Merge pull request #1031 from kchadha/block_serialization_compression

Block serialization compression

package.json

@@ -57,7 +57,7 @@
     "promise": "8.0.1",
     "scratch-audio": "latest",
     "scratch-blocks": "latest",
-    "scratch-parser": "latest",
+    "scratch-parser": "^3.0.0",
     "scratch-render": "latest",
     "scratch-storage": "^0.4.0",
     "script-loader": "0.7.2",

src/blocks/scratch3_data.js

@@ -74,7 +74,7 @@ class Scratch3DataBlocks {
     addToList (args, util) {
         const list = util.target.lookupOrCreateList(
             args.LIST.id, args.LIST.name);
-        list.value.push(args.ITEM);
+        if (list.value.length < Scratch3DataBlocks.LIST_ITEM_LIMIT) list.value.push(args.ITEM);
     }
 
     deleteOfList (args, util) {
@@ -98,7 +98,14 @@ class Scratch3DataBlocks {
         if (index === Cast.LIST_INVALID) {
             return;
         }
+        const listLimit = Scratch3DataBlocks.LIST_ITEM_LIMIT;
+        if (index > listLimit) return;
         list.value.splice(index - 1, 0, item);
+        if (list.value.length > listLimit) {
+            // If inserting caused the list to grow larger than the limit,
+            // remove the last element in the list
+            list.value.pop();
+        }
     }
 
     replaceItemOfList (args, util) {
@@ -144,6 +151,14 @@ class Scratch3DataBlocks {
         }
         return false;
     }
+
+    /**
+     * Type representation for list variables.
+     * @const {string}
+     */
+    static get LIST_ITEM_LIMIT () {
+        return 200000;
+    }
 }
 
 module.exports = Scratch3DataBlocks;

src/import/load-sound.js

@@ -12,6 +12,10 @@ const log = require('../util/log');
  */
 const loadSoundFromAsset = function (sound, soundAsset, runtime) {
     sound.assetId = soundAsset.assetId;
+    if (!runtime.audioEngine) {
+        log.error('No audio engine present; cannot load sound asset: ', sound.md5);
+        return Promise.resolve(sound);
+    }
     return runtime.audioEngine.decodeSound(Object.assign(
         {},
         sound,
@@ -35,10 +39,6 @@ const loadSound = function (sound, runtime) {
         log.error('No storage module present; cannot load sound asset: ', sound.md5);
         return Promise.resolve(sound);
     }
-    if (!runtime.audioEngine) {
-        log.error('No audio engine present; cannot load sound asset: ', sound.md5);
-        return Promise.resolve(sound);
-    }
     const idParts = StringUtil.splitFirst(sound.md5, '.');
     const md5 = idParts[0];
     const ext = idParts[1].toLowerCase();

src/serialization/deserialize-assets.js

@@ -30,11 +30,7 @@ const deserializeSound = function (sound, runtime, zip, assetFileName) {
         // This sound has already been cached.
         return Promise.resolve(null);
     }
-    if (!zip) {
-        // TODO adding this case to make integration tests pass, need to rethink
-        // the entire structure of saving/loading here (w.r.t. differences between
-        // loading from local zip file or from server)
-        log.error('Zipped assets were not provided.');
+    if (!zip) { // Zip will not be provided if loading project json from server
         return Promise.resolve(null);
     }
     const soundFile = zip.file(fileName);
@@ -93,11 +89,7 @@ const deserializeCostume = function (costume, runtime, zip, assetFileName) {
         return Promise.resolve(null);
     }
 
-    if (!zip) {
-        // TODO adding this case to make integration tests pass, need to rethink
-        // the entire structure of saving/loading here (w.r.t. differences between
-        // loading from local zip file or from server)
-        log.error('Zipped assets were not provided.');
+    if (!zip) { // Zip will not be provided if loading project json from server
         return Promise.resolve(null);
     }
 
@@ -106,14 +98,11 @@ const deserializeCostume = function (costume, runtime, zip, assetFileName) {
         log.error(`Could not find costume file associated with the ${costume.name} costume.`);
         return Promise.resolve(null);
     }
-    let dataFormat = null;
     let assetType = null;
     const costumeFormat = costume.dataFormat.toLowerCase();
     if (costumeFormat === 'svg') {
-        dataFormat = storage.DataFormat.SVG;
         assetType = storage.AssetType.ImageVector;
-    } else if (costumeFormat === 'png') {
-        dataFormat = storage.DataFormat.PNG;
+    } else if (['png', 'bmp', 'jpeg', 'jpg', 'gif'].indexOf(costumeFormat) >= 0) {
         assetType = storage.AssetType.ImageBitmap;
     } else {
         log.error(`Unexpected file format for costume: ${costumeFormat}`);
@@ -126,7 +115,8 @@ const deserializeCostume = function (costume, runtime, zip, assetFileName) {
     return costumeFile.async('uint8array').then(data => {
         storage.builtinHelper.cache(
             assetType,
-            dataFormat,
+            // TODO eventually we want to map non-png's to their actual file types?
+            costumeFormat,
             data,
             assetId
         );

