ODK-X Survey Form Processing

The XLSXConverter converts the XLSX file defining an Survey form into a JSON representation of that form. This representation is then layered on top of the generic Survey JavaScript framework to produce the JavaScript code that is executed when filling out the form.

The primary building blocks of this generic Survey JavaScript framework are:

  • bootstrap - for prompt UI and behavior

  • Handlebars - for HTML content rendering

  • Backbone - for event handling within prompts

  • requirejs - for module dependencies and loading

  • jQuery - generic utility functions

  • underscore - generic utility functions

  • moment - date and time handling support

Some additional libraries are use for specific widgets and capabilities (for example, d3 for graphing, combodate for calendar widgets).

The Survey JavaScript framework then adds form navigation, data validation, data storage and data retrieval functions. Central to this framework is the calling context which provides a continuation abstraction for chaining and resuming processing during asynchronous interactions.

Survey Calling Contexts (ctxt)

The success and failure callbacks used within the odkData API are also used throughout the Survey JavaScript. These are so common, that they are passed into functions as a single "calling context" argument, generally named ctxt. Whereas many libraries have success and failure callbacks:

  • obj.action(successCallbackFn, failureCallbackFn);

the Survey JavaScript would just pass in the ctxt object:

  • obj.action(ctxt);

This ctxt object consists, at a minimum, of a success function and a failure function. The failure function generally takes one argument which is an object containing a message field that holds an error message. The success function may pass in an argument or not.

These calling contexts are created, tracked and managed by the controller class via:

  • window.controller.newContext( event ) – when needed during event processing

  • window.controller.newCallbackContext() – on callbacks from Java shim

  • window.controller.newStartContext() – special case

  • window.controller.newFatalContext() – special case

The ctxt object extends the baseContext defined within controller, which has:

    contextChain: [],
    append: function( method, detail ) {...},
    success: function() {...},
    failure: function(msg) {...},

A well-written success() or failure(msg) function will perform its actions then call the success or failure function of the parent instance from which it is extended. So you will often see code like this in Survey JavaScript:

var that = this;
this.render($.extend({}, ctxt, { success: function() {
    }, failure: function(msg) {
        ctxt.append("mymethod", "unable to render");
} });

Where postRender(ctxt) will be responsible for calling the success or failure methods of the ctxt object that was extended and passed into the render() method. The failure(msg) code, in contrast, just logs a message to the context log (via append(), discussed below), and calls the parent instance’s failure function.

By always calling the parent instance’s success or failure function, you can do interesting things, like implement mutexes (an advanced software construct) – because you are always assured that if you extend a ctxt, that one of your failure(msg) and success() functions will always be called.

The failure(msg) function takes an argument, which is an object that may contain an optional ‘message’ parameter, which could be a description of what the failure was. This is used during validation.

The use of the ctxt object enables you to store values within the ctxt, and ensure that these are available later in your code, or, via extending it, to change the success function so that it takes an argument, etc., as needed by your code (the database layer quite frequently needs to pass values into the ctxt success method).

The append() function on the context enables you to append a log record to the context. The baseContext’s success() and failure(msg) methods both cause the accumulated log messages to be written via the odkCommon.log(). On Chrome, the log message is suppressed. On Android, it is written to the /opendatakit/appName/output/logging directory and emitted in the system log if an error or warning.

The ‘seq:’ and ‘seqAtEnd:’ values emitted in these logs are useful for understanding what events are processed concurrently within the JavaScript. ‘seq’ is the sequence number of this context, and ‘seqAtEnd’ is the sequence number of the newest context in-process at the time this context completes.

Note that when interacting with other asynchronous frameworks, it is easy to convert from ctxt-based style to the success/failure function style:

fwk.action( function() { ctxt.success(); }, function() { ctxt.failure(); } );

Finally, these calling contexts are very similar to JavaScript promises. However, within the Survey JavaScript, the typical construction is to insert processing steps before taking the success or failure action of the incoming calling context. In contrast, with promises, the typical construction is to append processing steps upon completion of the promise.

