package coq-serapi

  1. Overview
  2. Docs
Legend:
Library
Module
Module type
Parameter
Class
Class type

The SerAPI Protocol

SerAPI is a set of utilities designed to help users and tool creators to interact with Coq in a systematic way; in particular, SerAPI was designed to provide full serialization and de-serialization of key Coq structures, including user-level AST and kernel terms.

SerAPI also provides a reification of Coq's document building API, making it pretty easy to build and check systematically Coq documents.

As of today SerAPI does provide the following components:

  • serlib: A library providing serializers for core Coq structures; the main serialization formats are S-expressions and JSON. serlib is based on ppx_sexp_conv from Jane Street and `ppx_deriving_yojson`. serlib also provides for custom hash and equality functions for many Coq types.
  • sertop: A toplevel executable exposing a simple document building and querying protocol. This is the main component, we document it properly below.
  • sercomp: A simple compiler utility for .v files that can input and output Coq files in a variety of formats. See its manual for more help.
  • serload: TODO

History:

SerAPI was a JsCoq offspring project; JsCoq added experimental serialization of Coq terms, however we quickly realized that this facility would be helpful in the general setting; we also took advantage of the serialization facilities to specify the Coq building API as a DSL; the client for the tool was an experimental Emacs mode by Clément Pit-Claudel.

The next step was to provide reliable "round-trip" (de)serialization of full Coq documents; Karl Palmskog contributed the round trip testing infrastructure to make this happen.

Users:

SerAPI is a bit of a swiss army knife, in the sense that it is a general "talk to Coq" tool and can do many things; a good way to understand the tool is look at some of its users, see the list of them in the Project's README

Basic Overview of the Protocol:

SerAPI protocol can be divided in two main sets of operations: document creation and checking, and document querying.

Note that the protocol is fully specified as a DSL written in OCaml; thus, its canonical specification can be found below as documents to the OCaml code. In this section, we attempt a brief introduction, but the advanced user will without doubt want to look at the details just below.

Document creation and checking:

Before you can use SerAPI to extract any information about a Coq document, you must indeed have Coq parse and process the document. Coq's parsing process is quite complicated due to user-extensibility, but SerAPI tries to smooth the experience as much as possible.

A Coq document is basically a list of sentences which are uniquely identified by a Stateid.t object; for our purposes this identifier is an integer.

Note: In future versions, sentence id will be deprecated, and instead we will use Language Server Protocol-style locations inside the document to identify sentences.

Each sentence has a "parent", that is to say, a previous sentence; the initial sentence has as a parent sid = 1 (sid = sentence id).

Note that the parent is important for parsing as it may modify the parsing itself, for example it may be a Notation command.

Thus, to build or append to a Coq document, you should select a parent sentence and ask SerAPI to add some new ones. This is achieved with the (Add (opts) "text") command.

See below for a detailed overview of Add, but the basic idea is that Coq will parse and add to the current document as many sentences as you have sent to it. Unfortunately, sentence number for the newly added ones is not always predictable but there are workarounds for that.

If succesfull, Add will send back an Added message with the location and new sentence identifier. This is useful to let SerAPI do the splitting of sentences for you. A typical use thus is:

(Add () "Lemma addnC n : n + 0 = n. Proof. now induction n. Qed.")

This will return 4 answers.

Sentence Checking

Adding a set of sentences basically amounts to parsing, however in most cases Coq won't try to typecheck or run the tactics at hand. For that purpose you can use the (Exec sid) command. Taking a sentence id, Check will actually check sid and all the sentences sid depends upon.

Note that in some modes Coq can skip proofs here, so in order to get a fully-checked document you may have to issue Check for every sentence on it. Checking a sentence twice is usually a noop.

Modification of the Document

In order to modify a "live" document, SerAPI does provide a (Cancel sid) command. Cancel will take a sentence id and return the list of sentences that are not valid anymore.

Thus, you can edit a document by cancelling and re-adding sentences.

Caveats

Cancelling a non-executed part is poorly supported by the underlying Coq checking algorithm. In particular, Cancel will force execution up to the previous sentence; thus it is not possible to parse a list of sentences and then replace them without incurring in the cost of executing them. In particular, it could be even the case that after issuing Cancel sid, there is an error in the execution of an unrelated sentence. It should be possible to identify this sentence using the exception attributes. As of today, this remains a hard-limitation of the STM.

Querying documents:

For a particular point on the document, you can query Coq for information about it. Common query use cases are for example lists of tactics, AST, completion, etc... Querying is done using the (Query (opts) query) command. The full specification can be found below.

A particulary of Query is that the caller must set all the pertinent output options. For example, if the query should return for-humans data or machine-readable one.

Non-interactive use

In many cases, non-interactive use is very convenient; for that, we recommend you read the help of the `sercomp` compiler.

