Add bounded PRIMARY selection publication (#657) (#658)

Author: wieslawsoltes

docs/validation/code-oss-web-api-ledger.md

@@ -18,7 +18,7 @@ Static matches establish source reachability only. Runtime claims require the ev
 | `broadcast-channel` | yes | blocked | linux-x64, macos-arm64, windows-x64 | WebScene#81 | 0 | Cross-realm ordering, clone, teardown, queue, and product evidence are incomplete. |
 | `cache-storage` | yes | blocked | linux-x64, macos-arm64, windows-x64 | WebScene#266 | 0 | CacheStorage is required by webview resources and remains scheduled in the webview stack. |
 | `canvas-2d` | yes | blocked | linux-x64, macos-arm64, windows-x64 | WebScene#227 | 0 | Broad engine coverage exists, but the accepted Code OSS visual scenarios have not yet supplied revision-pinned product evidence here. |
-| `clipboard` | yes | blocked | linux-x64, macos-arm64, windows-x64 | WebScene#227 | 0 | The host ABI exists; editor, terminal, webview, denial, focus, and large-payload product qualification is incomplete. |
+| `clipboard` | yes | blocked | linux-x64, macos-arm64, windows-x64 | WebScene#227, WebScene#657 | 0 | The bounded host ABI and focused input/textarea PRIMARY publication source contract exist. AppScene#275 integration plus editor, terminal, webview, denial, focus, large-payload and packaged-product qualification remain incomplete. |
 | `crypto` | yes | blocked | linux-x64, macos-arm64, windows-x64 | WebScene#102 | 0 | The vetted-provider runtime surface is complete; configured unchanged-package MCP secret-input and browser connection-secret persistence/reopen evidence remains missing across the release RIDs. Stock Code OSS omits proprietary VSDA assets, and bundled Copilot HMAC executes in its Node extension host. |
 | `dom-core` | yes | blocked | linux-x64, macos-arm64, windows-x64 | WebScene#227, WebScene#260 | 0 | DOM coverage is extensive but no exhaustive reachable-surface and accepted-scenario reconciliation has been promoted. |
 | `drag-drop` | yes | blocked | linux-x64, macos-arm64, windows-x64 | WebScene#260 | 0 | Explorer/editor drag data, cancellation, focus, native input, and teardown evidence is incomplete. |

