Merge pull request 'feat: Pickling goal states' (#129) from serial/pickle into dev

Reviewed-on: #129

Pantograph/Protocol.lean

@@ -289,6 +289,19 @@ structure GoalDiag where
   instantiate: Bool := true
   printSexp: Bool := false
 
+structure GoalSave where
+  id: Nat
+  path: System.FilePath
+  deriving Lean.FromJson
+structure GoalSaveResult where
+  deriving Lean.ToJson
+structure GoalLoad where
+  path: System.FilePath
+  deriving Lean.FromJson
+structure GoalLoadResult where
+  id: Nat
+  deriving Lean.ToJson
+
 
 /-- Executes the Lean compiler on a single file -/
 structure FrontendProcess where

Pantograph/Serial.lean

@@ -2,6 +2,7 @@ import Lean.Environment
 import Lean.Replay
 import Init.System.IOError
 import Std.Data.HashMap
+import Pantograph.Goal
 
 /-!
 Input/Output functions
@@ -55,7 +56,7 @@ and when unpickling, we build a fresh `Environment` from the imports,
 and then add the new constants.
 -/
 @[export pantograph_env_pickle_m]
-def env_pickle (env : Environment) (path : System.FilePath) : IO Unit :=
+def environmentPickle (env : Environment) (path : System.FilePath) : IO Unit :=
   Pantograph.pickle path (env.header.imports, env.constants.map₂)
 
 /--
@@ -65,9 +66,97 @@ We construct a fresh `Environment` with the relevant imports,
 and then replace the new constants.
 -/
 @[export pantograph_env_unpickle_m]
-def env_unpickle (path : System.FilePath) : IO (Environment × CompactedRegion) := unsafe do
+def environmentUnpickle (path : System.FilePath) : IO (Environment × CompactedRegion) := unsafe do
   let ((imports, map₂), region) ← Pantograph.unpickle (Array Import × PHashMap Name ConstantInfo) path
   let env ← importModules imports {} 0
   return (← env.replay (Std.HashMap.ofList map₂.toList), region)
 
+
+open Lean.Core in
+structure CompactCoreState where
+  -- env             : Environment
+  nextMacroScope  : MacroScope     := firstFrontendMacroScope + 1
+  ngen            : NameGenerator  := {}
+  -- traceState      : TraceState     := {}
+  -- cache           : Cache     := {}
+  -- messages        : MessageLog     := {}
+  -- infoState       : Elab.InfoState := {}
+
+@[export pantograph_goal_state_pickle_m]
+def goalStatePickle (goalState : GoalState) (path : System.FilePath) : IO Unit :=
+  let {
+    savedState := {
+      term := {
+        meta := {
+          core,
+          meta,
+        }
+        «elab»,
+      },
+      tactic
+    }
+    root,
+    parentMVar?,
+    convMVar?,
+    calcPrevRhs?,
+  } := goalState
+  --let env := core.env
+  Pantograph.pickle path (
+    ({ core with } : CompactCoreState),
+    meta,
+    «elab»,
+    tactic,
+
+    root,
+    parentMVar?,
+    convMVar?,
+    calcPrevRhs?,
+  )
+
+@[export pantograph_goal_state_unpickle_m]
+def goalStateUnpickle (path : System.FilePath) (env : Environment)
+    : IO (GoalState × CompactedRegion) := unsafe do
+  let ((
+    compactCore,
+    meta,
+    «elab»,
+    tactic,
+
+    root,
+    parentMVar?,
+    convMVar?,
+    calcPrevRhs?,
+  ), region) ← Pantograph.unpickle (
+    CompactCoreState ×
+    Meta.State ×
+    Elab.Term.State ×
+    Elab.Tactic.State ×
+
+    MVarId ×
+    Option MVarId ×
+    Option (MVarId × MVarId × List MVarId) ×
+    Option (MVarId × Expr)
+  ) path
+  let goalState := {
+    savedState := {
+      term := {
+        meta := {
+          core := {
+            compactCore with
+            passedHeartbeats := 0,
+            env,
+          },
+          meta,
+        },
+        «elab»,
+      },
+      tactic,
+    },
+    root,
+    parentMVar?,
+    convMVar?,
+    calcPrevRhs?,
+  }
+  return (goalState, region)
+
 end Pantograph

