Hello, reCaptcha, and the rest of the world

Hello, reCaptcha, and the rest of the world

In this chapter, we will see how to plug an external API to the Opa platform -- here, Google reCaptcha, an API used to protect forms against spammers by attempting to determine whether the person filling the form is actually a human being. Along the way, we will introduce the Opa Binding System Library (or BSL) and some of the mechanisms provided by Opa to permit modular, safe programming.

As we will be interacting with JavaScript, some notions of the JavaScript syntax are useful to understand this chapter.

Overview

As in previous chapters, let us start with a picture of the application we will develop in this chapter:

This simple application prompts users to recognize words that are considered too difficult to read by a computer in the current state of the art of character recognition -- typically because Google's own resources have failed to make sense of these words -- and determines that a user is indeed a human being if the answer corresponds to that of sufficient other users through the world. The feature is provided by Google, as an API called reCaptcha. This API involves some code that must be executed on the client (to display the user interface and offer some interactivity) and some code that must be executed on the server (to contact Google's servers and check the validity of the answer).

If you are curious, this is the full source code of our application.

Run | Fork

Of course, since the features are provided by Google reCaptcha, the most interesting aspects of the code are not to be found in the source of the application itself, but in how we define the binding of reCaptcha for Opa.

This is done in two parts. At high-level, we find the Opa package.

At low level, we find the JavaScript binding.

In the rest of the chapter, we will walk you through all the concepts and constructions introduced in these listings.

Populating the BSL

The documentation of the reCaptcha API details five methods of a JavaScript object called Recaptcha, that should be called at distinct stages of the use of the API. Our first step will therefore be to bind each of these methods to Opa, through the mechanism of the Binding System Library.

TIP: About the BSL

The Binding System Library was developed to allow binding external features to Opa. The mechanism is used at low-level in Opa itself, to bind Opa to browser features, but also to server features.

At the time of this writing, the BSL provides native bindings with JavaScript and OCaml, and can be used to bind to most languages, including C.

This chapter shows how to plug Opa with some JavaScript client code.

For this purpose, we will create a file called recaptcha.js, and that we will populate with Opa-to-JavaScript bindings.

Let us start with initialization. The documentation states that any use of the JavaScript API must initialize the Recaptcha object as follows:

Recaptcha.create(pubkey,
   id,
   {
      theme: theme,
      callback: Recaptcha.focus_response_field
   }
);

where pubkey is a public key obtained from the reCaptcha admin site, id is the identifier of the UI component that will host the reCaptcha and theme is the name of the visual theme.

For our purpose, as we are binding the API, these three values should be function arguments, for a function accepting three character strings and returning nothing meaningful.

We specify this as follows:

Binding initialization (extract of recaptcha.js)

/**
 * @register {string, string, string -> void}
 */
function init(id, pubkey, theme)
{
  Recaptcha.create(pubkey,
      id,
      {
        theme: theme,
        callback: Recaptcha.focus_response_field
      }
  );
}

This JavaScript code registers a function called init. This function is implemented in JavaScript and its type is specified as string, string, string -> void.

The rest of the extract is regular JavaScript, copied directly from the documentation of reCaptcha.

CAUTION: About void

The only surprise may be that there is no final return. Indeed, in regular JavaScript, functions with no return value actually return undefined. By opposition, Opa is stricter and does not allow undefined values. Since our definition states that init always returns a void, the BSL will apply an automatic transformation of the Javascrit code so that once bined in Opa, the function returns void.

In Opa, functions always return a value, even if this value is void. This is true even of functions implemented in JavaScript.

Therefore, if a BSL JavaScript function has type ... -> void, once in Opa, it returns void.

The rest of the API is quite similar. Functions reload and destroy, which serve respectively to display a new challenge or to clean the screen once the reCaptcha has become useless, are bound as follows:

Binding reload, destroy (extract of recaptcha.js)

/**
 * @register { -> void}
 */
function reload()
{
    Recaptcha.reload();
}

/**
 * @register {-> void}
 */
function destroy()
{
    Recaptcha.destroy();
}

Binding functions get_challenge and get_response is quite similar. The first of these functions returns an opaque string that can be used by the reCaptcha server to determine which image has been sent to the user. The second returns the text entered by the user.