docs/validation/primary-selection-publication-657.md

@@ -0,0 +1,75 @@
+# Bounded PRIMARY selected-text publication source contract
+
+WebScene issue #657 adds the provider half of automatic Linux PRIMARY
+publication. The exact baseline is WebScene
+`dabd1435c8e6415163067ad7ad5befb498a77126`; AppScene #275 is the dependent
+native consumer. The provider is a product-neutral native-host signal and has
+no X11, browser-process, Electron, CEF, WebView or VS Code dependency.
+
+## Provider boundary
+
+A settled nonempty selection in the currently focused, connected, writable
+`input` or `textarea` emits the existing `webscene_host_request_v1` with this
+exact shape:
+
+- kind `WEBSCENE_HOST_REQUEST_CLIPBOARD_WRITE_V1`;
+- request ID and target node ID zero;
+- flags `WEBSCENE_HOST_REQUEST_CLIPBOARD_REPLACE_V1 |
+  WEBSCENE_HOST_REQUEST_CLIPBOARD_PRIMARY_V1`;
+- content type `text/plain`, no URL, and valid UTF-8 selected bytes;
+- a maximum payload of 64 KiB, with oversize selections refused rather than
+  truncated.
+
+The zero request ID makes the update one-way. It creates no JavaScript API,
+Promise, completion binding or user-activation requirement. Ordinary Clipboard
+API and shortcut reads/writes retain their nonzero request IDs, existing flags,
+16-operation completion cap and CLIPBOARD meaning.
+
+Password, readonly, disabled, inert, disconnected and unfocused controls are
+ineligible. A collapsed or empty selection, an invalid UTF-8 slice, a UTF-16
+boundary inside a surrogate pair, and an oversize slice emit nothing. These
+cases do not clear a previously published or pending PRIMARY value.
+
+## Coalescing and lifecycle
+
+Selection and value mutations mark only the active text control dirty. The
+runtime extracts its final selected slice after the current script, event and
+microtask work reaches a stable checkpoint. Extraction checks byte offsets and
+the 64 KiB cap before allocating the request payload.
+
+The mutex-protected typed-host FIFO recognizes only the exact zero-ID PRIMARY
+publication shape. If one is already pending, the runtime replaces that lease
+in place with the newest eligible payload. Otherwise it appends one only while
+the aggregate 1,024-request queue has capacity. Publication therefore uses at
+most one pending slot and cannot evict or consume capacity reserved by an
+existing completion-bearing request. Saturation drops the update without
+changing clipboard state or exposing selected bytes in diagnostics.
+
+Top-level navigation discards every publication still in WebScene's queue and
+clears its dirty observation. A queued child-frame publication carries a
+private frame-owner identity, outside the public ABI, so frame navigation or
+detach can remove it before retiring that realm. Engine teardown destroys the
+same bounded state. Once a host has taken a stable one-way lease, it represents
+the prior selection and may remain PRIMARY across document navigation,
+matching desktop selection ownership. Collapse, blur and teardown never send
+a clearing write.
+
+## Authored contracts and remaining acceptance
+
+`test_primary_selection_publication` covers the exact ABI, UTF-8 text,
+same-task and pending-request coalescing, retention on collapse, ineligible
+controls, detach, refusal above 64 KiB and admission at the cap. Existing
+clipboard contracts continue to cover ordinary Clipboard API flags, request
+IDs, completion and the PRIMARY middle-paste read path.
+
+Unchanged Code OSS mirrors short Monaco selections into its writable hidden
+textarea, so those exact selections reach this provider without a source
+patch. Monaco can elide large or multipage screen-reader textarea content and
+does not represent all multicursor model selections there. Exact large,
+multipage and multicursor Monaco publication, readonly controls,
+contenteditable and document `Selection` remain outside #657.
+
+Only `git diff --check` is run for this fast implementation slice. The authored
+native contract, compiler/build, browser-oracle comparison, 10,000-change CPU
+and allocation measurements, AppScene #275 integration, physical X11 clients,
+packaged unchanged Code OSS, VM and computer-use gates remain unexecuted.

experiments/WebScene.NativeEngine.Probe/native/webscene_native_engine.h

@@ -2689,8 +2689,9 @@ WEBSCENE_API void webscene_file_grant_release_request_release_v2(
     const webscene_file_grant_release_request_v2* request);
 
 /* Typed native desktop request ABI. Request memory is immutable and remains
- * valid until release. Byte payloads are capped at 16 MiB, strings are UTF-8,
- * and at most 16 completion-bearing operations may be pending per document. */
+ * valid until release. Byte payloads are capped at 16 MiB, with one-way
+ * selected-text publication capped at 64 KiB. Strings are UTF-8, and at most
+ * 16 completion-bearing operations may be pending per document. */
 enum {
     WEBSCENE_HOST_REQUEST_OPEN_EXTERNAL_URL_V1 = 1,
     WEBSCENE_HOST_REQUEST_CLIPBOARD_READ_V1 = 2,
@@ -2704,9 +2705,11 @@ enum {
 };
 enum {
     WEBSCENE_HOST_REQUEST_CLIPBOARD_REPLACE_V1 = 1U << 0U,
-    /* Selects the platform's primary/selection clipboard for a text/plain
-     * READ request. Ordinary Clipboard API and shortcut requests leave this
-     * clear and continue to address the normal clipboard. */
+    /* Selects the platform's primary/selection clipboard. It is valid either
+     * alone for a completion-bearing text/plain READ request or together with
+     * REPLACE for the engine's zero-ID, one-way selected-text WRITE. Ordinary
+     * Clipboard API and shortcut requests leave this clear and continue to
+     * address the normal clipboard. */
     WEBSCENE_HOST_REQUEST_CLIPBOARD_PRIMARY_V1 = 1U << 1U,
     WEBSCENE_HOST_REQUEST_NAVIGATION_REPLACE_V1 = 1U << 0U,
     WEBSCENE_HOST_REQUEST_EXTERNAL_NEW_CONTEXT_V1 = 1U << 0U,

experiments/WebScene.NativeEngine.Probe/native/webscene_v8_runtime.cpp

@@ -6526,6 +6526,126 @@ struct v8_dom_runtime::implementation final {
         return true;
     }
 