README.md

@@ -64,62 +64,7 @@ stat
 ```
 where the application of `assumption` should lead to a failure.
 
-### Commands
+For a list of commands, see [REPL Documentation](doc/repl.md).
-
-See `Pantograph/Protocol.lean` for a description of the parameters and return values in JSON.
-* `reset`: Delete all cached expressions and proof trees
-* `stat`: Display resource usage
-* `expr.echo {"expr": <expr>, "type": <optional expected type>, ["levels": [<levels>]]}`: Determine the
-  type of an expression and format it.
-* `env.catalog`: Display a list of all safe Lean symbols in the current environment
-* `env.inspect {"name": <name>, "value": <bool>}`: Show the type and package of a
-  given symbol; If value flag is set, the value is printed or hidden. By default
-  only the values of definitions are printed.
-* `options.set { key: value, ... }`: Set one or more options (not Lean options; those
-  have to be set via command line arguments.), for options, see `Pantograph/Protocol.lean`
-
-  One particular option for interest for machine learning researchers is the
-  automatic mode (flag: `"automaticMode"`).  By default it is turned on, with
-  all goals automatically resuming. This makes Pantograph act like a gym,
-  with no resumption necessary to manage your goals.
-* `options.print`: Display the current set of options
-* `goal.start {["name": <name>], ["expr": <expr>], ["levels": [<levels>]], ["copyFrom": <symbol>]}`:
-  Start a new proof from a given expression or symbol
-* `goal.tactic {"stateId": <id>, "goalId": <id>, ...}`: Execute a tactic string on a
-  given goal. The tactic is supplied as additional key-value pairs in one of the following formats:
-  - `{ "tactic": <tactic> }`: Execute an ordinary tactic
-  - `{ "expr": <expr> }`: Assign the given proof term to the current goal
-  - `{ "have": <expr>, "binderName": <name> }`: Execute `have` and creates a branch goal
-  - `{ "calc": <expr> }`: Execute one step of a `calc` tactic. Each step must
-    be of the form `lhs op rhs`. An `lhs` of `_` indicates that it should be set
-    to the previous `rhs`.
-  - `{ "conv": <bool> }`: Enter or exit conversion tactic mode. In the case of
-    exit, the goal id is ignored.
-* `goal.continue {"stateId": <id>, ["branch": <id>], ["goals": <names>]}`:
-  Execute continuation/resumption
-  - `{ "branch": <id> }`: Continue on branch state. The current state must have no goals.
-  - `{ "goals": <names> }`: Resume the given goals
-* `goal.remove {"stateIds": [<id>]}"`: Drop the goal states specified in the list
-* `goal.print {"stateId": <id>}"`: Print a goal state
-* `frontend.process { ["fileName": <fileName>",] ["file": <str>], invocations:
-  <bool>, sorrys: <bool> }`: Executes the Lean frontend on a file, collecting
-  either the tactic invocations (`"invocations": true`) or the sorrys into goal
-  states (`"sorrys": true`)
-
-### Errors
-
-When an error pertaining to the execution of a command happens, the returning JSON structure is
-
-``` json
-{ "error": "type", "desc": "description" }
-```
-Common error forms:
-* `command`: Indicates malformed command structure which results from either
-  invalid command or a malformed JSON structure that cannot be fed to an
-  individual command.
-* `index`: Indicates an invariant maintained by the output of one command and
-  input of another is broken. For example, attempting to query a symbol not
-  existing in the library or indexing into a non-existent proof state.
 
 ### Project Environment
 

Repl.lean

@@ -15,6 +15,16 @@ structure State where
 /-- Main state monad for executing commands -/
 abbrev MainM := ReaderT Context (StateT State Lean.CoreM)
 
+def newGoalState (goalState: GoalState) : MainM Nat := do
+  let state ← get
+  let stateId := state.nextId
+  set { state with
+    goalStates := state.goalStates.insert stateId goalState,
+    nextId := state.nextId + 1
+  }
+  return stateId
+
+
 -- HACK: For some reason writing `CommandM α := MainM (Except ... α)` disables
 -- certain monadic features in `MainM`
 abbrev CR α := Except Protocol.InteractionError α
