/- Copyright (c) 2018 Microsoft Corporation. All rights reserved. Released under Apache 2.0 license as described in the file LICENSE. Author: Sebastian Ullrich Syntax-tree creating parser combinators -/ prelude import init.lean.parser.basic import init.data.list.instances namespace lean namespace parser namespace combinators open hasTokens hasView monadParsec variables {α : Type} {m : Type → Type} local notation `parser` := m syntax variables [monad m] [monadExcept (parsec.message syntax) m] [monadParsec syntax m] [alternative m] def node (k : syntaxNodeKind) (rs : list parser) : parser := do args ← rs.mfoldl (λ (args : list syntax) r, do -- on error, append partial syntax tree to previous successful parses and rethrow a ← catch r $ λ msg, match args with -- do not wrap an error in the first argument to uphold the invariant documented at `syntaxNode` | [] := throw msg | _ := let args := msg.custom.get :: args in throw {msg with custom := syntax.mkNode k args.reverse}, pure (a::args) ) [], pure $ syntax.mkNode k args.reverse @[reducible] def seq : list parser → parser := node noKind instance node.tokens (k) (rs : list parser) [parser.hasTokens rs] : parser.hasTokens (node k rs) := ⟨tokens rs⟩ instance node.view (k) (rs : list parser) [i : hasView α k] : parser.hasView α (node k rs) := { view := i.view, review := i.review } -- Each parser combinator comes equipped with `hasView` and `hasTokens` instances private def many1Aux (p : parser) : list syntax → nat → parser | as 0 := error "unreachable" | as (n+1) := do a ← catch p (λ msg, throw {msg with custom := -- append `syntax.missing` to make clear that list is incomplete syntax.list (syntax.missing::msg.custom.get::as).reverse}), many1Aux (a::as) n <|> pure (syntax.list (a::as).reverse) def many1 (r : parser) : parser := do rem ← remaining, many1Aux r [] (rem+1) instance many1.tokens (r : parser) [parser.hasTokens r] : parser.hasTokens (many1 r) := ⟨tokens r⟩ instance many1.view (r : parser) [parser.hasView α r] : parser.hasView (list α) (many1 r) := { view := λ stx, match stx.asNode with | some n := n.args.map (hasView.view r) | _ := [hasView.view r syntax.missing], review := λ as, syntax.list $ as.map (review r) } def many (r : parser) : parser := many1 r <|> pure (syntax.list []) instance many.tokens (r : parser) [parser.hasTokens r] : parser.hasTokens (many r) := ⟨tokens r⟩ instance many.view (r : parser) [hasView α r] : parser.hasView (list α) (many r) := /- Remark: `many1.view` can handle empty list. -/ {..many1.view r} private def sepByAux (p : m syntax) (sep : parser) (allowTrailingSep : bool) : bool → list syntax → nat → parser | pOpt as 0 := error "unreachable" | pOpt as (n+1) := do let p := if pOpt then some <$> p <|> pure none else some <$> p, some a ← catch p (λ msg, throw {msg with custom := -- append `syntax.missing` to make clear that list is incomplete syntax.list (syntax.missing::msg.custom.get::as).reverse}) | pure (syntax.list as.reverse), -- I don't want to think about what the output on a failed separator parse should look like let sep := try sep, some s ← some <$> sep <|> pure none | pure (syntax.list (a::as).reverse), sepByAux allowTrailingSep (s::a::as) n def sepBy (p sep : parser) (allowTrailingSep := tt) : parser := do rem ← remaining, sepByAux p sep allowTrailingSep tt [] (rem+1) def sepBy1 (p sep : parser) (allowTrailingSep := tt) : parser := do rem ← remaining, sepByAux p sep allowTrailingSep ff [] (rem+1) instance sepBy.tokens (p sep : parser) (a) [parser.hasTokens p] [parser.hasTokens sep] : parser.hasTokens (sepBy p sep a) := ⟨tokens p ++ tokens sep⟩ structure sepBy.elem.view (α β : Type) := (item : α) (separator : option β := none) instance sepBy.elem.view.itemCoe {α β : Type} : hasCoeT α (sepBy.elem.view α β) := ⟨λ a, ⟨a, none⟩⟩ private def sepBy.viewAux {α β} (p sep : parser) [parser.hasView α p] [parser.hasView β sep] : list syntax → list (sepBy.elem.view α β) | [] := [] | [stx] := [⟨hasView.view p stx, none⟩] | (stx1::stx2::stxs) := ⟨hasView.view p stx1, some $ hasView.view sep stx2⟩::sepBy.viewAux stxs instance sepBy.view {α β} (p sep : parser) (a) [parser.hasView α p] [parser.hasView β sep] : parser.hasView (list (sepBy.elem.view α β)) (sepBy p sep a) := { view := λ stx, match stx.asNode with | some n := sepBy.viewAux p sep n.args | _ := [⟨view p syntax.missing, none⟩], review := λ as, syntax.list $ as.bind (λ a, match a with | ⟨v, some vsep⟩ := [review p v, review sep vsep] | ⟨v, none⟩ := [review p v]) } instance sepBy1.tokens (p sep : parser) (a) [parser.hasTokens p] [parser.hasTokens sep] : parser.hasTokens (sepBy1 p sep a) := ⟨tokens (sepBy p sep a)⟩ instance sepBy1.view {α β} (p sep : parser) (a) [parser.hasView α p] [parser.hasView β sep] : parser.hasView (list (sepBy.elem.view α β)) (sepBy1 p sep a) := {..sepBy.view p sep a} def optional (r : parser) : parser := do r ← optional $ -- on error, wrap in "some" catch r (λ msg, throw {msg with custom := syntax.list [msg.custom.get]}), pure $ match r with | some r := syntax.list [r] | none := syntax.list [] instance optional.tokens (r : parser) [parser.hasTokens r] : parser.hasTokens (optional r) := ⟨tokens r⟩ instance optional.view (r : parser) [parser.hasView α r] : parser.hasView (option α) (optional r) := { view := λ stx, match stx.asNode with | some {args := [], ..