Protocol Specification

Basic Protocol Objects

SerAPI can return different kinds of objects as an answer to queries; object type is usually distinguished by a tag, for example (CoqString "foo") or (CoqConstr (App ...)

Serialization representation is derived from the OCaml representation automatically, except for a few custom datatypes (see below). Thus, the best is to use Merlin or some OCaml-browsing tool as to know the internal of each type; we provide a brief description of each object:

type coq_object =
  1. | CoqString of string
    (*

    A string

    *)
  2. | CoqSList of string list
    (*

    A list of strings

    *)
  3. | CoqPp of Pp.t
    (*

    A Coq "Pretty Printing" Document type, main type used by Coq to submit formatted output

    *)
  4. | CoqLoc of Loc.t
    (*

    A Coq Location object, used for positions inside the document.

    *)
  5. | CoqTok of Tok.t CAst.t list
    (*

    Coq Tokens, as produced by the lexer

    *)
  6. | CoqDP of Names.DirPath.t
    (*

    Coq "Logical" Paths, used for module and section names

    *)
  7. | CoqAst of Vernacexpr.vernac_control
    (*

    Coq Abstract Syntax trees for statements, as produced by the parser

    *)
  8. | CoqOption of Goptions.option_name * Goptions.option_state
    (*

    Coq Options, as in Set Resolution Depth

    *)
  9. | CoqConstr of Constr.constr
    (*

    Coq Kernel terms, this is the fundamental representation for terms of the Calculus of Inductive constructions

    *)
  10. | CoqEConstr of EConstr.t
    (*

    Coq Kernel terms, but maybe open

    *)
  11. | CoqExpr of Constrexpr.constr_expr
    (*

    Coq term ASTs, this is the user-level parsing tree of terms

    *)
  12. | CoqMInd of Names.MutInd.t * Declarations.mutual_inductive_body
    (*

    Coq kernel-level inductive; this is a low-level object that contains all the details of an inductive.

    *)
  13. | CoqEnv of Environ.env
    (*

    Coq kernel-level enviroments: they do provide the full information about what the kernel know, heavy.

    *)
  14. | CoqTactic of Names.KerName.t * Ltac_plugin.Tacenv.ltac_entry
    (*

    Representation of an Ltac tactic definition

    *)
  15. | CoqLtac of Ltac_plugin.Tacexpr.raw_tactic_expr
    (*

    AST of an LTAC tactic definition

    *)
  16. | CoqGenArg of Genarg.raw_generic_argument
    (*

    Coq Generic argument, can contain any type

    *)
  17. | CoqQualId of Libnames.qualid
    (*

    Qualified identifier

    *)
  18. | CoqGlobRef of Names.GlobRef.t
    (*

    "Global Reference", which is a type that can point to a module, a constant, a variable, a constructor...

    *)
  19. | CoqGlobRefExt of Globnames.extended_global_reference
    (*

    "Extended Global Reference", as they can contain syntactic definitions too

    *)
  20. | CoqImplicit of Impargs.implicits_list
    (*

    Implicit status for a constant

    *)
  21. | CoqProfData of Ltac_plugin.Profile_ltac.treenode
    (*

    Ltac Profiler data

    *)
  22. | CoqNotation of Constrexpr.notation
    (*

    Representation of a notation (usually a string)

    *)
  23. | CoqUnparsing of Ppextend.notation_printing_rules * Notation_gram.notation_grammar
    (*

    Rules for notation printing and some internals

    *)
  24. | CoqGoal of Constr.t Serapi_goals.reified_goal Serapi_goals.ser_goals
    (*

    Goals, with types and terms in Kernel-level representation

    *)
  25. | CoqExtGoal of Constrexpr.constr_expr Serapi_goals.reified_goal Serapi_goals.ser_goals
    (*

    Goals, with types and terms in user-level, AST representation

    *)
  26. | CoqProof of EConstr.constr list
    (*

    Proof object: really low-level and likely to be deprecated.

    *)
  27. | CoqAssumptions of Serapi_assumptions.t
    (*

    Structured representation of the assumptions of a constant.

    *)
  28. | CoqComments of ((int * int) * string) list list
    (*

    List of comments in a document, the list will have one element for each call to Add; note that with the current model, it is hard to do better, as a call to Add can map to several sentences so comments are really mapped to each of those.

    See https://github.com/coq/coq/issues/12413 for updates on improved support

    *)
  29. | CoqLibObjects of {
    1. library_segment : Summary.Interp.frozen Lib.library_segment;
    2. path_prefix : Nametab.object_prefix;
    }
    (*

    Meta-logical Objects in Coq's library / module system

    *)

There are some Coq types that cannot be seralizaled properly, in this case, the types can be "opaque", or we will perform some manual serialization, such for GADTs.

In the past generic arguments were such a case, but that has been fixed in SerAPI 0.17. Please open an issue or pull request if you find such a discrepancy as to document it here.

