Bug 1940906 - Register a speech-recognition nsIMLModelResolver. r=media-playback-reviewers,alwu

SpeechModelResolver expands a model id (from LanguagesToSpeechModelId) to its
concrete ModelHub artifact for HWInferenceParent's generic model-resolution
lookup (@mozilla.org/ml/model-resolver;1?task=speech-recognition). Needed
starting with the next patch's RecvIsModelInstalled, which resolves ids
through that lookup and does nothing useful without a registered resolver.

Differential Revision: https://phabricator.services.mozilla.com/D315355

Author: padenot

dom/media/webspeech/recognition/SpeechModelResolver.cpp

@@ -0,0 +1,116 @@
+/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
+/* vim:set ts=2 sw=2 sts=2 et cindent: */
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+#include "SpeechModelResolver.h"
+
+#include <functional>
+
+#include "SpeechRecognitionModelMapping.h"
+#include "js/Conversions.h"
+#include "mozilla/Logging.h"
+#include "mozilla/dom/Promise-inl.h"
+#include "mozilla/dom/Promise.h"
+#include "mozilla/dom/WindowGlobalParent.h"
+#include "nsIMLModelHub.h"
+#include "nsServiceManagerUtils.h"
+
+namespace mozilla::dom {
+
+static LazyLogModule gSpeechModelDownloadLog("SpeechModelDownload");
+#define LOGD(...) \
+  MOZ_LOG_FMT(gSpeechModelDownloadLog, LogLevel::Debug, __VA_ARGS__)
+#define LOGE(...) \
+  MOZ_LOG_FMT(gSpeechModelDownloadLog, LogLevel::Error, __VA_ARGS__)
+
+NS_IMPL_ISUPPORTS(SpeechModelResolver, nsIMLModelResolver)
+
+NS_IMETHODIMP
+SpeechModelResolver::Resolve(const nsACString& aId, nsACString& aEngine,
+                             nsACString& aModel, nsACString& aRevision,
+                             nsACString& aFilename) {
+  SpeechModelIdentifier identifier;
+  if (!ResolveSpeechModelId(aId, identifier)) {
+    return NS_ERROR_NOT_AVAILABLE;
+  }
+  aEngine.Assign(kSpeechRecognitionEngineId);
+  aModel.Assign(identifier.mModelName);
+  aRevision.Assign(identifier.mRevision);
+  aFilename.Assign(identifier.mFileName);
+  return NS_OK;
+}
+
+// Runs aCallback with whether aModel is already downloaded to the local cache.
+// Must run on the parent main thread. Never rejects (false on any failure).
+// The testing mock is not consulted here: it lives in HWInferenceParent, which
+// short-circuits the already-installed case before this is reached.
+static void CheckInstalled(const SpeechModelIdentifier& aModel,
+                           std::function<void(bool)>&& aCallback) {
+  nsCOMPtr<nsIMLModelHub> hub = do_GetService("@mozilla.org/ml-modelhub;1");
+  if (!hub) {
+    LOGE("CheckInstalled - failed to get ModelHub service");
+    aCallback(false);
+    return;
+  }
+
+  RefPtr<Promise> promise;
+  nsresult rv = hub->IsModelInstalled(
+      kSpeechRecognitionEngineId, aModel.mModelName, aModel.mRevision,
+      aModel.mFileName, getter_AddRefs(promise));
+  if (NS_FAILED(rv) || !promise) {
+    LOGE("CheckInstalled - IsModelInstalled call failed rv={:x}",
+         static_cast<uint32_t>(rv));
+    aCallback(false);
+    return;
+  }
+
+  promise->AddCallbacksWithCycleCollectedArgs(
+      [aCallback](JSContext*, JS::Handle<JS::Value> aValue, ErrorResult&) {
+        aCallback(JS::ToBoolean(aValue));
+      },
+      [aCallback](JSContext*, JS::Handle<JS::Value>, ErrorResult&) {
+        aCallback(false);
+      });
+}
+
+NS_IMETHODIMP
+SpeechModelResolver::AuthorizeDownload(
+    const nsACString& aModel, const nsACString& aRevision,
+    const nsACString& aFilename, WindowGlobalParent* aWindow,
+    const nsAString& aProgressToken,
+    nsIMLModelDownloadAuthorizationCallback* aCallback) {
+  MOZ_ASSERT(NS_IsMainThread());
+  MOZ_ASSERT(aCallback);
+
+  // Speech recognition is only ever driven by content, so a request without a
+  // document to prompt for has nobody to ask.
+  if (!aWindow) {
+    LOGE("AuthorizeDownload - no requesting window");
+    aCallback->Resolve(false);
+    return NS_OK;
+  }
+
+  SpeechModelIdentifier model{nsCString(aModel), nsCString(aFilename),
+                              nsCString(aRevision),
+                              SpeechModelSizeMB(aModel, aRevision, aFilename)};
+  LOGD("AuthorizeDownload - model={} sizeMB={}", model.ToString().get(),
+       model.mSizeMB);
+
+  nsCOMPtr<nsIMLModelDownloadAuthorizationCallback> callback = aCallback;
+  std::function<void(bool)> resolve = [callback](bool aAllow) {
+    callback->Resolve(aAllow);
+  };
+
+  // Authorize unconditionally for now; asking the user before an actual
+  // download arrives with the permission prompt.
+  CheckInstalled(
+      model, [resolve = std::move(resolve)](bool) mutable { resolve(true); });
+  return NS_OK;
+}
+
+#undef LOGD
+#undef LOGE
+
+}  // namespace mozilla::dom