+    bool enqueue_primary_selection_publication(
+        std::unique_ptr<native_host_request> request)
+    {
+        if (!request) return false;
+        auto notify = false;
+        {
+            std::lock_guard lock(host_request_mutex);
+            const auto pending = std::find_if(
+                typed_host_requests.begin(),
+                typed_host_requests.end(),
+                [](const auto& candidate) {
+                    return candidate != nullptr
+                        && candidate->view.request_id == 0U
+                        && candidate->view.kind
+                            == WEBSCENE_HOST_REQUEST_CLIPBOARD_WRITE_V1
+                        && candidate->view.flags
+                            == (WEBSCENE_HOST_REQUEST_CLIPBOARD_REPLACE_V1
+                                | WEBSCENE_HOST_REQUEST_CLIPBOARD_PRIMARY_V1);
+                });
+            if (pending != typed_host_requests.end()) {
+                *pending = std::move(request);
+                return true;
+            }
+            if (host_requests.size() + typed_host_requests.size()
+                    + reserved_host_request_count
+                >= maximum_host_request_count) return false;
+            typed_host_requests.push_back(std::move(request));
+            notify = true;
+        }
+        if (notify && host_request_available) host_request_available();
+        return true;
+    }
+
+    void discard_primary_selection_publication_for_frame(uint32_t frame_id)
+    {
+        if (frame_id == 0U) return;
+        std::lock_guard lock(host_request_mutex);
+        std::erase_if(typed_host_requests, [&](const auto& request) {
+            return request != nullptr
+                && request->source_frame_owner_id == frame_id
+                && request->view.request_id == 0U
+                && request->view.kind
+                    == WEBSCENE_HOST_REQUEST_CLIPBOARD_WRITE_V1
+                && request->view.flags
+                    == (WEBSCENE_HOST_REQUEST_CLIPBOARD_REPLACE_V1
+                        | WEBSCENE_HOST_REQUEST_CLIPBOARD_PRIMARY_V1);
+        });
+    }
+
+    void mark_primary_selection_publication_dirty(
+        const dom_node* candidate) noexcept
+    {
+        if (candidate != nullptr && candidate == active_element)
+            primary_selection_publication_dirty = true;
+    }
+
+    void flush_primary_selection_publication()
+    {
+        if (!primary_selection_publication_dirty) return;
+        primary_selection_publication_dirty = false;
+        constexpr size_t maximum_primary_selection_bytes = 64U * 1024U;
+        auto* selected = active_element;
+        if (!is_writable_text_control(selected)
+            || !is_connected(*selected)
+            || !selected->form_control().input_focused
+            || forms::input_type_is(*selected, "password")
+            || focus_document_owner(selected) != focused_document_owner_id) {
+            return;
+        }
+        ensure_form_value(*selected);
+        const auto& control = selected->form_control();
+        const auto start = std::min(control.selection_start, control.value.size());
+        const auto end = std::min(control.selection_end, control.value.size());
+        if (start >= end || control.selection_start_utf16_suboffset != 0U
+            || control.selection_end_utf16_suboffset != 0U
+            || end - start > maximum_primary_selection_bytes) {
+            return;
+        }
+        const auto text = std::string_view(control.value).substr(start, end - start);
+        const auto valid_utf8 = [](std::string_view value) noexcept {
+            for (size_t offset = 0U; offset < value.size();) {
+                const auto first = static_cast<unsigned char>(value[offset]);
+                size_t length{};
+                uint32_t scalar{};
+                if (first <= 0x7fU) { length = 1U; scalar = first; }
+                else if (first >= 0xc2U && first <= 0xdfU) {
+                    length = 2U; scalar = first & 0x1fU;
+                } else if (first >= 0xe0U && first <= 0xefU) {
+                    length = 3U; scalar = first & 0x0fU;
+                } else if (first >= 0xf0U && first <= 0xf4U) {
+                    length = 4U; scalar = first & 0x07U;
+                } else return false;
+                if (offset + length > value.size()) return false;
+                for (size_t index = 1U; index < length; ++index) {
+                    const auto continuation =
+                        static_cast<unsigned char>(value[offset + index]);
+                    if ((continuation & 0xc0U) != 0x80U) return false;
+                    scalar = (scalar << 6U) | (continuation & 0x3fU);
+                }
+                if ((length == 2U && scalar < 0x80U)
+                    || (length == 3U && scalar < 0x800U)
+                    || (length == 4U && scalar < 0x10000U)
+                    || (scalar >= 0xd800U && scalar <= 0xdfffU)
+                    || scalar > 0x10ffffU) return false;
+                offset += length;
+            }
+            return true;
+        };
+        if (!valid_utf8(text)) return;
+        auto request = std::make_unique<native_host_request>();
+        request->view.kind = WEBSCENE_HOST_REQUEST_CLIPBOARD_WRITE_V1;
+        request->view.flags = WEBSCENE_HOST_REQUEST_CLIPBOARD_REPLACE_V1
+            | WEBSCENE_HOST_REQUEST_CLIPBOARD_PRIMARY_V1;
+        request->source_frame_owner_id = focus_document_owner(selected);
+        request->content_type = "text/plain";
+        request->bytes.assign(text.begin(), text.end());
+        static_cast<void>(
+            enqueue_primary_selection_publication(std::move(request)));
+    }
+
     bool enqueue_reserved_typed_host_request(
         std::unique_ptr<native_host_request> request)
     {
@@ -7481,14 +7601,16 @@ uint64_t v8_dom_runtime::last_resize_observers_nanoseconds() const noexcept
 