Printing Options

type print_format =
  1. | PpSer
    (*

    Output in serialized format usually sexp

    *)
  2. | PpStr
    (*

    Output a string with a human-friendly representation

    *)
  3. | PpTex
    (*

    Output a TeX expression

    *)
  4. | PpCoq
    (*

    Output a Coq Pp.t, representation-indepedent document

    *)

Query output format

type format_opt = {
  1. pp_format : print_format;
    (*

    Output format (default PpSer)

    *)
  2. pp_depth : int;
    (*

    Depth (default 0)

    *)
  3. pp_elide : string;
    (*

    Elipsis (default: "...")

    *)
  4. pp_margin : int;
    (*

    Margin (default: 72)

    *)
}

Printing options, not all options are relevant for all printing backends

type print_opt = {
  1. sid : Stateid.t;
    (*

    sid denotes the sentence id we are querying over, essential information as goals for example will vary.

    *)
  2. pp : format_opt;
    (*

    Printing format of the query, this can be used to select the type of the answer, as for example to show goals in human-form.

    *)
}
val gen_pp_obj : Environ.env -> Evd.evar_map -> coq_object -> Pp.t

Query Sub-Protocol

type query_pred =
  1. | Prefix of string
    (*

    Filter named objects based on the given prefix

    *)

Predicates on the queries. This is at the moment mostly a token functionality

type query_opt = {
  1. preds : query_pred list;
    (*

    List of predicates on queries, mostly a placeholder, will allow to add filtering conditions in the future

    *)
  2. limit : int option;
    (*

    Limit the number of results, should evolve into an API with resume functionality, maybe we adopt LSP conventions here

    *)
  3. sid : Stateid.t;
    (*

    sid denotes the sentence id we are querying over, essential information as goals for example will vary.

    *)
  4. pp : format_opt;
    (*

    Printing format of the query, this can be used to select the type of the answer, as for example to show goals in human-form.

    *)
  5. route : Feedback.route_id;
    (*

    Legacy/Deprecated STM query method

    *)
}

Query options, note the default values that help interactive use, however in mechanized use we do not recommend skipping any field

type query_cmd =
  1. | Option
    (*

    List of options Coq knows about

    *)
  2. | Search
    (*

    Query version of the Search command

    *)
  3. | Goals
    (*

    Current goals, in kernel form

    *)
  4. | EGoals
    (*

    Current goals, in AST form

    *)
  5. | Ast
    (*

    Ast for the current sentence

    *)
  6. | TypeOf of string
    (*

    Type of an expression (unimplemented?)

    *)
  7. | Names of string
    (*

    (Names prefix) will return the list of identifiers Coq knows that start with prefix

    *)
  8. | Tactics of string
    (*

    (Tactcis prefix) will return the list of tactics Coq knows that start with prefix

    *)
  9. | Locate of string
    (*

    Query version of the Locate commands

    *)
  10. | Implicits of string
    (*

    Return information of implicits for a given constant

    *)
  11. | Unparsing of string
    (*

    Return internal information for a given notation

    *)
  12. | Definition of string
    (*

    Return the definition for a given global

    *)
  13. | LogicalPath of string
    (*

    Returns Coq's "logical path" for a given file

    *)
  14. | PNotations
    (*

    Return a list of notations

    *)
  15. | ProfileData
    (*

    Return LTAC profile data, if any

    *)
  16. | Proof
    (*

    Return the proof object low-level

    *)
  17. | Vernac of string
    (*

    Execute an arbitrary Coq command in an isolated state.

    *)
  18. | Env
    (*

    Return the current enviroment

    *)
  19. | Assumptions of string
    (*

    Return the assumptions of a given global

    *)
  20. | Complete of string
    (*

    Naïve but efficient prefix-based completion of identifiers

    *)
  21. | Comments
    (*

    Get all comments of a document

    *)
  22. | Objects
    (*

    Get Coq meta-logical module objects

    *)

Query commands are mostly a tag and some arguments determining the result type.

Important Note that Query won't force execution of a particular state, thus for example if you do (Query ((sid 3)) Goals) and the sentence 3 wasn't evaluated, then the query will return zero answers.