src/serialization/sb2.js

@@ -215,7 +215,7 @@ const parseScratchObject = function (object, runtime, extensions, topLevel, zip)
     const sprite = new Sprite(blocks, runtime);
     // Sprite/stage name from JSON.
     if (object.hasOwnProperty('objName')) {
-        sprite.name = object.objName;
+        sprite.name = topLevel ? 'Stage' : object.objName;
     }
     // Costumes from JSON.
     const costumePromises = [];

src/serialization/sb3.js

@@ -1,6 +1,6 @@
 /**
  * @fileoverview
- * Partial implementation of a SB3 serializer and deserializer. Parses provided
+ * An SB3 serializer and deserializer. Parses provided
  * JSON and then generates all needed scratch-vm runtime structures.
  */
 
@@ -8,6 +8,8 @@ const vmPackage = require('../../package.json');
 const Blocks = require('../engine/blocks');
 const Sprite = require('../sprites/sprite');
 const Variable = require('../engine/variable');
+const log = require('../util/log');
+const uid = require('../util/uid');
 
 const {loadCostume} = require('../import/load-costume.js');
 const {loadSound} = require('../import/load-sound.js');
@@ -25,14 +27,156 @@ const {deserializeCostume, deserializeSound} = require('./deserialize-assets.js'
  * @property {Map.<string, string>} extensionURLs - map of ID => URL from project metadata. May not match extensionIDs.
  */
 