In the rare cases when it is necessary to append actions after a calling context chain completes (like the Promise model), two APIs are provided:

  • ctxt.setChainedContext(aCtxt);

  • ctxt.setTerminalContext(aCtxt);

Chained contexts are executed in-order, depth-first, from first registered to last registered, after which all terminal contexts are executed in the order in which they were collected from within all of the executed chained contexts. In practice, the Survey JavaScript framework only makes use of terminal contexts, and those usages only register a single terminal context.

Survey JavaScript Modules

All user forms processed within Survey load the same HTML file. Form-specific content and behaviors are specified via the window.location.hash portion of the URL. The common HTML file is here:


and its contents are:

<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN" "http://www.w3.org/TR/html4/loose.dtd">
    <meta http-equiv="content-type" content="text/html; charset=UTF-8">
    <meta name="viewport" content="width=device-width, initial-scale=1.0">
    <title>OpenDataKit Common Javascript Framework</title>
    <link rel="stylesheet" type="text/css" id="custom-styles" />
    <link rel="stylesheet" type="text/css" id="theme" href="libs/bootstrap-3.3.7-    dist/css/bootstrap.min.css" />
    <link rel="stylesheet" type="text/css" href="../config/assets/css/odk-survey.css" />
    <link rel="stylesheet" type="text/css" id="theme" href="libs/spinner/waitMe.css" />
    <script type="text/javascript" src="../config/assets/framework/frameworkDefinitions.js"></script>
    <script type="text/javascript" src="../config/assets/commonDefinitions.js"></script>
    <script type="text/javascript" src="js/odkCommon.js"></script>
    <script type="text/javascript" src="js/odkData.js"></script>
    <script type="text/javascript" src="tables/js/odkTables.js"></script>
    <script type="text/javascript" src="survey/js/odkSurvey.js"></script>
    <script type="text/javascript" src="survey/js/odkSurveyStateManagement.js"></script>
    <noscript>This page requires javascript and a Chrome or WebKit browser</noscript>
    <div id="block-ui"></div>
    <div class="odk-page">
        <div class="odk-screen">
            <div class="odk-toolbar"></div>
            <div class="odk-scroll">
                <div class="odk-container">Please wait...</div>
            <div class="odk-footer"></div>
    <script type="text/javascript" data-main="survey/js/main" src="libs/require.2.3.3.js"></script>

This loads a /config/assets/css/odk-survey.css file that users can customize, loads the common JavaScript wrapper objects and translation files, and finally triggers requirejs to load the framework and (eventually) process the window.location.hash to load and interpret the form definition.

The requirejs module management framework, under the direction of the /system/survey/js/main.js configuration and initialization file, loads the JavaScript files used by the Survey form framework.

Listed alphabetically, these are:

  • builder - responsible for reading the formDef.json and initializing the controller with the list of prompts in the survey.

  • controller - handles the logic for moving from one prompt to the next; this includes pre- and post- actions and performing the validation logic.

  • database - Handles the interactions with the odkData interface to the database. This also constructs and maintains the in-memory model description holding the form definition and the instance’s data and of the structure of the table in which it is stored.

  • databaseUtils - contains utility functions for transforming between the database storage strings and the JavaScript reconstructions in the model.

  • formulaFunctions - common functions accessible from the user's JavaScript eval environment (for use within their formulas).

  • handlebarsHelpers - Handlebars helper functions for use within handlebars templates. These are invoked via {{helperFunction arg1}} or {{helperFunction arg1 arg2}} within the handlebars templates.

  • main - the requirejs configuration and initialization file loaded via index.html that guides the JavaScript loading process. It waits for various components to load, cleans up the WebKit URL, and invokes parsequery.changeUrlHash(ctxt).

