IO (σ × β). The key idea is that σ × β contains all of the information needed to reconstruct the state and the new value.

Now lets define some values to generalise mapped_foo:

Write IO (σ × β) as IO (stM β)
Write StateT.run . s as mapInBase : StateT σ IO α → IO (stM β)
Define restoreM : IO (stM α) → StateT σ IO α as below

def stM (α : Type) := α × σ

def restoreM (x : IO (stM α)) : StateT σ IO α := do
  let (a,s) ← liftM x
  set s
  return a

To get:

def mapped_foo' : StateT σ IO β := do
  let s ← get
  let mapInBase := fun z => StateT.run z s
  restoreM <| foo <| mapInBase bar

and finally define

def control {α : Type}
  (f : ({β : Type} → StateT σ IO β → IO (stM β)) → IO (stM α))
  : StateT σ IO α := do
  let s ← get
  let mapInBase := fun {β} (z : StateT σ IO β) => StateT.run z s
  let r : IO (stM α) := f mapInBase
  restoreM r

Now we can write mapped_foo as:

def mapped_foo'' : StateT σ IO β :=
  control (fun mapInBase => foo (mapInBase bar))

The core idea of mapInBase is that given any β, it runs an instance of StateT σ IO β and 'packages' the result and state as IO (stM β) so that it can be piped through foo. Once it's been through foo we can then unpack the state again with restoreM. Hence we can apply foo to bar without losing track of the state.

Here stM β = σ × β is the 'packaged result state', but we can generalise: if we have a tower StateT σ₁ <| StateT σ₂ <| IO, then the composite packaged state is going to be stM₁₂ β := σ₁ × σ₂ × β or stM₁₂ := stM₁ ∘ stM₂.

MonadControl m n means that when programming in the monad n, we can switch to a base monad m using control, just like with liftM. In contrast to liftM, however, we also get a function runInBase that allows us to "lower" actions in n into m. This is really useful when we have large towers of monad transformers, as we do in the metaprogramming library.

For example there is a function withNewMCtxDepthImp : MetaM α → MetaM α that runs the input monad instance in a new nested metavariable context. We can lift this to withNewMctxDepth : n α → n α using MonadControlT MetaM n (MonadControlT is the transitive closure of MonadControl). Which means that we can also run withNewMctxDepth in the Tactic monad without needing to faff around with lifts and all the other boilerplate needed in mapped_foo.

Relationship to `MonadFunctor` #

A stricter form of MonadControl is MonadFunctor, which defines monadMap {α} : (∀ {β}, m β → m β) → n α → n α. Using monadMap it is also possible to define mapped_foo above. However there are some mappings which can't be derived using MonadFunctor. For example:

 @[inline] def map1MetaM [MonadControlT MetaM n] [Monad n] (f : forall {α}, (β → MetaM α) → MetaM α) {α} (k : β → n α) : n α :=
   control fun runInBase => f fun b => runInBase <| k b

 @[inline] def map2MetaM [MonadControlT MetaM n] [Monad n] (f : forall {α}, (β → γ → MetaM α) → MetaM α) {α} (k : β → γ → n α) : n α :=
   control fun runInBase => f fun b c => runInBase <| k b c

In monadMap, we can only 'run in base' a single computation in n into the base monad m. Using control means that runInBase can be used multiple times.

source

class MonadControl (m : semiOutParam (Type u → Type v)) (n : Type u → Type w) :

Type (max (max (u + 1) v) w)

MonadControl is a way of stating that the monad m can be 'run inside' the monad n.

This is the same as MonadBaseControl in Haskell. To learn about MonadControl, see the comment above this docstring.

stM : Type u → Type u
liftWith : {α : Type u} → (({β : Type u} → n β → m (MonadControl.stM m n β)) → m α) → n α
restoreM : {α : Type u} → m (MonadControl.stM m n α) → n α

Instances

source

class MonadControlT (m : Type u → Type v) (n : Type u → Type w) :

Type (max (max (u + 1) v) w)

Transitive closure of MonadControl.

stM : Type u → Type u
liftWith : {α : Type u} → (({β : Type u} → n β → m (stM m n β)) → m α) → n α
restoreM : {α : Type u} → stM m n α → n α

Instances

source

@[always_inline]

instance instMonadControlTOfMonadControl (m : Type u_1 → Type u_2) (n : Type u_1 → Type u_3) (o : Type u_1 → Type u_4) [MonadControl n o] [MonadControlT m n] :

MonadControlT m o

Equations

One or more equations did not get rendered due to their size.

source

instance instMonadControlTOfPure (m : Type u → Type v) [Pure m] :

MonadControlT m m

Equations

One or more equations did not get rendered due to their size.

source

@[inline]

def controlAt (m : Type u → Type v) {n : Type u → Type w} [MonadControlT m n] [Bind n] {α : Type u} (f : ({β : Type u} → n β → m (stM m n β)) → m (stM m n α)) :

n α

Equations

controlAt m f = liftWith f >>= restoreM

Instances For

source

@[inline]

def control {m : Type u → Type v} {n : Type u → Type w} [MonadControlT m n] [Bind n] {α : Type u} (f : ({β : Type u} → n β → m (stM m n β)) → m (stM m n α)) :

n α

Equations

control f = controlAt m f

Instances For

source

class ForM (m : Type u → Type v) (γ : Type w₁) (α : outParam (Type w₂)) :

Type (max (max (max (u + 1) v) w₁) w₂)

Typeclass for the polymorphic forM operation described in the "do unchained" paper. Remark:

γ is a "container" type of elements of type α.
α is treated as an output parameter by the typeclass resolution procedure. That is, it tries to find an instance using only m and γ.