@@ -50,6 +60,8 @@ def execute (command: Protocol.Command): MainM Lean.Json := do
     | "goal.continue" => run goal_continue
     | "goal.delete"   => run goal_delete
     | "goal.print"    => run goal_print
+    | "goal.save"     => run goal_save
+    | "goal.load"     => run goal_load
     | "frontend.process"  => run frontend_process
     | cmd =>
       let error: Protocol.InteractionError :=
@@ -62,14 +74,6 @@ def execute (command: Protocol.Command): MainM Lean.Json := do
   errorCommand := errorI "command"
   errorIndex := errorI "index"
   errorIO := errorI "io"
-  newGoalState (goalState: GoalState) : MainM Nat := do
-    let state ← get
-    let stateId := state.nextId
-    set { state with
-      goalStates := state.goalStates.insert stateId goalState,
-      nextId := state.nextId + 1
-    }
-    return stateId
   -- Command Functions
   reset (_: Protocol.Reset): MainM (CR Protocol.StatResult) := do
     let state ← get
@@ -90,10 +94,10 @@ def execute (command: Protocol.Command): MainM Lean.Json := do
     Environment.addDecl args
   env_save (args: Protocol.EnvSaveLoad): MainM (CR Protocol.EnvSaveLoadResult) := do
     let env ← Lean.MonadEnv.getEnv
-    env_pickle env args.path
+    environmentPickle env args.path
     return .ok {}
   env_load (args: Protocol.EnvSaveLoad): MainM (CR Protocol.EnvSaveLoadResult) := do
-    let (env, _) ← env_unpickle args.path
+    let (env, _) ← environmentUnpickle args.path
     Lean.setEnv env
     return .ok {}
   expr_echo (args: Protocol.ExprEcho): MainM (CR Protocol.ExprEchoResult) := do
@@ -203,11 +207,7 @@ def execute (command: Protocol.Command): MainM Lean.Json := do
     match nextState? with
     | .error error => return .error <| errorI "structure" error
     | .ok nextGoalState =>
-      let nextStateId := state.nextId
+      let nextStateId ← newGoalState nextGoalState
-      set { state with
-        goalStates := state.goalStates.insert nextStateId nextGoalState,
-        nextId := state.nextId + 1
-      }
       let goals ← goalSerialize nextGoalState (options := state.options)
       return .ok {
         nextStateId,
@@ -224,6 +224,16 @@ def execute (command: Protocol.Command): MainM Lean.Json := do
       return .error $ errorIndex s!"Invalid state index {args.stateId}"
     let result ← runMetaInMainM <| goalPrint goalState state.options
     return .ok result
+  goal_save (args: Protocol.GoalSave): MainM (CR Protocol.GoalSaveResult) := do
+    let state ← get
+    let .some goalState := state.goalStates[args.id]? |
+      return .error $ errorIndex s!"Invalid state index {args.id}"
+    goalStatePickle goalState args.path
+    return .ok {}
+  goal_load (args: Protocol.GoalLoad): MainM (CR Protocol.GoalLoadResult) := do
+    let (goalState, _) ← goalStateUnpickle args.path (← Lean.MonadEnv.getEnv)
+    let id ← newGoalState goalState
+    return .ok { id }
   frontend_process (args: Protocol.FrontendProcess): MainM (CR Protocol.FrontendProcessResult) := do
     let options := (← get).options
     try

Test/Common.lean

@@ -125,11 +125,20 @@ def mvarUserNameAndType (mvarId: MVarId): MetaM (Name × String) := do
 
 abbrev TestT := StateT LSpec.TestSeq
 
-def addTest [Monad m] (test: LSpec.TestSeq): TestT m Unit := do
+def addTest [Monad m] (test: LSpec.TestSeq) : TestT m Unit := do
   set $ (← get) ++ test
 
+def checkEq [Monad m] [DecidableEq α] (desc : String) (lhs rhs : α) : TestT m Unit := do
+  addTest $ LSpec.check desc (lhs == rhs)
+def checkTrue [Monad m] (desc : String) (flag : Bool) : TestT m Unit := do
+  addTest $ LSpec.check desc flag
+def fail [Monad m] (desc : String) : TestT m Unit := do
+  addTest $ LSpec.check desc false
+
 def runTest [Monad m] (t: TestT m Unit): m LSpec.TestSeq :=
   Prod.snd <$> t.run LSpec.TestSeq.done
+def runTestWithResult { α } [Monad m] (t: TestT m α): m (α × LSpec.TestSeq) :=
+  t.run LSpec.TestSeq.done
 
 def runTestTermElabM (env: Environment) (t: TestT Elab.TermElabM Unit):
   IO LSpec.TestSeq :=

Test/Main.lean

@@ -1,11 +1,12 @@
 import LSpec
+import Test.Delate
 import Test.Environment
