Do not export 'CompositeArrayBuffer' from tfjs-core

Author: mattsoulanille

tfjs-converter/src/executor/graph_model_test.ts

@@ -586,8 +586,8 @@ describe('Model', () => {
         expect(handler.savedArtifacts.modelTopology).toEqual(CUSTOM_OP_MODEL);
         expect(handler.savedArtifacts.weightSpecs).toEqual(weightsManifest);
         tfc.test_util.expectArraysClose(
-            new Int32Array(new io.CompositeArrayBuffer(
-                handler.savedArtifacts.weightData).slice()), bias.dataSync());
+            new Int32Array(io.concatenateArrayBuffers(
+                handler.savedArtifacts.weightData)), bias.dataSync());
       });
     });
   });
@@ -617,8 +617,8 @@ describe('Model', () => {
         });
         expect(handler.savedArtifacts.weightSpecs).toEqual(weightsManifest);
         tfc.test_util.expectArraysClose(
-          new Int32Array(new io.CompositeArrayBuffer(
-            handler.savedArtifacts.weightData).slice()), bias.dataSync());
+          new Int32Array(io.concatenateArrayBuffers(
+            handler.savedArtifacts.weightData)), bias.dataSync());
       });
     });
 
@@ -906,8 +906,8 @@ describe('Model', () => {
         });
         expect(handler.savedArtifacts.weightSpecs).toEqual(weightsManifest);
         tfc.test_util.expectArraysClose(
-          new Int32Array(new io.CompositeArrayBuffer(handler.savedArtifacts
-            .weightData).slice()), bias.dataSync());
+          new Int32Array(io.concatenateArrayBuffers(handler.savedArtifacts
+            .weightData)), bias.dataSync());
       });
     });
 

tfjs-converter/src/operations/executors/spy_ops.ts

@@ -15,23 +15,11 @@
  * =============================================================================
  */
 
-// The opposite of Extract<T, U>
-type Without<T, U> = T extends U ? never : T;
-
-// Do not spy on CompositeArrayBuffer because it is a class constructor.
-type NotSpiedOn = 'CompositeArrayBuffer';
-
 export type RecursiveSpy<T> =
-  T extends Function ? jasmine.Spy :
-  {[K in Without<keyof T, NotSpiedOn>]: RecursiveSpy<T[K]>} &
-  {[K in Extract<keyof T, NotSpiedOn>]: T[K]};
+    T extends Function ? jasmine.Spy : {[K in keyof T]: RecursiveSpy<T[K]>};
 
 export function spyOnAllFunctions<T>(obj: T): RecursiveSpy<T> {
   return Object.fromEntries(Object.entries(obj).map(([key, val]) => {
-    // TODO(mattSoulanille): Do not hard code this
-    if (key === 'CompositeArrayBuffer') {
-      return val;
-    }
     if (val instanceof Function) {
       return [key, jasmine.createSpy(`${key} spy`, val).and.callThrough()];
     } else if (val instanceof Array) {

tfjs-core/src/io/io.ts

@@ -27,13 +27,11 @@ import {fromMemory, fromMemorySync, withSaveHandler, withSaveHandlerSync} from '
 import {getLoadHandlers, getSaveHandlers, registerLoadRouter, registerSaveRouter} from './router_registry';
 import {IOHandler, IOHandlerSync, LoadHandler, LoadOptions, ModelArtifacts, ModelArtifactsInfo, ModelJSON, ModelStoreManager, OnProgressCallback, RequestDetails, SaveConfig, SaveHandler, SaveResult, TrainingConfig, WeightGroup, WeightsManifestConfig, WeightsManifestEntry, WeightData} from './types';
 import {loadWeights, weightsLoaderFactory} from './weights_loader';
-import {CompositeArrayBuffer} from './composite_array_buffer';
 
 export {copyModel, listModels, moveModel, removeModel} from './model_management';
 export {
   browserFiles,
   browserHTTPRequest,
-  CompositeArrayBuffer,
   concatenateArrayBuffers,
   decodeWeights,
   encodeWeights,

