Pantograph

A Machine-to-Machine interaction system for Lean 4.

Pantograph provides interfaces to execute proofs, construct expressions, and examine the symbol list of a Lean project for machine learning.

Installation

For Nix users, run

nix build .#{sharedLib,executable}

to build either the shared library or executable.

Install elan and lake, and run

lake build

This builds the executable in .lake/build/bin/pantograph-repl.

Executable Usage

pantograph MODULES|LEAN_OPTIONS

The REPL loop accepts commands as single-line JSON inputs and outputs either an Error: (indicating malformed command) or a JSON return value indicating the result of a command execution. The command can be passed in one of two formats

command { ... }
{ "cmd": command, "payload": ... }

The list of available commands can be found in Pantograph/Protocol.lean and below. An empty command aborts the REPL.

The pantograph executable must be run with a list of modules to import. It can also accept lean options of the form --key=value e.g. --pp.raw=true.

Example: (~5k symbols)

$ pantograph Init
env.catalog
env.inspect {"name": "Nat.le_add_left"}

Example with mathlib4 (~90k symbols, may stack overflow, see troubleshooting)

$ pantograph Mathlib.Analysis.Seminorm
env.catalog

Example proving a theorem: (alternatively use goal.start {"copyFrom": "Nat.add_comm"}) to prime the proof

$ pantograph Init
goal.start {"expr": "∀ (n m : Nat), n + m = m + n"}
goal.tactic {"stateId": 0, "goalId": 0, "tactic": "intro n m"}
goal.tactic {"stateId": 1, "goalId": 0, "tactic": "assumption"}
goal.delete {"stateIds": [0]}
stat {}
goal.tactic {"stateId": 1, "goalId": 0, "tactic": "rw [Nat.add_comm]"}
stat

where the application of assumption should lead to a failure.

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.

• 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. options.set { "automaticMode": true }. This makes Pantograph act like 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

{ "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

To use Pantograph in a project environment, setup the LEAN_PATH environment variable so it contains the library path of lean libraries. The libraries must be built in advance. For example, if mathlib4 is stored at ../lib/mathlib4, the environment might be setup like this:

LIB="../lib"
LIB_MATHLIB="$LIB/mathlib4/lake-packages"
export LEAN_PATH="$LIB/mathlib4/build/lib:$LIB_MATHLIB/aesop/build/lib:$LIB_MATHLIB/Qq/build/lib:$LIB_MATHLIB/std/build/lib"

LEAN_PATH=$LEAN_PATH build/bin/pantograph $@

The $LEAN_PATH executable of any project can be extracted by

lake env printenv LEAN_PATH

Troubleshooting

If lean encounters stack overflow problems when printing catalog, execute this before running lean:

ulimit -s unlimited

Library Usage

Pantograph/Library.lean exposes a series of interfaces which allow FFI call with Pantograph which mirrors the REPL commands above. It is recommended to call Pantograph via this FFI since it provides a tremendous speed up.

The executable can be used as-is, but linking against the shared library requires the presence of lean-all.

Inject any project path via the pantograph_init_search function.

Developing

A Lean development shell is provided in the Nix flake.

Testing

The tests are based on LSpec. To run tests, use either

nix flake check

or

lake test

You can run an individual test by specifying a prefix

lake test -- "Tactic/No Confuse"