dom/media/webspeech/recognition/SpeechModelResolver.h

@@ -0,0 +1,40 @@
+/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
+/* vim:set ts=2 sw=2 sts=2 et cindent: */
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+#ifndef DOM_MEDIA_WEBSPEECH_RECOGNITION_SPEECHMODELRESOLVER_H_
+#define DOM_MEDIA_WEBSPEECH_RECOGNITION_SPEECHMODELRESOLVER_H_
+
+#include "nsIMLModelResolver.h"
+
+namespace mozilla::dom {
+
+// The speech-recognition implementation of the generic HWInference model
+// resolver (registered for the "speech-recognition" task). Lives in the
+// trusted parent (main) process.
+//
+// Resolve() expands an id from the in-tree model table (see
+// SpeechRecognitionModelMapping) to its concrete ModelHub artifact; it throws
+// for any other id. This is what keeps the (less trusted) utility process from
+// ever naming a model/revision/filename itself.
+//
+// AuthorizeDownload() skips the prompt when the model is already installed,
+// honors the testing prefs, and otherwise shows the model-download permission
+// doorhanger, allowing the download only on a real user Allow. Content can
+// only ask; it never authorizes a download.
+class SpeechModelResolver final : public nsIMLModelResolver {
+ public:
+  NS_DECL_ISUPPORTS
+  NS_DECL_NSIMLMODELRESOLVER
+
+  SpeechModelResolver() = default;
+
+ private:
+  ~SpeechModelResolver() = default;
+};
+
+}  // namespace mozilla::dom
+
+#endif  // DOM_MEDIA_WEBSPEECH_RECOGNITION_SPEECHMODELRESOLVER_H_

dom/media/webspeech/recognition/components.conf

@@ -0,0 +1,19 @@
+# -*- Mode: python; indent-tabs-mode: nil; tab-width: 4 -*-
+# vim: set filetype=python:
+# This Source Code Form is subject to the terms of the Mozilla Public
+# License, v. 2.0. If a copy of the MPL was not distributed with this
+# file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+Classes = [
+    {
+        "cid": "{f330141f-9b06-4ebd-baa4-df932e9eb78e}",
+        "contract_ids": [
+            "@mozilla.org/ml/model-resolver;1?task=speech-recognition"
+        ],
+        "type": "mozilla::dom::SpeechModelResolver",
+        "headers": [
+            "/dom/media/webspeech/recognition/SpeechModelResolver.h"
+        ],
+        "processes": ProcessSelector.MAIN_PROCESS_ONLY,
+    },
+]

dom/media/webspeech/recognition/moz.build

@@ -43,6 +43,7 @@ EXPORTS.mozilla.hwinference += [
 UNIFIED_SOURCES += [
     "SpeechGrammar.cpp",
     "SpeechGrammarList.cpp",
+    "SpeechModelResolver.cpp",
     "SpeechRecognition.cpp",
     "SpeechRecognitionAlternative.cpp",
     "SpeechRecognitionBackend.cpp",
@@ -59,6 +60,10 @@ IPDL_SOURCES = [
     "PSpeechRecognition.ipdl",
 ]
 
+XPCOM_MANIFESTS += [
+    "components.conf",
+]
+
 # Bug 1244242 is about enabling on Android
 if CONFIG["MOZ_WIDGET_TOOLKIT"] != "android":
     DIRS += ["/third_party/parakeet.cpp"]