  • odkSurvey - simple wrapper for invoking the various media capture actions exposed by Survey

  • odkSurveyStateManagement - this is used only within App Designer to simulate the injected Java interface of the same name.

  • opendatakit - a random collection of methods that don't quite belong anywhere. Some of these cache and wrap requests to the odkCommon layer.

  • parsequery - responsible for parsing the hash fragment and triggering the building of the form, the triggering the initialization of the data table, changing of the viewed page, etc.

  • prompts - the core set of prompts defined by the Survey JavaScript framework. The first of these, base, defines the basic operation of a prompt.

  • promptTypes - due to the way requirejs works, this defines an empty object into which the prompts (above) are inserted.

  • screenManager - handles the rendering of a screen, including any please-wait or other in-progress notifications, and the events that initiate actions on that screen (for example, change language, swipe left/right, back/forward button clicks). Many of those actions invoke methods on the controller to complete. Note that rendering of the prompts within a screen (equivalent to an ODK Collect field-list) are handled within the definition of the screen.

  • screens - the core set of screen renderers defined by the Survey JavaScript framework. This includes the templating screen for customized layouts and the standard screen renderer.

  • screenTypes - due to the way requirejs works, this defines an empty object into which the screens (above) are inserted.

Survey Control Flow Overview

index.html Initialization Sequence

The index.html file explicitly loads these script files:
  • frameworkDefinitions.js - translations for standard Survey buttons and prompts

  • commonDefinitions.js - application-wide translations defined by the user

  • odkCommon.js - wrapper object for odkCommonIf injected Java interface

  • odkData.js - wrapper object for odkDataIf injected Java interface

  • odkTables.js - wrapper object for odkTablesIf injected Java interface and convenience methods for Tables navigation actions.

  • odkSurvey.js - wrapper object providing convenience methods for media capture interactions.

  • odkSurveyStateManagement.js - mock object used only within App Designer to provide functionality equivalent to the injected Java interface by the same name.

  • require.js - the requirejs module management library

  • main.js - loaded indirectly by requirejs to begin the module-load process

The relatively rapid loading of index.html very quickly presents ‘Please wait…’ to the user. This is not internationalized. Once the Survey framework is initialized, this will change to an internationalized prompt (using the waiting_text translations), and then be replaced by the requested screen in the form (or first screen of the form) when the form definition is fully processed.


The main.js file declares the interdependencies among the various JavaScript frameworks. It relies on requirejs for package dependency management and loading. The code first loads jQuery and an extended regex library (for Unicode strings). Once those are loaded, it then loads additional 3rd party libraries and the main Survey JavaScript framework files via:

require([ 'spinner', 'databaseUtils', 'opendatakit', 'database', 'parsequery',
                        'builder', 'controller', 'd3', 'jqueryCsv', 'combodate'],
  function(...) {...})

Once the ODK-X frameworks has loaded, the body of the function is executed. The body then initializes the parsequery object (needed to avoid circular references):


And then either triggers a reload to clean up the window.location value or initiates the parsing of the formDef.json specified in the URL location.hash via:



parsequery has two main entry points. The first:

parsequery.changeUrlHash(ctxt) {
    // when complete:
    that.controller.registerQueuedActionAvailableListener(ctxt, opendatakit.getRefId());

parses the formDef and calls the controller to initiate the processing of data callbacks from the Java layer.