 bool v8_dom_runtime::dispatch_input(const webscene_input_event& event, bool defer_cursor_update)
 {
-    return impl_->dispatch_input(event, defer_cursor_update)
-        && impl_->promote_pending_promise_error();
+    const auto dispatched = impl_->dispatch_input(event, defer_cursor_update);
+    impl_->flush_primary_selection_publication();
+    return dispatched && impl_->promote_pending_promise_error();
 }
 
 bool v8_dom_runtime::dispatch_drag(native_drag_event& event)
 {
-    return impl_->dispatch_drag(event)
-        && impl_->promote_pending_promise_error();
+    const auto dispatched = impl_->dispatch_drag(event);
+    impl_->flush_primary_selection_publication();
+    return dispatched && impl_->promote_pending_promise_error();
 }
 
 void v8_dom_runtime::refresh_pointer_cursor_after_layout()

experiments/WebScene.NativeEngine.Probe/native/webscene_v8_runtime.h

@@ -55,6 +55,7 @@ struct native_host_completion {
 };
 struct native_host_request {
     webscene_host_request_v1 view{};
+    uint32_t source_frame_owner_id{};
     std::string content_type;
     std::vector<uint8_t> bytes;
     std::string url;

experiments/WebScene.NativeEngine.Probe/native/webscene_v8_runtime_css_cascade.inc

@@ -2409,6 +2409,8 @@
         std::optional<bool> user_validity_interacted = std::nullopt,
         std::optional<bool> value_changed_by_user = std::nullopt)
     {
+        if (previous_value != current_value)
+            mark_primary_selection_publication_dirty(&subject);
         document.mark_dirty();
         const auto empty_state_changed=previous_value.empty()!=current_value.empty();
         const auto previous_range=forms::range_state(subject,previous_value);

experiments/WebScene.NativeEngine.Probe/native/webscene_v8_runtime_dom_core.inc

@@ -2825,6 +2825,7 @@
                     || (target != nullptr && subtree_contains(frame, target));
             });
         }
+        discard_primary_selection_publication_for_frame(frame.id);
         {
             std::lock_guard lock(document_generation_mutex);
             ++frame_navigation_generations.try_emplace(frame.id, 0U)
@@ -2894,6 +2895,7 @@
 