-const serializeBlock = function (block) {
+// Constants used during serialization and deserialization
+const INPUT_SAME_BLOCK_SHADOW = 1; // unobscured shadow
+const INPUT_BLOCK_NO_SHADOW = 2; // no shadow
+const INPUT_DIFF_BLOCK_SHADOW = 3; // obscured shadow
+// There shouldn't be a case where block is null, but shadow is present...
+
+// Constants referring to 'primitive' blocks that are usually shadows,
+// or in the case of variables and lists, appear quite often in projects
+// math_number
+const MATH_NUM_PRIMITIVE = 4; // there's no reason these constants can't collide
+// math_positive_number
+const POSITIVE_NUM_PRIMITIVE = 5; // with the above, but removing duplication for clarity
+// math_whole_number
+const WHOLE_NUM_PRIMITIVE = 6;
+// math_integer
+const INTEGER_NUM_PRIMITIVE = 7;
+// math_angle
+const ANGLE_NUM_PRIMITIVE = 8;
+// colour_picker
+const COLOR_PICKER_PRIMITIVE = 9;
+// text
+const TEXT_PRIMITIVE = 10;
+// event_broadcast_menu
+const BROADCAST_PRIMITIVE = 11;
+// data_variable
+const VAR_PRIMITIVE = 12;
+// data_listcontents
+const LIST_PRIMITIVE = 13;
+
+// Map block opcodes to the above primitives and the name of the field we can use
+// to find the value of the field
+const primitiveOpcodeInfoMap = {
+    math_number: [MATH_NUM_PRIMITIVE, 'NUM'],
+    math_positive_number: [POSITIVE_NUM_PRIMITIVE, 'NUM'],
+    math_whole_number: [WHOLE_NUM_PRIMITIVE, 'NUM'],
+    math_integer: [INTEGER_NUM_PRIMITIVE, 'NUM'],
+    math_angle: [ANGLE_NUM_PRIMITIVE, 'NUM'],
+    colour_picker: [COLOR_PICKER_PRIMITIVE, 'COLOUR'],
+    text: [TEXT_PRIMITIVE, 'TEXT'],
+    event_broadcast_menu: [BROADCAST_PRIMITIVE, 'BROADCAST_OPTION'],
+    data_variable: [VAR_PRIMITIVE, 'VARIABLE'],
+    data_listcontents: [LIST_PRIMITIVE, 'LIST']
+};
+
+/**
+ * Serializes primitives described above into a more compact format
+ * @param {object} block the block to serialize
+ * @return {array} An array representing the information in the block,
+ * or null if the given block is not one of the primitives described above.
+ */
+const serializePrimitiveBlock = function (block) {
+    // Returns an array represeting a primitive block or null if not one of
+    // the primitive types above
+    if (primitiveOpcodeInfoMap.hasOwnProperty(block.opcode)) {
+        const primitiveInfo = primitiveOpcodeInfoMap[block.opcode];
+        const primitiveConstant = primitiveInfo[0];
+        const fieldName = primitiveInfo[1];
+        const field = block.fields[fieldName];
+        const primitiveDesc = [primitiveConstant, field.value];
+        if (block.opcode === 'event_broadcast_menu') {
+            primitiveDesc.push(field.id);
+        } else if (block.opcode === 'data_variable' || block.opcode === 'data_listcontents') {
+            primitiveDesc.push(field.id);
+            if (block.topLevel) {
+                primitiveDesc.push(block.x ? Math.round(block.x) : 0);
+                primitiveDesc.push(block.y ? Math.round(block.y) : 0);
+            }
+        }
+        return primitiveDesc;
+    }
+    return null;
+};
+
+/**
+ * Serializes the inputs field of a block in a compact form using
+ * constants described above to represent the relationship between the
+ * inputs of this block (e.g. if there is an unobscured shadow, an obscured shadow
+ * -- a block plugged into a droppable input -- or, if there is just a block).
+ * Based on this relationship, serializes the ids of the block and shadow (if present)
+ *
+ * @param {object} inputs The inputs to serialize
+ * @return {object} An object representing the serialized inputs
+ */
+const serializeInputs = function (inputs) {
+    const obj = Object.create(null);
+    for (const inputName in inputs) {
+        if (!inputs.hasOwnProperty(inputName)) continue;
+        // if block and shadow refer to the same block, only serialize one
+        if (inputs[inputName].block === inputs[inputName].shadow) {
+            // has block and shadow, and they are the same
+            obj[inputName] = [
+                INPUT_SAME_BLOCK_SHADOW,
+                inputs[inputName].block
+            ];
+        } else if (inputs[inputName].shadow === null) {
+            // does not have shadow
+            obj[inputName] = [
+                INPUT_BLOCK_NO_SHADOW,
+                inputs[inputName].block
+            ];
+        } else {
+            // block and shadow are both present and are different
+            obj[inputName] = [
+                INPUT_DIFF_BLOCK_SHADOW,
+                inputs[inputName].block,
+                inputs[inputName].shadow
+            ];
+        }
+    }
+    return obj;
+};
+
+/**
+ * Serialize the fields of a block in a more compact form.
+ * @param {object} fields The fields object to serialize
+ * @return {object} An object representing the serialized fields
+ */
+const serializeFields = function (fields) {
+    const obj = Object.create(null);
+    for (const fieldName in fields) {
+        if (!fields.hasOwnProperty(fieldName)) continue;
+        obj[fieldName] = [fields[fieldName].value];
+        if (fields[fieldName].hasOwnProperty('id')) {
+            obj[fieldName].push(fields[fieldName].id);
+        }
+    }
+    return obj;
+};
+
+/**
+ * Serialize the given block in the SB3 format with some compression of inputs,
+ * fields, and primitives.
+ * @param {object} block The block to serialize
+ * @return {object | array} A serialized representation of the block. This is an
+ * array if the block is one of the primitive types described above or an object,
+ * if not.
+ */
+const serializeBlock = function (block) {
+    const serializedPrimitive = serializePrimitiveBlock(block);
+    if (serializedPrimitive) return serializedPrimitive;
+    // If serializedPrimitive is null, proceed with serializing a non-primitive block
     const obj = Object.create(null);
-    obj.id = block.id;
     obj.opcode = block.opcode;
+    // NOTE: this is extremely important to serialize even if null;
+    // not serializing `next: null` results in strange behavior with block
+    // execution
     obj.next = block.next;
     obj.parent = block.parent;
-    obj.inputs = block.inputs;
-    obj.fields = block.fields;
+    obj.inputs = serializeInputs(block.inputs);
+    obj.fields = serializeFields(block.fields);
     obj.topLevel = block.topLevel ? block.topLevel : false;
     obj.shadow = block.shadow;
     if (block.topLevel) {
@@ -49,15 +193,108 @@ const serializeBlock = function (block) {
     return obj;
 };
 
