Aug 23, 2026 – Aug 29, 2026

Unify code for PodGroups and CompositePodGroups in workloadForestGo · 383 + / 709 −

Introduces 32 new declarations in pkg/scheduler/framework/types.go.

Adds new systems design rather than adjusting what was there. Tests changed with it, with code of their own.

pkg/scheduler/framework/types.go ↗ · 24 files

@@ -1161,73 +1064,129 @@ func findNodeAndParent(curr, parent *PodGroupInfo, name string) (*PodGroupInfo, // It returns a flat slice of all QueuedPodInfo elements that were successfully removed. func (pgqi *QueuedPodGroupInfo) deleteSubtreePods(curr *PodGroupInfo) []*QueuedPodInfo { 	removedPods := make([]*QueuedPodInfo, 0)-	if curr.GetPodGroup() != nil {-		key := fwk.PodGroupKey(curr.Namespace, curr.Name)+	if curr.GetType() == fwk.PodGroupKeyType {+		key := fwk.PodGroupKey(curr.GetNamespace(), curr.GetName()) 		if pods, ok := pgqi.QueuedPodInfos[key]; ok { 			removedPods = append(removedPods, pods...) 			delete(pgqi.QueuedPodInfos, key) 		}+		return removedPods 	} 	for _, child := range curr.Children { 		removedPods = append(removedPods, pgqi.deleteSubtreePods(child)...) 	} 	return removedPods } -func newQueuedPodGroupInfo(pg *schedulingv1beta1.PodGroup) *QueuedPodGroupInfo {-	return &QueuedPodGroupInfo{-		PodGroupInfo: &PodGroupInfo{-			Namespace: pg.Namespace,-			Name:      pg.Name,-			Type:      fwk.PodGroupKeyType,-			PodGroup:  pg,-			Children:  make([]*PodGroupInfo, 0),-		},-		QueuedPodInfos: make(map[fwk.EntityKey][]*QueuedPodInfo),-	}-}--// PodGroupInfo is a wrapper around the PodGroup API object together with a list of pods that belong to the pod group.

Add an opt-in per-frame pixel cap (cap_pixels_per_frame) to the Qwen3-VL video processorPython · 381 + / 63 −

Introduces 8 new declarations in src/transformers/models/cohere_compass/video_processing_cohere_compass.py.

Adds new core implementation rather than adjusting what was there. Tests changed with it, with code of their own.

src/transformers/models/cohere_compass/video_processing_cohere_compass.py ↗ · 8 files

@@ -296,12 +331,55 @@ def _preprocess(         processed_grids = reorder_videos(processed_grids, grouped_videos_index)         pixel_values_videos = torch.cat(processed_videos, dim=0)         video_grid_thw = torch.tensor(processed_grids)-        data = {-            "pixel_values_videos": pixel_values_videos,-            "video_grid_thw": video_grid_thw,-        } -        return BatchFeature(data=data, tensor_type=return_tensors)+        return BatchFeature(+            data={"pixel_values_videos": pixel_values_videos, "video_grid_thw": video_grid_thw},+            tensor_type=return_tensors,+        )++    def get_num_of_video_patches(self, num_frames: int, height: int, width: int, videos_kwargs=None):+        """+        A utility that returns number of video patches a given video size.++        Args:+            num_frames (`int`):+                Number of frames in the input video.+            height (`int`):+                Height of the input video.+            width (`int`):+                Width of the input video.+            videos_kwargs (`dict`, *optional*)+                Any kwargs to override defaults of the video processor.+        Returns:+            `Tuple(int, int)`: Number of placeholder tokens required and number of patches per image.+        """+        videos_kwargs = videos_kwargs if videos_kwargs is not None else {}+        min_pixels = videos_kwargs.get("min_pixels", None) or self.size["shortest_edge"]+        max_pixels = videos_kwargs.get("max_pixels", None) or self.size["longest_edge"]

fix(cli): don't duplicate passthrough args for deno deploy/sandboxRust · 79 + / 5 −

Introduces 4 new declarations in libs/cli_parser/src/tests_full.rs.

Adds new language design rather than adjusting what was there. Read the linked PR for the surrounding test context.

libs/cli_parser/src/tests_full.rs ↗ · 2 files

@@ -9767,6 +9767,79 @@ fn use_env_proxy_flags() {   assert!(r.is_err()); } +#[test]+fn deploy_subcommand() {+  let r = flags_from_vec(svec!["deno", "deploy"]);+  assert_eq!(+    r.unwrap(),+    Flags {+      subcommand: DenoSubcommand::Deploy(DeployFlags { sandbox: false }),+      ..Flags::default()+    }+  );++  // `deploy` is a passthrough subcommand: every arg after it is forwarded+  // verbatim, exactly once. Regression test for a duplication bug where the+  // args were written to `argv` both by `deploy_parse` and by the generic+  // trailing-arg handling, turning `--prod` into `--prod --prod`.+  let r = flags_from_vec(svec!["deno", "deploy", "--prod"]);+  assert_eq!(+    r.unwrap(),+    Flags {+      subcommand: DenoSubcommand::Deploy(DeployFlags { sandbox: false }),+      argv: svec!["--prod"],+      ..Flags::default()+    }+  );++  let r =+    flags_from_vec(svec!["deno", "deploy", "--project=myapp", "--prod", "-y"]);+  assert_eq!(+    r.unwrap(),+    Flags {

feat(desktop): clipboard apiRust · 332 + / 3 −

Introduces 12 new declarations in cli/rt/desktop.rs.