We would ideally evolve towards a true query language, likley having query_cmd and coq_object be typed such that query : 'a query -> 'a coq_object.

module QueryUtil : sig ... end

Control Sub-Protocol

Adding a new sentence
type parse_entry =
  1. | Vernac
  2. | Constr
type parse_opt = {
  1. ontop : Stateid.t option;
  2. entry : parse_entry;
}

parse ontop of the given sentence with entry entry

type add_opts = {
  1. lim : int option;
    (*

    Parse lim sentences at most (None == no limit)

    *)
  2. ontop : Stateid.t option;
    (*

    parse ontop of the given sentence

    *)
  3. newtip : Stateid.t option;
    (*

    Make newtip the new sentence id, very useful to avoid synchronous operations

    *)
  4. verb : bool;
    (*

    verb internal Coq parameter, be verbose on parsing

    *)
}

Add will take a string and parse all the sentences on it, until an error of the end is found. Options for Add are:

Creating a new document

experimental

type newdoc_opts = {
  1. top_name : Coqargs.top;
    (*

    name of the top-level module of the new document

    *)
  2. ml_load_path : string list option;
    (*

    Initial ML loadpath

    *)
  3. vo_load_path : Loadpath.vo_path list option;
    (*

    Initial LoadPath for the document

    *)
  4. require_libs : Coqargs.require_injection list option;
    (*

    Libraries to load in the initial document state

    *)
}
type save_opts = {
  1. prefix_output_dir : string option;
    (*

    prefix a directory to the saved vo file.

    *)
  2. sid : Stateid.t;
    (*

    sid of the point to save the document

    *)
}

Save options, Coq must save a module `Foo` to a concrete module path determined by -R / -Q options , so we don't have a lot of choice here.

Top Level Protocol

The top level protocol is the main input command to SerAPI, we detail each of the commands below.

The main interaction loop is as: 1. submit tagged command (tag (Cmd args)) 2. receive tagged ack (Answer tag Ack) 3. receive tagged results, usually (Answer tag (ObjList ...) or (Answer tag (CoqExn ...) 4. receive tagged completion event (Answer tag Completed)

The Ack and Completed events are always produced, and provide a kind of "bracking" for command execution.

type cmd =
  1. | NewDoc of newdoc_opts
    (*

    Create a new document, experimental, only usable when --no_init was used.

    *)
  2. | SaveDoc of save_opts
    (*

    Save the .vo file corresponding to the current document, note that proofs must be closed etc... in order for this to succeed.

    *)
  3. | Add of add_opts * string
    (*

    Add a set of sentences to the current document

    *)
  4. | Cancel of Stateid.t list
    (*

    Remove a set of sentences from the current document

    *)
  5. | Exec of Stateid.t
    (*

    Execute a particular sentence

    *)
  6. | Query of query_opt * query_cmd
    (*

    Query a Coq document

    *)
  7. | Print of print_opt * coq_object
    (*

    Print some object

    *)
  8. | Parse of parse_opt * string
    (*

    Parse

    *)
  9. | Join
    (*

    Be sure that a document is consistent

    *)
  10. | Finish
    (*

    Internal

    *)
  11. | ReadFile of string
  12. | Tokenize of string
  13. | Noop
  14. | Help

Each top level command will produce an answers, see below for answer description.

exception NoSuchState of Stateid.t

raised when referring to a Stateid.t unknown to SerAPI

exception CannotSaveVo

raised when trying to save a module without a corresponding --topfile

module ExnInfo : sig ... end
type answer_kind =
  1. | Ack
    (*

    The command was received, Coq is processing it.

    *)
  2. | Completed
    (*

    The command was completed.

    *)
  3. | Added of Stateid.t * Loc.t * Stm.add_focus
    (*

    A sentence was added, with corresponding sentence id and location.

    *)
  4. | Canceled of Stateid.t list
    (*

    A set of sentences are not valid anymore.

    *)
  5. | ObjList of coq_object list
    (*

    Set of objects, usually the answer to a query

    *)
  6. | CoqExn of ExnInfo.t
    (*

    The command produced an error, optionally at a document location

    *)

State of the evaluator

module State : sig ... end

Entry points to the DSL evaluator

val exec_cmd : State.t -> cmd -> answer_kind list * State.t

exec_cmd cmd execute SerAPI command

type cmd_tag = string

Each command and answer are tagged by a user-provided identifier

type tagged_cmd = cmd_tag * cmd

We introduce our own feedback type to overcome some limitations of Coq's Feedback, for now we only modify the Message data

type feedback_content =
  1. | Processed
  2. | Incomplete
  3. | Complete
  4. | ProcessingIn of string
  5. | InProgress of int
  6. | WorkerStatus of string * string
  7. | AddedAxiom
  8. | FileDependency of string option * string
  9. | FileLoaded of string * string
  10. | Message of {
    1. level : Feedback.level;
    2. loc : Loc.t option;
    3. pp : Pp.t;
    4. str : string;
    }
type feedback = {
  1. doc_id : Feedback.doc_id;
  2. span_id : Stateid.t;
  3. route : Feedback.route_id;
  4. contents : feedback_content;
}
type answer =
  1. | Answer of cmd_tag * answer_kind
    (*

    The answer is comming from a user-issued command

    *)
  2. | Feedback of feedback
    (*

    Output produced by Coq (asynchronously)

    *)

General answers of the protocol can be responses to commands, or Coq messages

OCaml

Innovation. Community. Security.