Binding get_challenge and get_response (extract of recaptcha.js)

/**
 * @register{-> string}
 */
function get_challenge()
{
    return (Recaptcha.get_challenge()||"")
}

/**
 * @register {-> string}
 */
function get_response()
{
    return (Recaptcha.get_response()||"")
}

Note that we do not return simply Recaptcha.get_challenge() or Recaptcha.get_response(). Indeed, experience with the reCaptcha API shows that, in some (undocumented) cases, these functions return value null, which is definitely not a valid Opa string. For this purpose, we normalize the null value to the empty string "".

CAUTION: About null

In Opa, the JavaScript value null is meaningless. An Opa function implemented as JavaScript and which returns null (or an object in which some fields are null) is a programmer error.

We are now done with JavaScript.

Typing the API

The next step is to connect the BSL to the server component and wrap the features as a nice Opa API.

Looking at the documentation of reCaptcha, we may see that a reCaptcha accepts exactly three meaningful arguments:

• a private key, which we need to obtain manually from reCaptcha, and which should never be transmitted to the client, for security reasons;
• a public key, which we obtain along with the private key;
• an optional theme name.

We group these arguments as a record type, as follows:

The reCaptcha configuration

type Recaptcha.config =
{
   {
     string privkey
   }
   cfg_private,

   {
      string pubkey,
      option(string) theme
   }
   cfg_public
}

By convention, records which simply serve to group arguments under meaningful names are called configurations and their name ends with .config. Here, in order to avoid confusions, we have split this record in two subrecords, called respectively cfg_private, for information that we want to keep on the server, and cfg_public, for information that can leave the client without breaching security.

Also, looking at the documentation of the reCaptcha server-side API, we find out that communications with the reCaptcha server can yield the following results:

• a success;
• a failure, which may either mean that the user failed to identify the text, or that some other issue took place, including a communication error between Google servers.

Two additional error cases may appear:

• a communication error between your application and Google (typically, due to a network outage);
• a message returned by Google which does not match the documentation (unlikely but possible);
• an empty answer provided by the user, in which case communication should not even take place.

While these distinct results are represented as strings in the API, in Opa, we will prefer a sum type, which we define as follows:

reCaptcha results

type Recaptcha.success = {captcha_solved} /**The captcha is correct.*/
type Recaptcha.failure =
    { WebClient.failure captcha_not_reachable } /**Could not reach the distant server.*/
 or { string upstream }      /**Upstream failure. Could be a user error, but the code is not meant to be exploited.*/
 or { list(string) unknown } /**Server could be reached, but produced an error that doesn't match the specifications
                                 provided by Google. Possible cause: proxy problem.*/
 or { empty_answer }         /**Recaptcha guidelines mention that we should never send answers that are empty.*/

type Recaptcha.result = { Recaptcha.success success } or { Recaptcha.failure failure }

And finally, we define one last type, that of the reCaptcha implementation, as follows:

Type of the reCaptcha implementation (first version)

type Recaptcha.implementation = {
      /**Place a request to the reCaptcha server to verify that user entry is correct.
        @param challenge
        @param response
        @param callback*/
     (string, string, (Recaptcha.result -> void) -> void) validate,
      /**Reload the reCaptcha, displaying a different challenge*/
     (-> void) reload,
      /**Destroy the reCaptcha*/
     (-> void) destroy
}

The role of this type is to encapsulate the functions of the reCaptcha after construction.

This type offers three fields. The first two will map respectively to the reload method we have bound earlier and to the destroy method we have bound earlier. The third, validate, is a more powerful function whose role will be to send to the reCaptcha server the challenge, the response entered by the user, to wait for the server response and to trigger a function with the result.

TIP: Objects in Opa

Opa is not an Object-Oriented Programming Language in the traditional sense of the term. However, it is a Higher-Order Programming Language, which means among other things that all the major features of Object-Oriented Programming can be found in Opa, and more.

In our listing, values of type Recaptcha.implementation contain fields, some of which are functions -- not unlike Objects in OO languages may contain fields and methods.

Although this is a slight misues of the term, we often call such records, that is records containing fields, some of which are functions, Objects.