+/**
+ * Compresses the serialized inputs replacing block/shadow ids that refer to
+ * one of the primitives with the primitive itself. E.g.
+ *
+ * blocks: {
+ *      aUidForMyBlock: {
+ *          inputs: {
+ *               MYINPUT: [1, 'aUidForAnUnobscuredShadowPrimitive']
+ *          }
+ *      },
+ *      aUidForAnUnobscuredShadowPrimitive: [4, 10]
+ *      // the above is a primitive representing a 'math_number' with value 10
+ * }
+ *
+ * becomes:
+ *
+ * blocks: {
+ *      aUidForMyBlock: {
+ *          inputs: {
+ *               MYINPUT: [1, [4, 10]]
+ *          }
+ *      }
+ * }
+ * Note: this function modifies the given blocks object in place
+ * @param {object} block The block with inputs to compress
+ * @param {objec} blocks The object containing all the blocks currently getting serialized
+ * @return {object} The serialized block with compressed inputs
+ */
+const compressInputTree = function (block, blocks) {
+    // This is the second pass on the block
+    // so the inputs field should be an object of key - array pairs
+    const serializedInputs = block.inputs;
+    for (const inputName in serializedInputs) {
+        // don't need to check for hasOwnProperty because of how we constructed
+        // inputs
+        const currInput = serializedInputs[inputName];
+        // traverse currInput skipping the first element, which describes whether the block
+        // and shadow are the same
+        for (let i = 1; i < currInput.length; i++) {
+            if (!currInput[i]) continue; // need this check b/c block/shadow can be null
+            const blockOrShadowID = currInput[i];
+            // replace element of currInput directly
+            // (modifying input block directly)
+            const blockOrShadow = blocks[blockOrShadowID];
+            if (Array.isArray(blockOrShadow)) {
+                currInput[i] = blockOrShadow;
+                // Modifying blocks in place!
+                delete blocks[blockOrShadowID];
+            }
+        }
+    }
+    return block;
+};
+
+/**
+ * Serialize the given blocks object (representing all the blocks for the target
+ * currently being serialized.)
+ * @param {object} blocks The blocks to be serialized
+ * @return {object} The serialized blocks with compressed inputs and compressed
+ * primitives.
+ */
 const serializeBlocks = function (blocks) {
     const obj = Object.create(null);
     for (const blockID in blocks) {
         if (!blocks.hasOwnProperty(blockID)) continue;
-        obj[blockID] = serializeBlock(blocks[blockID]);
+        obj[blockID] = serializeBlock(blocks[blockID], blocks);
+    }
+    // once we have completed a first pass, do a second pass on block inputs
+    for (const blockID in obj) {
+        // don't need to do the hasOwnProperty check here since we
+        // created an object that doesn't get extra properties/functions
+        const serializedBlock = obj[blockID];
+        // caution, this function deletes parts of this object in place as
+        // it's traversing it
+        obj[blockID] = compressInputTree(serializedBlock, obj);
+        // second pass on connecting primitives to serialized inputs directly
+    }
+    // Do one last pass and remove any top level shadows (these are caused by
+    // a bug: LLK/scratch-vm#1011, and this pass should be removed once that is
+    // completely fixed)
+    for (const blockID in obj) {
+        const serializedBlock = obj[blockID];
+        // If the current block is serialized as a primitive (e.g. it's an array
+        // instead of an object), AND it is not one of the top level primitives
+        // e.g. variable getter or list getter, then it should be deleted as it's
+        // a shadow block, and there are no blocks that reference it, otherwise
+        // they would have been compressed in the last pass)
+        if (Array.isArray(serializedBlock) &&
+            [VAR_PRIMITIVE, LIST_PRIMITIVE].indexOf(serializedBlock) < 0) {
+            log.warn(`Found an unexpected top level primitive with block ID: ${
+                blockID}; deleting it from serialized blocks.`);
+            delete obj[blockID];
+        }
     }
     return obj;
 };
 
+/**
+ * Serialize the given costume.
+ * @param {object} costume The costume to be serialized.
+ * @return {object} A serialized representation of the costume.
+ */
 const serializeCostume = function (costume) {
     const obj = Object.create(null);
     obj.assetId = costume.assetId;
@@ -69,17 +306,22 @@ const serializeCostume = function (costume) {
     // but that change should be made carefully since it is very
     // pervasive
     obj.md5ext = costume.md5;
-    obj.dataFormat = costume.dataFormat;
+    obj.dataFormat = costume.dataFormat.toLowerCase();
     obj.rotationCenterX = costume.rotationCenterX;
     obj.rotationCenterY = costume.rotationCenterY;
     return obj;
 };
 