} := none | some {args := [stx], ..} := some $ hasView.view r stx | _ := some $ view r syntax.missing, review := λ a, match a with | some a := syntax.list [review r a] | none := syntax.list [] } instance optional.viewDefault (r : parser) [parser.hasView α r] : parser.hasViewDefault (optional r) (option α) none := ⟨⟩ /-- Parse a list `[p1, ..., pn]` of parsers as `p1 <|> ... <|> pn`. Note that there is NO explicit encoding of which parser was chosen; parsers should instead produce distinct node names for disambiguation. -/ def anyOf (rs : list parser) : parser := match rs with | [] := error "anyOf" | (r::rs) := rs.foldl (<|>) r instance anyOf.tokens (rs : list parser) [parser.hasTokens rs] : parser.hasTokens (anyOf rs) := ⟨tokens rs⟩ instance anyOf.view (rs : list parser) : parser.hasView syntax (anyOf rs) := default _ /-- Parse a list `[p1, ..., pn]` of parsers with `monadParsec.longestMatch`. If the result is ambiguous, wrap it in a `choice` node. Note that there is NO explicit encoding of which parser was chosen; parsers should instead produce distinct node names for disambiguation. -/ def longestMatch (rs : list parser) : parser := do stxs ← monadParsec.longestMatch rs, match stxs with | [stx] := pure stx | _ := pure $ syntax.mkNode choice stxs instance longestMatch.tokens (rs : list parser) [parser.hasTokens rs] : parser.hasTokens (longestMatch rs) := ⟨tokens rs⟩ instance longestMatch.view (rs : list parser) : parser.hasView syntax (longestMatch rs) := default _ def choiceAux : list parser → nat → parser | [] _ := error "choice: empty list" | (r::rs) i := do { stx ← r, pure $ syntax.mkNode ⟨name.mkNumeral name.anonymous i⟩ [stx] } <|> choiceAux rs (i+1) /-- Parse a list `[p1, ..., pn]` of parsers as `p1 <|> ... <|> pn`. The result will be wrapped in a node with the index of the successful parser as the name. Remark: Does not have a `hasView` instance because we only use it in `nodeChoice!` macros that define their own views. -/ def choice (rs : list parser) : parser := choiceAux rs 0 instance choice.tokens (rs : list parser) [parser.hasTokens rs] : parser.hasTokens (choice rs) := ⟨tokens rs⟩ /-- Like `choice`, but using `longestMatch`. Does not create choice nodes, prefers the first successful parser. -/ def longestChoice (rs : list parser) : parser := do stx::stxs ← monadParsec.longestMatch $ rs.enum.map $ λ ⟨i, r⟩, do { stx ← r, pure $ syntax.mkNode ⟨name.mkNumeral name.anonymous i⟩ [stx] } | error "unreachable", pure stx instance longestChoice.tokens (rs : list parser) [parser.hasTokens rs] : parser.hasTokens (longestChoice rs) := ⟨tokens rs⟩ instance try.tokens (r : parser) [parser.hasTokens r] : parser.hasTokens (try r) := ⟨tokens r⟩ instance try.view (r : parser) [i : parser.hasView α r] : parser.hasView α (try r) := {..i} instance label.tokens (r : parser) (l) [parser.hasTokens r] : parser.hasTokens (label r l) := ⟨tokens r⟩ instance label.view (r : parser) (l) [i : parser.hasView α r] : parser.hasView α (label r l) := {..i} instance recurse.tokens (α δ m a) [monadRec α δ m] : parser.hasTokens (recurse a : m δ) := default _ -- recursive use should not contribute any new tokens instance recurse.view (α δ m a) [monadRec α δ m] : parser.hasView syntax (recurse a : m δ) := default _ instance monadLift.tokens {m' : Type → Type} [hasMonadLiftT m m'] (r : m syntax) [parser.hasTokens r] : parser.hasTokens (monadLift r : m' syntax) := ⟨tokens r⟩ instance monadLift.view {m' : Type → Type} [hasMonadLiftT m m'] (r : m syntax) [i : parser.hasView α r] : parser.hasView α (monadLift r : m' syntax) := {..i} instance seqLeft.tokens {α : Type} (x : m α) (p : m syntax) [parser.hasTokens p] : parser.hasTokens (p <* x) := ⟨tokens p⟩ instance seqLeft.view {α β : Type} (x : m α) (p : m syntax) [i : parser.hasView β p] : parser.hasView β (p <* x) := {..i} instance seqRight.tokens {α : Type} (x : m α) (p : m syntax) [parser.hasTokens p] : parser.hasTokens (x *> p) := ⟨tokens p⟩ instance seqRight.view {α β : Type} (x : m α) (p : m syntax) [i : parser.hasView β p] : parser.hasView β (x *> p) := {..i} instance coe.tokens {β} (r : parser) [parser.hasTokens r] [hasCoeT parser β]: parser.hasTokens (coe r : β) := ⟨tokens r⟩ instance coe.view {β} (r : parser) [i : parser.hasView α r] [hasCoeT parser β] : parser.hasView α (coe r : β) := {..i} instance coe.viewDefault {β} (d : α) (r : parser) [hasView α r] [parser.hasViewDefault r α d] [hasCoeT parser β] : parser.hasViewDefault (coe r : β) α d := ⟨⟩ end combinators end parser end lean