Now that the types are ready, we can write the functions that manipulate them.

Implementing the API

The documentation of reCaptcha mentions a JavaScript library, provided by reCaptcha, which needs to be loaded prior to initializing the reCaptcha. We handle this in a function onready which we will use shortly:

Function onready (first version)

function void onready(string id, string pubkey, string theme)
{
   Client.Script.load_uri_then(path_js_uri,
     function()
     {
       (%% Recaptcha.init %%)(id, pubkey, theme)
     }
   )
}

This function makes use of Client.Script.load_uri_then, a function provided by the library to load a distant script and, once loading is complete, to invoke a second function. First argument, path_js_uri, is a constant representing the URI at which the JS is available. We will define it a bit later. The second argument is a function that makes use of a construction you have never seen:

(%% Recaptcha.init %%)

This is simply function init, as we defined it a few minutes ago in JavaScript.

TIP: Calling the BSL

To use an Opa value defined in the BSL (that is, in JavaScript, C, OCaml, or any other language), the syntax is

(%% NameOfThePlugin.name_of_the_value %%)

Replace NameOfThePlugin by the name of the file you have passed to opa-plugin-builder and name_of_the_value by the name you have registered with ##register. Capitalization of plug-in names is ignored.

The contains between the two %% is called the key of the external primitive.

In other words, this function onready loads the JavaScript provided by Google, then initializes the reCaptcha. To call this function, we will make use of a Recaptcha.config and a ready event, to ensure that it is called only on a browser that actually makes use of the reCaptcha. We will need this initialization in our constructor for Recaptcha.implementation.

TIP: Multiple loads

Module Client.Script provides several functions for loading JavaScript. These functions ensure that a JavaScript URI will only be loaded once. In other words, you do not have to worry about the same JavaScript being loaded and executed multiple times.

The second important function, which we will also need to build our Recaptcha.implementation, is the validation. We implement it as follows:

function validate(challenge, response, (Recaptcha.result -> void) callback)
{
  //By convention, do not even post a request if the data is empty
  if (String.is_empty(challenge) || String.is_empty(response))
  {
    callback({failure: {empty_answer}})
  }
  else
  {
    /**POST request, formatted as per API specifications*/
    data = [ ("privatekey", privkey)
           , ("remoteip",   "{HttpRequest.get_ip()?(127.0.0.1)}")
           , ("challenge",  challenge)
           , ("response",   response)
           ]
    /**Handle POST failures, decode reCaptcha responses, convert this to [reCaptcha.result].*/
    function with_result(res)
    {
      match (res)
      {
        case ~{failure}:
          callback({failure: {captcha_not_reachable: failure}})
        case ~{success}:
          details = String.explode("\n", success.content)
          match (details)
          {
            case ["true" | _]: callback({success: {captcha_solved}})
            case ["false", code | _]: callback({failure: {upstream: code}})
            default: callback({failure: {unknown: details}})
          }
      }
    }
    /**Encode arguments, POST them*/
    WebClient.Post.try_post_with_options_async(path_validate_uri,
          WebClient.Post.of_form({WebClient.Post.default_options with content: {some: data}}),
          with_result)
  }
}

Although this listing uses several functions that you have not seen yet, it should not surprise you. The first part responds immediately if the challenge or the response is negative. This complies with the specification of reCaptcha, in addition to saving precious resources on your server. We then build data, a list of association of the arguments expected by the reCaptcha API, and with_result, a function that handles the return of our {post} request by analyzing the resulting string. Note the use of function WebClient.Post.of_form, which converts a request using a list of associations, as are built by HTML forms, into a raw, string-based request. Also note function String.explode, which splits a string in a list of substrings.