+/**
+ * Serialize the given sound.
+ * @param {object} sound The sound to be serialized.
+ * @return {object} A serialized representation of the sound.
+ */
 const serializeSound = function (sound) {
     const obj = Object.create(null);
     obj.assetId = sound.assetId;
     obj.name = sound.name;
-    obj.dataFormat = sound.dataFormat;
+    obj.dataFormat = sound.dataFormat.toLowerCase();
     obj.format = sound.format;
     obj.rate = sound.rate;
     obj.sampleCount = sound.sampleCount;
@@ -92,17 +334,63 @@ const serializeSound = function (sound) {
     return obj;
 };
 
+/**
+ * Serialize the given variables object.
+ * @param {object} variables The variables to be serialized.
+ * @return {object} A serialized representation of the variables. They get
+ * separated by type to compress the representation of each given variable and
+ * reduce duplicate information.
+ */
+const serializeVariables = function (variables) {
+    const obj = Object.create(null);
+    // separate out variables into types at the top level so we don't have
+    // keep track of a type for each
+    obj.variables = Object.create(null);
+    obj.lists = Object.create(null);
+    obj.broadcasts = Object.create(null);
+    for (const varId in variables) {
+        const v = variables[varId];
+        if (v.type === Variable.BROADCAST_MESSAGE_TYPE) {
+            obj.broadcasts[varId] = v.value; // name and value is the same for broadcast msgs
+            continue;
+        }
+        if (v.type === Variable.LIST_TYPE) {
+            obj.lists[varId] = [v.name, v.value];
+            continue;
+        }
+
+        // otherwise should be a scalar type
+        obj.variables[varId] = [v.name, v.value];
+        // only scalar vars have the potential to be cloud vars
+        if (v.isPersistent) obj.variables[varId].push(true);
+    }
+    return obj;
+};
+
+/**
+ * Serialize the given target. Only serialize properties that are necessary
+ * for saving and loading this target.
+ * @param {object} target The target to be serialized.
+ * @return {object} A serialized representation of the given target.
+ */
 const serializeTarget = function (target) {
     const obj = Object.create(null);
     obj.isStage = target.isStage;
-    obj.name = target.name;
-    obj.variables = target.variables; // This means that uids for variables will persist across saves/loads
+    obj.name = obj.isStage ? 'Stage' : target.name;
+    const vars = serializeVariables(target.variables);
+    obj.variables = vars.variables;
+    obj.lists = vars.lists;
+    obj.broadcasts = vars.broadcasts;
     obj.blocks = serializeBlocks(target.blocks);
     obj.currentCostume = target.currentCostume;
     obj.costumes = target.costumes.map(serializeCostume);
     obj.sounds = target.sounds.map(serializeSound);
-    if (!obj.isStage) {
-        // Stage does not need the following properties
+    if (target.hasOwnProperty('volume')) obj.volume = target.volume;
+    if (obj.isStage) { // Only the stage should have these properties
+        if (target.hasOwnProperty('tempo')) obj.tempo = target.tempo;
+        if (target.hasOwnProperty('videoTransparency')) obj.videoTransparency = target.videoTransparency;
+        if (target.hasOwnProperty('videoState')) obj.videoState = target.videoState;
+    } else { // The stage does not need the following properties, but sprites should
         obj.visible = target.visible;
         obj.x = target.x;
         obj.y = target.y;
@@ -117,7 +405,7 @@ const serializeTarget = function (target) {
 /**
  * Serializes the specified VM runtime.
  * @param  {!Runtime} runtime VM runtime instance to be serialized.
- * @return {object}    Serialized runtime instance.
+ * @return {object} Serialized runtime instance.
  */
 const serialize = function (runtime) {
     // Fetch targets
@@ -142,6 +430,226 @@ const serialize = function (runtime) {
     return obj;
 };
 