tfjs-layers/src/models_test.ts

@@ -1293,7 +1293,7 @@ describeMathCPUAndWebGL2('Saving+loading model with optimizer', () => {
     // The second part comes from the bias of the dense layer, which has 1
     // element and is also 4 bytes.
     const weightData = savedArtifacts.weightData;
-    expect(new io.CompositeArrayBuffer(weightData).byteLength)
+    expect(io.concatenateArrayBuffers(weightData).byteLength)
       .toEqual(4 * 8 + 4 * 1 + 4);
 
     // Load the model back, with the optimizer.
@@ -1353,7 +1353,7 @@ describeMathCPUAndWebGL2('Saving+loading model with optimizer', () => {
     // The second part comes from the bias of the dense layer, which has 1
     // element and is also 4 bytes.
     const weightData = savedArtifacts.weightData;
-    expect(new io.CompositeArrayBuffer(weightData)
+    expect(io.concatenateArrayBuffers(weightData)
       .byteLength).toEqual(4 + 4 * 8 * 3 + 4 * 1 * 3);
 
     // Load the model back, with the optimizer.
@@ -1413,7 +1413,7 @@ describeMathCPUAndWebGL2('Saving+loading model with optimizer', () => {
     // The second part comes from the bias of the dense layer, which has 1
     // element and is also 4 bytes.
     const weightData = savedArtifacts.weightData;
-    expect(new io.CompositeArrayBuffer(weightData).byteLength)
+    expect(io.concatenateArrayBuffers(weightData).byteLength)
       .toEqual(4 + 4 * 8 * 3 + 4 * 1 * 3);
 
     // Load the model back, with the optimizer.
@@ -1471,7 +1471,7 @@ describeMathCPUAndWebGL2('Saving+loading model with optimizer', () => {
     // The second part comes from the bias of the dense layer, which has 1
     // element and is also 4 bytes.
     const weightData = savedArtifacts.weightData;
-    expect(new io.CompositeArrayBuffer(weightData).byteLength)
+    expect(io.concatenateArrayBuffers(weightData).byteLength)
       .toEqual(4 + 4 * 8 * 2 + 4 * 1 * 2);
 
     // Load the model back, with the optimizer.
@@ -1534,7 +1534,7 @@ describeMathCPUAndWebGL2('Saving+loading model with optimizer', () => {
     // The second part comes from the bias of the dense layer, which has 1
     // element and is also 4 bytes.
     const weightData = savedArtifacts.weightData;
-    expect(new io.CompositeArrayBuffer(weightData).byteLength)
+    expect(io.concatenateArrayBuffers(weightData).byteLength)
       .toEqual(4 + 4 * 8 * 3 + 4 * 1 * 3);
 
     // Load the model back, with the optimizer.
@@ -1593,7 +1593,7 @@ describeMathCPUAndWebGL2('Saving+loading model with optimizer', () => {
     // The second part comes from the bias of the dense layer, which has 1
     // element and is also 4 bytes.
     const weightData = savedArtifacts.weightData;
-    expect(new io.CompositeArrayBuffer(weightData).byteLength)
+    expect(io.concatenateArrayBuffers(weightData).byteLength)
       .toEqual(4 + 4 * 8 * 3 + 4 * 1 * 3);
 
     // Load the model back, with the optimizer.
@@ -1651,7 +1651,7 @@ describeMathCPUAndWebGL2('Saving+loading model with optimizer', () => {
     // The second part comes from the bias of the dense layer, which has 1
     // element and is also 4 bytes.
     const weightData = savedArtifacts.weightData;
-    expect(new io.CompositeArrayBuffer(weightData).byteLength)
+    expect(io.concatenateArrayBuffers(weightData).byteLength)
       .toEqual(4 + 4 * 8 * 2 + 4 * 1 * 2);
 
     // Load the model back, with the optimizer.