From validate, as well as our JavaScript implementations of reload and destroy, we may now construct our Recaptcha.implementation, as follows:

function Recaptcha.implementation make_implementation(string privkey)
{
   function validate(challenge, response, (Recaptcha.result -> void) callback)
   {
     //By convention, do not even post a request if the data is empty
     if (String.is_empty(challenge) || String.is_empty(response))
     {
       callback({failure: {empty_answer}})
     }
     else
     {
       /**POST request, formatted as per API specifications*/
       data = [ ("privatekey", privkey)
              , ("remoteip",   "{HttpRequest.get_ip()?(127.0.0.1)}")
              , ("challenge",  challenge)
              , ("response",   response)
              ]
       /**Handle POST failures, decode reCaptcha responses, convert this to [reCaptcha.result].*/
       function with_result(res)
       {
         match (res)
         {
           case ~{failure}:
             callback({failure: {captcha_not_reachable: failure}})
           case ~{success}:
             details = String.explode("\n", success.content)
             match (details)
             {
               case ["true" | _]: callback({success: {captcha_solved}})
               case ["false", code | _]: callback({failure: {upstream: code}})
               default: callback({failure: {unknown: details}})
             }
         }
       }
       /**Encode arguments, POST them*/
       WebClient.Post.try_post_with_options_async(path_validate_uri,
             WebClient.Post.of_form({WebClient.Post.default_options with content: {some: data}}),
             with_result)
     }
   }
   {~validate, reload:cl_reload, destroy:cl_destroy}
}

With this function, we implement a new function make, to construct both the Recaptcha.implementation and the user interface xhtml component that connects to this implementation:

function (Recaptcha.implementation, xhtml) make(Recaptcha.config config) {
    id = Dom.fresh_id();
    xhtml = <div id={id} onready={function(_) { onready(id, config.cfg_public.pubkey, config.cfg_public.theme?"red") }}/>;
    (make_implementation(config.cfg_private.privkey), xhtml);
}

One more utility function will be useful, to read the user interface and clean it up immediately:

function {string challenge, string response} get_token() {
    result = ({ challenge: (%%Recaptcha.get_challenge%%)()
                , response: (%%Recaptcha.get_response%%)()
              });
    (%%Recaptcha.destroy%%)();
    result;
}

With this function, we now have exposed all the features we need to use a reCaptcha. However, at this stage, we are exposing a great deal of the implementation. Consequently, our next step will be to make the implementation abstract and to encapsulate the features in a module.

Modularizing the features

It is generally a good idea to split large (or even small) projects into packages, as follows:

Wrapping our code as a package

package tutorial.recaptcha

TIP: About packages

In Opa, a package is a unit of compilation and abstraction. Packages can be distributed separately as compiled code, packages can hold private values, abstract types, etc.

The following declaration states that the current file is part of package package_name:

package package_name

Conversely, the following declaration states that the current file makes use of package package_name:

import package_name

Generally, in Opa, packages are assembled into package hierarchies, using reverse domain notation.

Now that we have placed our code in a package, we may decide that some of our types are abstract, by adding an @abstract directive as follows:

abstract type Recaptcha.implementation = { ... }

TIP: About abstract types

An abstract type is a type whose definition can only be used in the package in which it was defined, although its name might be used outside of the package.

This feature provides a very powerful mechanism for avoiding errors and ensuring that data invariants remain unbroken, but also to ensure that third-party developers do not base their work on characteristics that may change at a later stage.

Consider the following example:

 package example

 abstract type cost = float

In this example, type cost is defined as a synonym of float. In package example, any float can be used as a cost and reciprocally. However, outside of package example, cost and float are two distinct types. In particular, we have ensured that only the code of package example can create new values of type cost. If our specifications expect that a cost is always strictly positive, we have succesfully restriced the perimeter of the code we need to check to ensure that this invariant remains unbroken: only package example needs to be verified.

With this change, methods validate, destroy and reload can now be called only from our package. As these methods offer important features, we certainly wish to provide some form of access to the methods, as follows:

Exporting features (first version)

function void validate(Recaptcha.implementation implementation, (Recaptcha.result -> void) callback) {
    t = get_token();
    implementation.validate(t.challenge, t.response, callback);
}

function void reload(Recaptcha.implementation implementation) {
    implementation.reload();
}

function void destroy(Recaptcha.implementation implementation) {
    implementation.destroy();
}