Adds new language design rather than adjusting what was there. Read the linked PR for the surrounding test context.

cli/rt/desktop.rs ↗ · 7 files

@@ -696,6 +698,75 @@ pub const DESKTOP_JS: &str = r#"     console.error("[deno desktop] failed to install navigator.permissions:", e);   } +  // --- navigator.clipboard (text only) ---+  //+  // Spec surface: `navigator.clipboard` is a `Clipboard` (extends EventTarget)+  // exposing async `readText()` / `writeText()`. The ops behind them are+  // genuinely async: laufey's clipboard calls block their calling thread, and+  // on X11/Wayland a read is serviced by whichever app owns the selection, so+  // an unresponsive owner would otherwise freeze the whole runtime behind a+  // Promise that looks like it couldn't. They reject rather than resolve if+  // that owner never answers — `""` is indistinguishable from an empty+  // clipboard, and a resolved `writeText()` has to mean the write happened.+  //+  // The richer `read()` / `write()` (`ClipboardItem` / arbitrary MIME types)+  // aren't backed by laufey, so they're omitted rather than stubbed. Per spec+  // the read/write are gated on the `clipboard-read` / `clipboard-write`+  // permissions, but laufey has no clipboard permission model, so access+  // isn't gated here (mirroring how the desktop Notification surface+  // degrades).+  const webidl = internals.webidl;+  class Clipboard extends EventTarget {+    constructor() {+      super();+      webidl.illegalConstructor();+    }++    async readText() {+      webidl.assertBranded(this, ClipboardPrototype);+      return (await op_desktop_read_clipboard_text()) ?? "";+    }++    async writeText(data) {

sessions: Improve GitHub context attachmentsTypeScript · 63 + / 27 −

Introduces 1 new declaration in src/vs/sessions/contrib/chat/browser/newChatContextAttachments.ts.

Adds new core implementation rather than adjusting what was there. Tests changed with it, with code of their own.

src/vs/sessions/contrib/chat/browser/newChatContextAttachments.ts ↗ · 7 files

@@ -132,26 +134,41 @@ export class NewChatContextAttachments extends Disposable implements INewChatAtt 		for (const entry of visibleAttachments) { 			const pill = dom.append(this._container, dom.$('.sessions-chat-attachment-pill')); 			const resource = URI.isUri(entry.value) ? entry.value : isLocation(entry.value) ? entry.value.uri : undefined;+			const githubContextResource = entry.id.startsWith(GITHUB_CONTEXT_ID_PREFIX)+				? URI.parse(entry.id.slice(GITHUB_CONTEXT_ID_PREFIX.length))+				: undefined;+			const openResource = resource ?? githubContextResource;+			const imageData = entry.kind === 'image' ? coerceImageBuffer(entry.value) : undefined;+			const canOpen = Boolean(imageData || openResource || isPastedTextArtifact(entry));+			let content: HTMLElement;+			if (canOpen) {+				const openButton = dom.append(pill, dom.$<HTMLButtonElement>('button.sessions-chat-attachment-open'));+				openButton.type = 'button';+				openButton.setAttribute('aria-label', localize('openNamedAttachment', "Open {0}", entry.name));+				content = openButton;+				pill.classList.add('openable');+			} else {+				content = dom.append(pill, dom.$('span.sessions-chat-attachment-content'));+			} 			if (entry.kind === 'image') {-				const icon = dom.append(pill, renderIcon(Codicon.fileMedia));-				dom.append(pill, dom.$('span.sessions-chat-attachment-name', undefined, entry.name));-				const buffer = coerceImageBuffer(entry.value);-				if (buffer) {+				const icon = dom.append(content, renderIcon(Codicon.fileMedia));+				dom.append(content, dom.$('span.sessions-chat-attachment-name', undefined, entry.name));+				if (imageData) { 					// Swap the generic icon for a thumbnail once the shared helper 					// has decoded one, matching the workbench attachment pill.-					const preview = createImageHoverContent(resource, entry.name, buffer, entry.id, undefined, undefined, (url, isThumbnail) => {+					const preview = createImageHoverContent(resource, entry.name, imageData, entry.id, undefined, undefined, (url, isThumbnail) => { 						if (isThumbnail) { 							icon.replaceWith(dom.$('img.sessions-chat-attachment-image', { src: url, alt: '' }));

model: qwen4exp: reduce number of graph splitsC++ · 27 + / 10 −

Reworks 28 lines of existing logic in src/models/qwen4exp.cpp.