+/**
+ * Deserialize a block input descriptors. This is either a
+ * block id or a serialized primitive, e.g. an array
+ * (see serializePrimitiveBlock function).
+ * @param {string | array} inputDescOrId The block input descriptor to be serialized.
+ * @param {string} parentId The id of the parent block for this input block.
+ * @param {boolean} isShadow Whether or not this input block is a shadow.
+ * @param {object} blocks The entire blocks object currently in the process of getting serialized.
+ * @return {object} The deserialized input descriptor.
+ */
+const deserializeInputDesc = function (inputDescOrId, parentId, isShadow, blocks) {
+    if (!Array.isArray(inputDescOrId)) return inputDescOrId;
+    const primitiveObj = Object.create(null);
+    const newId = uid();
+    primitiveObj.id = newId;
+    primitiveObj.next = null;
+    primitiveObj.parent = parentId;
+    primitiveObj.shadow = isShadow;
+    primitiveObj.inputs = Object.create(null);
+    // need a reference to parent id
+    switch (inputDescOrId[0]) {
+    case MATH_NUM_PRIMITIVE: {
+        primitiveObj.opcode = 'math_number';
+        primitiveObj.fields = {
+            NUM: {
+                name: 'NUM',
+                value: inputDescOrId[1]
+            }
+        };
+        primitiveObj.topLevel = false;
+        break;
+    }
+    case POSITIVE_NUM_PRIMITIVE: {
+        primitiveObj.opcode = 'math_positive_number';
+        primitiveObj.fields = {
+            NUM: {
+                name: 'NUM',
+                value: inputDescOrId[1]
+            }
+        };
+        primitiveObj.topLevel = false;
+        break;
+    }
+    case WHOLE_NUM_PRIMITIVE: {
+        primitiveObj.opcode = 'math_whole_number';
+        primitiveObj.fields = {
+            NUM: {
+                name: 'NUM',
+                value: inputDescOrId[1]
+            }
+        };
+        primitiveObj.topLevel = false;
+        break;
+    }
+    case INTEGER_NUM_PRIMITIVE: {
+        primitiveObj.opcode = 'math_integer';
+        primitiveObj.fields = {
+            NUM: {
+                name: 'NUM',
+                value: inputDescOrId[1]
+            }
+        };
+        primitiveObj.topLevel = false;
+        break;
+    }
+    case ANGLE_NUM_PRIMITIVE: {
+        primitiveObj.opcode = 'math_angle';
+        primitiveObj.fields = {
+            NUM: {
+                name: 'NUM',
+                value: inputDescOrId[1]
+            }
+        };
+        primitiveObj.topLevel = false;
+        break;
+    }
+    case COLOR_PICKER_PRIMITIVE: {
+        primitiveObj.opcode = 'colour_picker';
+        primitiveObj.fields = {
+            COLOUR: {
+                name: 'COLOUR',
+                value: inputDescOrId[1]
+            }
+        };
+        primitiveObj.topLevel = false;
+        break;
+    }
+    case TEXT_PRIMITIVE: {
+        primitiveObj.opcode = 'text';
+        primitiveObj.fields = {
+            TEXT: {
+                name: 'TEXT',
+                value: inputDescOrId[1]
+            }
+        };
+        primitiveObj.topLevel = false;
+        break;
+    }
+    case BROADCAST_PRIMITIVE: {
+        primitiveObj.opcode = 'event_broadcast_menu';
+        primitiveObj.fields = {
+            BROADCAST_OPTION: {
+                name: 'BROADCAST_OPTION',
+                value: inputDescOrId[1],
+                id: inputDescOrId[2],
+                variableType: Variable.BROADCAST_MESSAGE_TYPE
+            }
+        };
+        primitiveObj.topLevel = false;
+        break;
+    }
+    case VAR_PRIMITIVE: {
+        primitiveObj.opcode = 'data_variable';
+        primitiveObj.fields = {
+            VARIABLE: {
+                name: 'VARIABLE',
+                value: inputDescOrId[1],
+                id: inputDescOrId[2],
+                variableType: Variable.SCALAR_TYPE
+            }
+        };
+        if (inputDescOrId.length > 3) {
+            primitiveObj.topLevel = true;
+            primitiveObj.x = inputDescOrId[3];
+            primitiveObj.y = inputDescOrId[4];
+        }
+        break;
+    }
+    case LIST_PRIMITIVE: {
+        primitiveObj.opcode = 'data_listcontents';
+        primitiveObj.fields = {
+            LIST: {
+                name: 'LIST',
+                value: inputDescOrId[1],
+                id: inputDescOrId[2],
+                variableType: Variable.LIST_TYPE
+            }
+        };
+        if (inputDescOrId.length > 3) {
+            primitiveObj.topLevel = true;
+            primitiveObj.x = inputDescOrId[3];
+            primitiveObj.y = inputDescOrId[4];
+        }
+        break;
+    }
+    default: {
+        log.error(`Found unknown primitive type during deserialization: ${JSON.stringify(inputDescOrId)}`);
+        return null;
+    }
+    }
+    blocks[newId] = primitiveObj;
+    return newId;
+};
+
+/**
+ * Deserialize the given block inputs.
+ * @param {object} inputs The inputs to deserialize.
+ * @param {string} parentId The block id of the parent block
+ * @param {object} blocks The object representing the entire set of blocks currently
+ * in the process of getting deserialized.
+ * @return {object} The deserialized and uncompressed inputs.
+ */
+const deserializeInputs = function (inputs, parentId, blocks) {
+    // Explicitly not using Object.create(null) here
+    // because we call prototype functions later in the vm
+    const obj = {};
+    for (const inputName in inputs) {
+        if (!inputs.hasOwnProperty(inputName)) continue;
+        const inputDescArr = inputs[inputName];
+        let block = null;
+        let shadow = null;
+        const blockShadowInfo = inputDescArr[0];
+        if (blockShadowInfo === INPUT_SAME_BLOCK_SHADOW) {
+            // block and shadow are the same id, and only one is provided
+            block = shadow = deserializeInputDesc(inputDescArr[1], parentId, true, blocks);
+        } else if (blockShadowInfo === INPUT_BLOCK_NO_SHADOW) {
+            block = deserializeInputDesc(inputDescArr[1], parentId, false, blocks);
+        } else { // assume INPUT_DIFF_BLOCK_SHADOW