 import Test.Frontend
 import Test.Integration
 import Test.Library
 import Test.Metavar
 import Test.Proofs
-import Test.Delate
+import Test.Serial
 import Test.Tactic
 
 -- Test running infrastructure
@@ -51,6 +52,7 @@ def main (args: List String) := do
     ("Metavar", Metavar.suite env_default),
     ("Proofs", Proofs.suite env_default),
     ("Delate", Delate.suite env_default),
+    ("Serial", Serial.suite env_default),
     ("Tactic/Congruence", Tactic.Congruence.suite env_default),
     ("Tactic/Motivated Apply", Tactic.MotivatedApply.suite env_default),
     ("Tactic/No Confuse", Tactic.NoConfuse.suite env_default),

Test/Serial.lean

@@ -0,0 +1,109 @@
+import LSpec
+import Test.Common
+import Lean
+import Pantograph.Library
+
+open Lean
+
+namespace Pantograph.Test.Serial
+
+def tempPath : IO System.FilePath := do
+  Prod.snd <$> IO.FS.createTempFile
+
+structure MultiState where
+  coreContext : Core.Context
+  env: Environment
+
+abbrev TestM := TestT $ StateRefT MultiState $ IO
+
+instance : MonadEnv TestM where
+  getEnv      := return (← getThe MultiState).env
+  modifyEnv f := do modifyThe MultiState fun s => { s with env := f s.env }
+
+def runCoreM { α } (state : Core.State) (testCoreM : TestT CoreM α) : TestM (α × Core.State) := do
+  let multiState ← getThe MultiState
+  let coreM := runTestWithResult testCoreM
+  match ← (coreM.run multiState.coreContext state).toBaseIO with
+  | .error e => do
+    throw $ .userError $ ← e.toMessageData.toString
+  | .ok ((a, tests), state') => do
+    set $ (← getThe LSpec.TestSeq) ++ tests
+    return (a, state')
+
+def test_environment_pickling : TestM Unit := do
+  let coreSrc : Core.State := { env := ← getEnv }
+  let coreDst : Core.State := { env := ← getEnv }
+
+  let name := `mystery
+  let envPicklePath ← tempPath
+  let ((), _) ← runCoreM coreSrc do
+    let type: Expr := .forallE `p (.sort 0) (.forallE `h (.bvar 0) (.bvar 1) .default) .default
+    let value: Expr := .lam `p (.sort 0) (.lam `h (.bvar 0) (.bvar 0) .default) .default
+    let c := Lean.Declaration.defnDecl <| Lean.mkDefinitionValEx
+      (name := name)
+      (levelParams := [])
+      (type := type)
+      (value := value)
+      (hints := Lean.mkReducibilityHintsRegularEx 1)
+      (safety := Lean.DefinitionSafety.safe)
+      (all := [])
+    let env' ← match (← getEnv).addDecl (← getOptions) c with
+      | .error e => do
+        let error ← (e.toMessageData (← getOptions)).toString
+        throwError error
+      | .ok env' => pure env'
+    environmentPickle env' envPicklePath
+
+  let _ ← runCoreM coreDst do
+    let (env', _) ← environmentUnpickle envPicklePath
+    checkTrue s!"Has symbol {name}" (env'.find? name).isSome
+    let anotherName := `mystery2
+    checkTrue s!"Doesn't have symbol {anotherName}" (env'.find? anotherName).isNone
+
+  IO.FS.removeFile envPicklePath
+
+def test_goal_state_pickling_simple : TestM Unit := do
+  let coreSrc : Core.State := { env := ← getEnv }
+  let coreDst : Core.State := { env := ← getEnv }
+  let statePath ← tempPath
+
+  let type: Expr := .forallE `p (.sort 0) (.forallE `h (.bvar 0) (.bvar 1) .default) .default
+  let stateGenerate : MetaM GoalState := runTermElabMInMeta do
+    GoalState.create type
+
+  let ((), _) ← runCoreM coreSrc do
+    let state ← stateGenerate.run'
+    goalStatePickle state statePath
+
+  let ((), _) ← runCoreM coreDst do
+    let (goalState, _) ← goalStateUnpickle statePath (← getEnv)
+    let metaM : MetaM (List Expr) := do
+      goalState.goals.mapM λ goal => goalState.withContext goal goal.getType
+    let types ← metaM.run'
+    checkTrue "Goals" $ types[0]!.equal type
+
+  IO.FS.removeFile statePath
+
+structure Test where
+  name : String
+  routine: TestM Unit
+
+protected def Test.run (test: Test) (env: Lean.Environment) : IO LSpec.TestSeq := do
+  -- Create the state
+  let state : MultiState := {
+    coreContext := ← createCoreContext #[],
+    env,
+  }
+  match ← ((runTest $ test.routine).run' state).toBaseIO with
+  | .ok e => return e
+  | .error e =>