     void retire_iframe_navigation(uint32_t frame_id)
     {
+        discard_primary_selection_publication_for_frame(frame_id);
         {
             std::lock_guard lock(document_generation_mutex);
             frame_navigation_generations.erase(frame_id);
@@ -5404,6 +5406,7 @@
         control.selection_end_utf16_suboffset = end_offset.utf16_suboffset;
         control.selection_direction = text_selection_direction::none;
         control.selection_explicitly_set = true;
+        self->mark_primary_selection_publication_dirty(node);
         if(self->is_connected(*node)) {
             self->recascade_form_value_selector_transition(
                 *node,previous_value,control.value,std::nullopt,false);

experiments/WebScene.NativeEngine.Probe/native/webscene_v8_runtime_dom_properties.inc

@@ -3883,6 +3883,7 @@
         v8::Local<v8::Value> value,
         const v8::PropertyCallbackInfo<void>& info)
     {
+        auto* self = current(info.GetIsolate());
         auto* node = unwrap_node(info.Holder());
         if (node == nullptr) return;
         const auto index = value->Int32Value(info.GetIsolate()->GetCurrentContext()).FromMaybe(0);
@@ -3908,6 +3909,7 @@
              node->mutable_form_control().selection_end_utf16_suboffset})) {
             node->mutable_form_control().selection_direction = text_selection_direction::none;
         }
+        self->mark_primary_selection_publication_dirty(node);
     }
 
     static void get_selection_end(
@@ -3930,6 +3932,7 @@
         v8::Local<v8::Value> value,
         const v8::PropertyCallbackInfo<void>& info)
     {
+        auto* self = current(info.GetIsolate());
         auto* node = unwrap_node(info.Holder());
         if (node == nullptr) return;
         const auto index = value->Int32Value(info.GetIsolate()->GetCurrentContext()).FromMaybe(0);
@@ -3955,6 +3958,7 @@
              node->mutable_form_control().selection_end_utf16_suboffset})) {
             node->mutable_form_control().selection_direction = text_selection_direction::none;
         }
+        self->mark_primary_selection_publication_dirty(node);
     }
 
     static void get_selection_direction(
@@ -3979,6 +3983,7 @@
         v8::Local<v8::Value> value,
         const v8::PropertyCallbackInfo<void>& info)
     {
+        auto* self = current(info.GetIsolate());
         auto* node = unwrap_node(info.Holder());
         if (node == nullptr) return;
         const auto direction = to_utf8(info.GetIsolate(), value);
@@ -3995,6 +4000,7 @@
              node->mutable_form_control().selection_end_utf16_suboffset})) {
             node->mutable_form_control().selection_direction = text_selection_direction::none;
         }
+        self->mark_primary_selection_publication_dirty(node);
     }
 
     static void get_boolean_attribute(

experiments/WebScene.NativeEngine.Probe/native/webscene_v8_runtime_navigation.inc

@@ -143,7 +143,11 @@
         fullscreen_element = nullptr;
         document_fullscreen = false;
         { std::lock_guard lock(host_request_mutex); host_requests.clear(); }
-        { std::lock_guard lock(host_request_mutex); typed_host_requests.clear(); }
+        {
+            std::lock_guard lock(host_request_mutex);
+            typed_host_requests.clear();
+            primary_selection_publication_dirty = false;
+        }
         { std::lock_guard lock(host_request_mutex); download_requests.clear(); }
         { std::lock_guard lock(host_request_mutex); outbound_drag_requests.clear(); }
         object_url_file_data.clear();

experiments/WebScene.NativeEngine.Probe/native/webscene_v8_runtime_semantics.inc

@@ -1110,6 +1110,7 @@
                 node->mutable_form_control().selection_direction =
                     text_selection_direction::none;
                 node->mutable_form_control().selection_explicitly_set = true;
+                mark_primary_selection_publication_dirty(node);
                 document.mark_dirty();
                 break;
             }