+            block = deserializeInputDesc(inputDescArr[1], parentId, false, blocks);
+            shadow = deserializeInputDesc(inputDescArr[2], parentId, true, blocks);
+        }
+        obj[inputName] = {
+            name: inputName,
+            block: block,
+            shadow: shadow
+        };
+    }
+    return obj;
+};
+
+/**
+ * Deserialize the given block fields.
+ * @param {object} fields The fields to be deserialized
+ * @return {object} The deserialized and uncompressed block fields.
+ */
+const deserializeFields = function (fields) {
+    // Explicitly not using Object.create(null) here
+    // because we call prototype functions later in the vm
+    const obj = {};
+    for (const fieldName in fields) {
+        if (!fields.hasOwnProperty(fieldName)) continue;
+        const fieldDescArr = fields[fieldName];
+        obj[fieldName] = {
+            name: fieldName,
+            value: fieldDescArr[0]
+        };
+        if (fieldDescArr.length > 1) {
+            obj[fieldName].id = fieldDescArr[1];
+        }
+        if (fieldName === 'BROADCAST_OPTION') {
+            obj[fieldName].variableType = Variable.BROADCAST_MESSAGE_TYPE;
+        } else if (fieldName === 'VARIABLE') {
+            obj[fieldName].variableType = Variable.SCALAR_TYPE;
+        } else if (fieldName === 'LIST') {
+            obj[fieldName].variableType = Variable.LIST_TYPE;
+        }
+    }
+    return obj;
+};
+
 /**
  * Parse a single "Scratch object" and create all its in-memory VM objects.
  * @param {!object} object From-JSON "Scratch object:" sprite, stage, watcher.
@@ -167,6 +675,26 @@ const parseScratchObject = function (object, runtime, extensions, zip) {
         sprite.name = object.name;
     }
     if (object.hasOwnProperty('blocks')) {
+        for (const blockId in object.blocks) {
+            if (!object.blocks.hasOwnProperty(blockId)) continue;
+            const blockJSON = object.blocks[blockId];
+            if (Array.isArray(blockJSON)) {
+                // this is one of the primitives
+                // delete the old entry in object.blocks and replace it w/the
+                // deserialized object
+                delete object.blocks[blockId];
+                deserializeInputDesc(blockJSON, null, false, object.blocks);
+                continue;
+            }
+            blockJSON.id = blockId; // add id back to block since it wasn't serialized
+            const serializedInputs = blockJSON.inputs;
+            const deserializedInputs = deserializeInputs(serializedInputs, blockId, object.blocks);
+            blockJSON.inputs = deserializedInputs;
+            const serializedFields = blockJSON.fields;
+            const deserializedFields = deserializeFields(serializedFields);
+            blockJSON.fields = deserializedFields;
+        }
+        // Take a second pass to create objects and add extensions
         for (const blockId in object.blocks) {
             if (!object.blocks.hasOwnProperty(blockId)) continue;
             const blockJSON = object.blocks[blockId];
@@ -178,7 +706,6 @@ const parseScratchObject = function (object, runtime, extensions, zip) {
                 extensions.extensionIDs.add(extensionId);
             }
         }
-        // console.log(blocks);
     }
     // Costumes from JSON.
     const costumePromises = (object.costumes || []).map(costumeSource => {
@@ -237,19 +764,58 @@ const parseScratchObject = function (object, runtime, extensions, zip) {
     // Create the first clone, and load its run-state from JSON.
     const target = sprite.createClone();
     // Load target properties from JSON.
+    if (object.hasOwnProperty('tempo')) {
+        target.tempo = object.tempo;
+    }
+    if (object.hasOwnProperty('volume')) {
+        target.volume = object.volume;
+    }
+    if (object.hasOwnProperty('videoTransparency')) {
+        target.videoTransparency = object.videoTransparency;
+    }
+    if (object.hasOwnProperty('videoState')) {
+        target.videoState = object.videoState;
+    }
     if (object.hasOwnProperty('variables')) {
-        for (const j in object.variables) {
-            const variable = object.variables[j];
+        for (const varId in object.variables) {
+            const variable = object.variables[varId];
             const newVariable = new Variable(
-                variable.id,
-                variable.name,
-                variable.type,
-                variable.isPersistent
+                varId, // var id is the index of the variable desc array in the variables obj
+                variable[0], // name of the variable
+                Variable.SCALAR_TYPE, // type of the variable
+                (variable.length === 3) ? variable[2] : false // isPersistent/isCloud
             );
-            newVariable.value = variable.value;
+            newVariable.value = variable[1];
             target.variables[newVariable.id] = newVariable;
         }
     }
+    if (object.hasOwnProperty('lists')) {
+        for (const listId in object.lists) {
+            const list = object.lists[listId];
+            const newList = new Variable(
+                listId,
+                list[0],
+                Variable.LIST_TYPE,
+                false
+            );
+            newList.value = list[1];
+            target.variables[newList.id] = newList;
+        }
+    }
+    if (object.hasOwnProperty('broadcasts')) {
+        for (const broadcastId in object.broadcasts) {
+            const broadcast = object.broadcasts[broadcastId];
+            const newBroadcast = new Variable(
+                broadcastId,
+                broadcast,
+                Variable.BROADCAST_MESSAGE_TYPE,
+                false
+            );
+            // no need to explicitly set the value, variable constructor
+            // sets the value to the same as the name for broadcast msgs
+            target.variables[newBroadcast.id] = newBroadcast;
+        }
+    }
     if (object.hasOwnProperty('x')) {
         target.x = object.x;
     }
@@ -285,7 +851,6 @@ const parseScratchObject = function (object, runtime, extensions, zip) {
 