The second entry point is _prepAndSwitchUI, which is called deep within the processing performed inside changeUrlHash(ctxt) and also by the controller when opening a specific instanceId within a form. That entry point assumes that the tableId and formId have not changed from what they currently are.

parsequery._parseParameters(ctxt) has the following flow (accomplished with many asynchronous processing steps – arguments are omitted):

parsequery._parseParameters() {
    if ( !sameForm ) {
        controller.reset( function() {
            // webpage now displays "Please wait..." with translations
    } else {

// called to load the (new) formDef.json
parseQuery._parseFormDefFile() {
    requirejs( "formDef.json", function() {

// called to interpret hash parameters after formDef.json loaded
// If the tableId is changed, load information about the tableId
// from the database layer so we know what fields are in it.
// Otherwise, interpret the formDef.json and construct the
// javascript objects that are used to render that form.
// And, once the object tree is initialized, call
// _prepAndSwitchUI() to render the specified screen in that form.
parseQuery._parseQueryParameterContinuation() {
    if ( !sameTable ) {
        controller.reset( function() {
            // webpage now displays ‘Please wait...’ with translations
            // Load information about the tableId from the database
            // layer so we know what fields are in it.
            database.initializeTables(function() {
                // parse and construct form objects
                builder.buildSurvey( function() {
                    // render the specified screen in this form
    } else if ( !sameForm ) {
        controller.reset( function() {
            // webpage now displays ‘Please wait...’ with translations
            // parse and construct form objects
            builder.buildSurvey( function() {
                // render the specified screen in this form
    } else if ( !sameInstance ) {
        controller.reset( function() {
            // webpage now displays ‘Please wait...’ with translations
            // render the specified screen in this form
    } else {
        // render the specified screen in this form

// retrieve and cache information for the instanceId (row)
// being manipulated (if any) and render the specified screen
// in the current form
parseQuery._prepAndSwitchUI() {
    database.initializeInstance( function() {
        controller.startAtScreenPath(ctxt, screenPath);

From this flow, you can see that the rough sequence of flow is:

  1. controller.reset() is called to display ‘Please wait…’

  2. database.initializeTables() to retrieve metadata about the tableId.

  3. builder.buildSurvey() to process the raw formDef.json file.

  4. database.initializeInstance() creates the initial (largely empty) row of an instanceId (if it is new) and reads the data for the instanceId from the database (if it is pre-existing), sets the current instance id and populates the mdl with the values for that instance id.

  5. controller.startAtScreenPath() is called to direct the Survey JavaScript framework to display the requested screen.

  6. controller.registerQueuedActionAvailableListener() is called to initiate the processing of any Java data callbacks (for instance, responses from intents).


Builder's only entry point is buildSurvey. This attempts to load several well-known files and then processes the formDef.json.

It begins by attempting to load (in order):


The file tableSpecificDefinitions.js contains the translations described earlier.

The customScreenTypes.js file contains user-defined screen types. These should follow the constructions of the basic screens defined in /system/survey/js/screens.js and should be stored as property fields inside the screenTypes object.

The customPromptTypes.js file contains user-defined prompt types. These should follow the constructions of the basic prompts defined in /system/survey/js/prompts .js and should be stored as property fields inside the promptTypes object.

The column_types field in the specification object within the formDef.json is a map consisting of column names and their expected column types. This is used to convert ordinary text describing a calculation into JavaScript functions that perform the calculation (via eval). For simplicity, these column names are interpreted independent of the sheet within the XLSX file from which the formDef.json is constructed. The allowed values for column types is only partially extensible as it must be interpreted and processed within the builder. The valid column types are:

  • function

  • formula

  • formula(arg1[, arg2[,…]])

  • requirejs_path

Columns with the function type are expected to contain column values ({columnValue}) that are a text string that can be evaluated as a function definition – for example, {columnValue} would be something like: function() { return 3; }.

The formula type and the formula(...) type are expected to have {columnValue} be an expression that is the return value of a function. These are wrapped by the builder to construct either

function() { return ({columnValue}); }


function(arg1[, arg2[,...]) { return ({columnValue}); }

Function and formula column types have their content evaluated in the context of the methods exposed by formulaFunctions to produce JavaScript functions. Because they are evaluated within the formulaFunctions context, they only have limited access to the internals of the Survey framework. This intentionally limits their power and the potential for damage that they might otherwise wreak.