+    return LSpec.check s!"Emitted exception: {e.toString}" (e.toString == "")
+
+def suite (env : Lean.Environment): List (String × IO LSpec.TestSeq) :=
+  let tests: List Test := [
+    { name := "environment_pickling", routine := test_environment_pickling, },
+    { name := "goal_state_pickling_simple", routine := test_goal_state_pickling_simple, },
+  ]
+  tests.map (fun test => (test.name, test.run env))
+
+end Pantograph.Test.Serial

doc/repl.md

@@ -0,0 +1,63 @@
+# REPL
+
+## Commands
+
+See `Pantograph/Protocol.lean` for a description of the parameters and return values in JSON.
+* `reset`: Delete all cached expressions and proof trees
+* `stat`: Display resource usage
+* `expr.echo {"expr": <expr>, "type": <optional expected type>, ["levels": [<levels>]]}`: Determine the
+  type of an expression and format it.
+* `env.catalog`: Display a list of all safe Lean symbols in the current environment
+* `env.inspect {"name": <name>, "value": <bool>}`: Show the type and package of a
+  given symbol; If value flag is set, the value is printed or hidden. By default
+  only the values of definitions are printed.
+* `env.save { path }`, `env.load { path }`: Save/Load the current environment
+  to/from a file
+* `options.set { key: value, ... }`: Set one or more options (not Lean options; those
+  have to be set via command line arguments.), for options, see `Pantograph/Protocol.lean`
+
+  One particular option for interest for machine learning researchers is the
+  automatic mode (flag: `"automaticMode"`).  By default it is turned on, with
+  all goals automatically resuming. This makes Pantograph act like a gym,
+  with no resumption necessary to manage your goals.
+* `options.print`: Display the current set of options
+* `goal.start {["name": <name>], ["expr": <expr>], ["levels": [<levels>]], ["copyFrom": <symbol>]}`:
+  Start a new proof from a given expression or symbol
+* `goal.tactic {"stateId": <id>, "goalId": <id>, ...}`: Execute a tactic string on a
+  given goal. The tactic is supplied as additional key-value pairs in one of the following formats:
+  - `{ "tactic": <tactic> }`: Execute an ordinary tactic
+  - `{ "expr": <expr> }`: Assign the given proof term to the current goal
+  - `{ "have": <expr>, "binderName": <name> }`: Execute `have` and creates a branch goal
+  - `{ "calc": <expr> }`: Execute one step of a `calc` tactic. Each step must
+    be of the form `lhs op rhs`. An `lhs` of `_` indicates that it should be set
+    to the previous `rhs`.
+  - `{ "conv": <bool> }`: Enter or exit conversion tactic mode. In the case of
+    exit, the goal id is ignored.
+* `goal.continue {"stateId": <id>, ["branch": <id>], ["goals": <names>]}`:
+  Execute continuation/resumption
+  - `{ "branch": <id> }`: Continue on branch state. The current state must have no goals.
+  - `{ "goals": <names> }`: Resume the given goals
+* `goal.remove {"stateIds": [<id>]}"`: Drop the goal states specified in the list
+* `goal.print {"stateId": <id>}"`: Print a goal state
+* `goal.save{ id, path }`, `goal.load { path }`: Save/Load a goal state to/from a
+  file. The environment is not carried with the state. The user is responsible
+  to ensure the sender/receiver instances share the same environment.
+* `frontend.process { ["fileName": <fileName>",] ["file": <str>], invocations:
+  <bool>, sorrys: <bool> }`: Executes the Lean frontend on a file, collecting
+  either the tactic invocations (`"invocations": true`) or the sorrys into goal
+  states (`"sorrys": true`)
+
+## Errors
+
+When an error pertaining to the execution of a command happens, the returning JSON structure is
+
+``` json
+{ "error": "type", "desc": "description" }
+```
+Common error forms:
+* `command`: Indicates malformed command structure which results from either
+  invalid command or a malformed JSON structure that cannot be fed to an
+  individual command.
+* `index`: Indicates an invariant maintained by the output of one command and
+  input of another is broken. For example, attempting to query a symbol not
+  existing in the library or indexing into a non-existent proof state.