For further modularization, we will group all our functions as a module, and take the opportunity to hide the function and constants that should not be called from the outside of the module:

module Recaptcha
{
   function (Recaptcha.implementation, xhtml) make(Recaptcha.config config) {
       id = Dom.fresh_id();
       xhtml = <div id={id} onready={function(_) { onready(id, config.cfg_public.pubkey, config.cfg_public.theme?"red") }}/>;
       (make_implementation(config.cfg_private.privkey), xhtml);
   }

   function void validate(Recaptcha.implementation implementation, (Recaptcha.result -> void) callback) {
       t = get_token();
       implementation.validate(t.challenge, t.response, callback);
   }

   function void reload(Recaptcha.implementation implementation) {
       implementation.reload();
   }

   function void destroy(Recaptcha.implementation implementation) {
       implementation.destroy();
   }

   private path_validate_uri =
        Option.get(Parser.try_parse(Uri.uri_parser, "http://www.google.com/recaptcha/api/verify"));

   private path_js_uri =
        Option.get(Parser.try_parse(Uri.uri_parser, "http://www.google.com/recaptcha/api/js/recaptcha_ajax.js"));

   private function void onready(string id, string pubkey, string theme) {
      Client.Script.load_uri_then(path_js_uri,
        function()
        {
          (%% Recaptcha.init %%)(id, pubkey, theme)
        }
       );
   }

   private cl_reload =
     %%Recaptcha.reload%% /**Implementation of [reload]*/

   private cl_destroy =
     %%Recaptcha.destroy%% /**Implementation of [destroy]*/

   private function Recaptcha.implementation make_implementation(string privkey) {
      function validate(challenge, response, (Recaptcha.result -> void) callback) {
        //By convention, do not even post a request if the data is empty
        if (String.is_empty(challenge) || String.is_empty(response)) {
          callback({failure: {empty_answer}})
        } else {
          /**POST request, formatted as per API specifications*/
          data = [ ("privatekey", privkey)
                 , ("remoteip",   "{HttpRequest.get_ip()?(127.0.0.1)}")
                 , ("challenge",  challenge)
                 , ("response",   response)
                 ]
          /**Handle POST failures, decode reCaptcha responses, convert this to [reCaptcha.result].*/
          function with_result(res) {
              match (res) {
              case ~{failure}:
                  callback({failure: {captcha_not_reachable: failure}})
              case ~{success}:
                  details = String.explode("\n", success.content)
                  match (details) {
                  case ["true" | _]: callback({success: {captcha_solved}})
                  case ["false", code | _]: callback({failure: {upstream: code}})
                  default: callback({failure: {unknown: details}})
                  }
              }
          }
          /**Encode arguments, POST them*/
          WebClient.Post.try_post_with_options_async(path_validate_uri,
                WebClient.Post.of_form({WebClient.Post.default_options with content: {some: data}}),
                with_result)
        }
      }
      {~validate, reload:cl_reload, destroy:cl_destroy}
   }

   private function {string challenge, string response} get_token() {
     result = ({ challenge: (%%Recaptcha.get_challenge%%)()
               , response: (%%Recaptcha.get_response%%)()
               });
     (%%Recaptcha.destroy%%)();
     result;
   }

}

TIP: About modules

A module is an extended form of record introduced with module ModuleName { ... } instead of { ... }.

Modules offer a few syntactic enrichments:

• any field may be declared as private, which forbids from using it outside of the module;
• fields can be declared in any order, even if they have dependencies (including circular dependencies).

In addition, modules are typed slightly differently from regular records. We will detail this in another chapter.

And with this, our binding is complete.

We may compile our package with

opa recaptcha.js hello_recaptcha.opa

Let us recapitulate the Opa source code: Source

Run | Fork

Questions

Why an object?

As mentioned, we have defined Recaptcha.implementation as an object. This is a good reflex when extending the Opa platform through additional BSL bindings that use data structures can be implemented only on one side.

In Opa, data can be transmitted transparently between the client and the server. This is impossible for data that is meaningful only on the client. This is the case here, as JavaScript object Recaptcha, by definition, exists only on the client. However, wrapping the JavaScript data structure and the functions that manipulate it as an object ensures that the user only ever needs to access methods -- and such methods can always be transmitted from the client to the server, making the object data structure side-independent.

TIP: Making objects

When a BSL extension to the Opa platform introduces a data structure implemented only on one side, the user must never manipulate this data structure directly. Always hide this data structure behind an object, whose only fields are functions.