The requirejs_path type causes builder to prefix the path to the form's directory. This supports referencing custom prompt templates and, potentially, images and other media, that are stored in the form directory.

The default column_types map can be extended in the XLSX file by defining a column_types sheet with headings that are column names and a single row beneath that defines the column type for that column name.

The default column_types map consists of:

    _screen_block: 'function',
    condition: 'formula',
    constraint: 'formula',
    required: 'formula',
    calculation: 'formula', // 'assign' prompt and on calculates sheet.
    newRowInitialElementKeyToValueMap: 'formula',
    openRowInitialElementKeyToValueMap: 'formula',
    selectionArgs: 'formula',
    url: 'formula', // external_link prompt
    uri: 'formula', // queries
    callback: 'formula(context)', // queries
    choice_filter: 'formula(choice_item)', // expects "choice_item" context arg.
    templatePath: 'requirejs_path'

Builder uses the column_types field in the specification object within the formDef.json to convert fields (column names) into their appropriate types. This conversion consists of a a full traversal of content from the calculates, settings, choices, queries, and all the survey sheets in the original XLSX file.

Next, for each of the survey sheets, builder creates Backbone instances of the prompt types referenced on those sheets, one instance for each declared prompt. These instances fold the field definitions the user specified in the XLSX file on top of the default values provided by the prompt definitions (and custom prompt definitions), allowing the user to customize the prompt through explicit changes in the XLSX file. These prompt instances are used when rendering the survey.

Lastly, the builder attempts to load:


It then attempts to load:


Or, if that doesn't exist, it examines the formDef.json to see if there was a theme defined on the settings sheet of the XLSX file and attempts to load:


And, lastly, it examines the formDef.json to see if there was a font-size defined on the settings sheet of the XLSX file and attempts to set it in the body:

$('body').css("font-size", fontSize.value);


The Survey database layer is a fairly thin wrapper around the odkData object. It maintains a cache of all of the field values in the referenced instanceId (row) within the current form. This cache is synchronously referenced and modified within the presentation layer and asynchronously updated via calls to the odkData object. In general, these asynchronous writes occur during lose-focus event processing.

Additionally, it maintains a copy of the properties of that table (for example, display name of the table and display names of the fields) and a description of the field types in the database table (the table definition). These are returned via the odkData object. This information is used within Survey to enable formulas to refer to field values either via their elementPath or via the database column in which they are stored (elementKey). A prime example of this is a geopoint. If the name of the geopoint field is mylocation then the individual latitude, longitude, etc. values are maintained within the cache as individual keys within a mylocation object – you can refer to them naturally as mylocation.latitude, mylocation.longitude, etc. This is the elementPath representation of these fields. However, within the database layer, these are stored as individual columns with column names of mylocation_latitude, mylocation_longitude etc. That is the elementKey representation. A similar transformation occurs for file attachments and any user-defined complex data type (multi-valued prompts). Simple select-multiple prompts, which manipulate arrays of values, have an elementPath representation within the cache as a Javascript array of selected values. Within the database layer, their elementKey representation is a JSON serialization of this array (in contrast, select-multiple prompts that reference linked tables would not store their selections in the dominant data table but rely upon filter conditions and storing a (foreign) key in the subordinate table, or in an association table, to establish their linkage).

The support this synchronous cache and this data abstraction, the main entry points for this layer can be divided into 4 sections:

Retrieving Information About a Database Table

Two methods:

  • initializeTables(ctxt, formDef, tableId, formPath)

  • readTableDefinition(ctxt, formDef, tableId, formPath)

The first is called during the initial loading of the form; the second is used by linked table prompts.