Changes how existing runtime behaves. Read the linked PR for the surrounding test context.

src/models/qwen4exp.cpp ↗ · 2 files

@@ -1111,7 +1114,18 @@ ggml_tensor * llama_model_qwen4exp::graph::build_ple(     // gather then flatten the heads: get_rows lays the head dimension out slowest, as the reference does     ggml_tensor * emb = ggml_get_rows(ctx0, model.per_layer_tok_embd, rows);     emb = ggml_reshape_2d(ctx0, emb, hparams.ple_head_dim * n_heads, n_tokens);-    cb(emb, "ple_embd", il);+    cb(emb, "ple_embd", -1);++    return emb;+}++ggml_tensor * llama_model_qwen4exp::graph::build_ple(+        llm_graph_input_rs * inp,+        ggml_tensor *        emb,+        ggml_tensor *        hidden,+        int                  il) {+    const int64_t hc      = hparams.dsv4_hc_mult;+    const int64_t hc_dim  = hc * n_embd;      ggml_tensor * key   = build_lora_mm(model.layers[il].ple_key,   emb);     ggml_tensor * value = build_lora_mm(model.layers[il].ple_value, emb);

apiextensions: report why the test server's healthz wait failedGo · 21 + / 7 −

Reworks 22 lines of existing logic in staging/src/k8s.io/apiextensions-apiserver/pkg/cmd/server/testing/testserver.go.

Changes how existing runtime behaves. Read the linked PR for the surrounding test context.

staging/src/k8s.io/apiextensions-apiserver/pkg/cmd/server/testing/testserver.go ↗ · 1 files

@@ -216,23 +216,37 @@ func StartTestServer(t Logger, _ *TestServerInstanceOptions, customFlags []strin 	if err != nil { 		return result, fmt.Errorf("failed to create a client: %v", err) 	}-	err = wait.Poll(100*time.Millisecond, time.Minute, func() (bool, error) {+	// Keep the outcome of the last probe around: on timeout it is the only+	// clue about what the server was still waiting for, and /healthz names+	// the checks that did not pass in its body.+	var lastStatus int+	var lastBody []byte+	var lastErr error+	err = wait.PollUntilContextTimeout(ctx, 100*time.Millisecond, time.Minute, false, func(ctx context.Context) (bool, error) { 		select { 		case err := <-errCh: 			return false, err 		default: 		} -		result := client.CoreV1().RESTClient().Get().AbsPath("/healthz").Do(context.TODO())+		res := client.CoreV1().RESTClient().Get().AbsPath("/healthz").Do(ctx) 		status := 0-		result.StatusCode(&status)-		if status == 200 {-			return true, nil+		res.StatusCode(&status)+		body, rerr := res.Raw()+		// The probe that runs into the poll deadline is aborted mid-flight and+		// carries no useful outcome; keep the last one that actually completed.+		if ctx.Err() == nil {+			lastStatus, lastBody, lastErr = status, body, rerr 		}-		return false, nil+		return status == 200, nil 	})

sqlite: keep sessions alive across SQLite callbacksC++ · 26 + / 2 −

Reworks 10 lines of existing logic in src/node_sqlite.cc.

Changes how existing runtime behaves. Tests changed with it, with code of their own.

src/node_sqlite.cc ↗ · 3 files

@@ -1024,6 +1024,15 @@ void DatabaseSync::RemoveBackup(BackupJob* job) {   backups_.erase(job); } +std::vector<BaseObjectPtr<Session>> DatabaseSync::PinSessions() const {+  std::vector<BaseObjectPtr<Session>> pinned;+  pinned.reserve(sessions_.size());+  for (Session* session : sessions_) {+    pinned.emplace_back(session);+  }+  return pinned;+}+ void DatabaseSync::DeleteSessions() {   // all attached sessions need to be deleted before the database is closed   // https://www.sqlite.org/session/sqlite3session_create.html

Revert "[rust-compiler] Bail out on nested TypeScript `this` parameters (#37232)"Rust · 4 + / 12 −

Reworks 16 lines of existing logic in compiler/crates/react_compiler_lowering/src/build_hir.rs.

Changes how existing language design behaves. Read the linked PR for the surrounding test context.

compiler/crates/react_compiler_lowering/src/build_hir.rs ↗ · 1 files

@@ -2589,25 +2589,17 @@ fn lower_block_statement(     block: &react_compiler_ast::statements::BlockStatement,     parent_scope: Option<react_compiler_ast::scope::ScopeId>, ) -> Result<(), CompilerError> {-    Ok(lower_block_statement_inner(-        builder,-        block,-        None,-        parent_scope,-    )?)+    let _ = lower_block_statement_inner(builder, block, None, parent_scope);+    Ok(()) }  fn lower_block_statement_with_scope(     builder: &mut HirBuilder,     block: &react_compiler_ast::statements::BlockStatement,     scope_override: react_compiler_ast::scope::ScopeId, ) -> Result<(), CompilerError> {-    Ok(lower_block_statement_inner(-        builder,-        block,-        Some(scope_override),-        None,-    )?)+    let _ = lower_block_statement_inner(builder, block, Some(scope_override), None);+    Ok(()) }  fn lower_block_statement_inner(