 /**
  * Deserialize the specified representation of a VM runtime and loads it into the provided runtime instance.
- * TODO: parse extension info (also, design extension info storage...)
  * @param  {object} json - JSON representation of a VM runtime.
  * @param  {Runtime} runtime - Runtime instance
  * @param {JSZip} zip - Sb3 file describing this project (to load assets from)

src/sprites/rendered-target.js

@@ -215,9 +215,9 @@ class RenderedTarget extends Target {
      */
     static get VIDEO_STATE () {
         return {
-            'OFF': 'off',
-            'ON': 'on',
-            'ON-FLIPPED': 'on-flipped'
+            OFF: 'off',
+            ON: 'on',
+            ON_FLIPPED: 'on-flipped'
         };
     }
 
@@ -1007,7 +1007,12 @@ class RenderedTarget extends Target {
             variables: this.variables,
             lists: this.lists,
             costumes: costumes,
-            sounds: this.getSounds()
+            sounds: this.getSounds(),
+            tempo: this.tempo,
+            volume: this.volume,
+            videoTransparency: this.videoTransparency,
+            videoState: this.videoState
+
         };
     }
 

src/virtual-machine.js

@@ -257,7 +257,6 @@ class VirtualMachine extends EventEmitter {
         const zip = new JSZip();
 
         // Put everything in a zip file
-        // TODO compression?
         zip.file('project.json', projectJson);
         for (let i = 0; i < soundDescs.length; i++) {
             const currSound = soundDescs[i];
@@ -268,7 +267,13 @@ class VirtualMachine extends EventEmitter {
             zip.file(currCostume.fileName, currCostume.fileContent);
         }
 
-        return zip.generateAsync({type: 'blob'});
+        return zip.generateAsync({
+            type: 'blob',
+            compression: 'DEFLATE',
+            compressionOptions: {
+                level: 6 // Tradeoff between best speed (1) and best compression (9)
+            }
+        });
     }
 
     /**

test/unit/serialization_sb3.js

@@ -1,11 +1,13 @@
 const test = require('tap').test;
+const path = require('path');
 const VirtualMachine = require('../../src/index');
 const sb3 = require('../../src/serialization/sb3');
-const demoSb3 = require('../fixtures/demo.json');
+const readFileToBuffer = require('../fixtures/readProjectFile').readFileToBuffer;
+const projectPath = path.resolve(__dirname, '../fixtures/clone-cleanup.sb2');
 
 test('serialize', t => {
     const vm = new VirtualMachine();
-    vm.loadProject(JSON.stringify(demoSb3))
+    vm.loadProject(readFileToBuffer(projectPath))
         .then(() => {
             const result = sb3.serialize(vm.runtime);
             // @todo Analyze