Creating and Deleting a Database Row

Five methods:

  • initializeInstance(ctxt, model, formId, instanceId, sameInstance, keyValueMap)

  • get_linked_instances(ctxt, dbTableName, selection, selectionArgs, displayElementName, orderBy)

  • save_all_changes(ctxt, model, formId, instanceId, asComplete)

  • ignore_all_changes(ctxt, model, formId, instanceId)

  • delete_checkpoints_and_row(ctxt, model, instanceId)

The first method, initializeInstance is used to initialize the synchronous cache with data values. It takes a boolean, sameInstance that is true if this is a reload of values for the current instanceId (row). It also takes a map of data changes keyValueMap to apply to this instance.

If sameInstance is true, this array is ignored.

If sameInstance is false and instanceId is null (we are not yet editing a row) then any initial values for the form's session variables that are specified in the keyValueMap are applied, and any initial values for any of the row's fields are ignored.

If sameInstance is false and instanceId is not null, the row's values are fetched from the database. If the row does not exist, it is initialized with the default values specified in the form for each of the row's fields, and then those changes are overlaid with the changes specified in the keyValueMap. And, finally, any initial values for the form's session variables that are specified within the keyValueMap are applied.

The second method, get_linked_instances is used by linked table prompts to retrieve rows from other data tables (for example, for linked table prompts).

The remaining methods (save_all_changes, ignore_all_changes and delete_checkpoints_and_row) manage the retention and deletion of the row in the database table.

Getting and Modifying Fields In a Database Row

Five methods:

  • setValueDeferredChange( name, value )

  • getDataValue(name)

  • getInstanceMetaDataValue(name)

  • applyDeferredChanges(ctxt)

  • setInstanceMetaData(ctxt, name, value)

The first 3 of these methods are the standard setters and getters of values. In general, the metadata fields of a row are read-only within Survey JavaScript. For this reason, there is no synchronous setter method for these fields.

The last 2 methods, applyDeferredChanges and setInstanceMetaData, are used internally within the Survey JavaScript framework to flush the changes in the synchronous cache through to the database via calls to odkData. Nearly all manipulation of a row's instance metadata is done within the Java layer. The exception is the changing of the current row's locale, which is effected via the call to setInstanceMetaData.

Utility Functions for Parsing Selection and Order-By Clauses

Two methods:

  • convertSelectionString(linkedModel, selection)

  • convertOrderByString(linkedModel, order_by)

These functions examine where clauses and order-by clauses to replace any elementPath expressions with elementKey values. Because this is not within the database layer, these conversions are not entirely fool-proof.


The initial load of a form ends with a call to controller.startAtScreenPath() followed by a call to controller.registerQueuedActionAvailableListener().

The controller object is responsible for navigating the form, ensuring that required fields are populated, that constraints are applied, that all validation logic is executed, and that appropriate actions are taken when the user launches an external application (for example, media capture), launches a sub-form, saves the form, exits without saving, or elects to delete a row from the database.

To implement back button functionality, the controller maintains a history of how the user has navigated through the form. This navigation history is necessary because there is no fixed execution path through an Survey form (user-directed navigation is one of the big changes between the javarosa-based tools and Survey). The odkSurveyStateManagement injected Java interface provides the underlying storage mechanism for this functionality and is directly called by controller during its processing.

The types of actions that the controller can perform, and how these are defined in the formDef.json will be described later in this document. At this time, it is sufficient to know that the controller is executing a program that performs actions, such as the rendering of a screen containing one or more prompts, as well as performing conditional and unconditional branches within that program.

The controller's progress through this program is tracked by the history stack maintained within odkSurveyStateManagement and the top of that history stack identifies the operation which the controller is currently executing. The controller's (vastly simplified) form processing flow is as follows:

controller.startAtScreenPath(ctxt, screenPath) {
    var op = operation corresponding to screenPath.
// starting at the operation referenced by 'op',
// execute operations until a screen is rendered
controller._doActionAt(op) {
    // when the above completes, we are
    // given a screenOp (screen rendering
    // operation) to transition to, or
    // have already produced a pop-up to
    // communicate an error to the user.
    if ( screenOp !== null ) {
        controller.setScreenWithMessagePopup(ctxt, screenOp, ...);
// main execution loop
controller._doActionAtLoop(op) {
    while () {
        switch ( op._token_type ) {
        case "goto_label":
            // jump (possibly conditionally)
            // to another operation
        // other control flow options
        // some of these can return out
        // of this while without returning
        // a screen rendering operation.
        // any that do will have already
        // produced an alert or error pop-up
        case "assign":
            // do assignment
        case "begin_screen":
            // render a screen
            return op; // the ‘screenOp’ in _doActionAt();
// render a screen
controller.setScreenWithMessagePopup(ctxt, screenOp, options, msg) {
    // set up a 500ms delay timer to render the ‘msg’ pop-up
    // so that the UI can settle on the new page before we
    // display the message. Otherwise, it might be lost
    // during the rendering of the screen.
    setTimeout(function() {
    }, 500);
    screenManager.setScreen(ctxt, screenOp, options);

Simply put, the processing flow eventually calls screenManager to display a screen (via setScreen(ctxt, screenOp, options)) and perhaps also shows a pop-up with some sort of alert or error message (via showScreenPopup(m)).

When the next button is pressed or the screen is swiped forwards, the framework calls controller.gotoNextScreen() which verifies that all required fields are filled-in and all constraints are applied. It then triggers much the same processing sequence – calling doActionAt() with the operation after the currently-rendered screen.

When the back button is pressed or the screen is swiped backward, the framework calls controller.gotoPreviousScreen() which pops the operation history stack for the current survey sheet until a screen-rendering operation is found, and that screen is then rendered. And, if the history for the current survey sheet is exhausted, then the contents screen for that sheet is displayed.

Finally, returning to the discussion of the control flow on the initial load of a form, after the current screen is rendered, the call to controller.registerQueuedActionAvailableListener() causes an action listener to be registered with odkCommon and then calls that listener to process any results that became available before the listener was registered. If there are any results from a previous odkCommon.doAction(...intentArgs...) request (for example, a media-file capture request), then the controller's action listener will interpret the results to identify what prompt in the current screen should receive and process these results and then invoke that prompt to complete the processing. Otherwise, if there are no results, no additional actions are taken. This completes the control flow on the initial load of the form.


The screenManager provides event handling for swiping and the navigation bars at the top and bottom of a screen. It delegates to the screen object to construct the DOM representation for that content and also delegates to the screen object to register and unregister event handlers for any other DOM elements via calls to recursiveUndelegateEvents() and recursiveDelegateEvents(). Those event handlers are expected to be defined in the Backbone-based screen objects and prompt objects.

The high-level actions of the screen manager are:

screenManager.setScreen(ctxt, screen) {
    // show "loading..." spinner
    // stop processing all events on the current screen
    // construct the DOM objects in the page (heavily nested)
    screen.buildRenderContext(... {
        screen.render(... {
            screenManager.activeScreen = screen;
            // replace the screen
    // and via a ctxt.terminalContext()  registration
    // so that the DOM replacement and redraw can take effect


The screen object determines the set of prompts that should be displayed and lays them out. The custom screen example shows how this can be done within an arbitrary HTML template by using ids on DOM elements to identify where the inner HTML for a prompt should be injected.

Immediately prior to screen rendering, any unsaved changes in data values are asynchronously flushed to the database.

The screen object also enforces required fields and constraints and can reject any attempts by the controller object to move off of this screen or pop-up a confirmation for the user to accept.

See the screens.js file.


Prompts register event handlers for their DOM elements and are responsible for restoring and saving values displayed in those DOM elements into the synchronous data cache and for validating those values and enforcing any constraints (if so directed).

See the prompts.js file.

Survey Controller Actions

As mentioned earlier, the main processing loop within the controller executes a program derived from the form's XLSX file and encoded in the formDef.json. The 10 primitive operations in this program are described in ODK-X Survey Controller Actions.