7-day edition

Sep 8, 2026 – Sep 14, 2026

The edition exactly as it was published. Nothing here is re-selected on a later reading.

Sep 15, 20267 reads · 571 merged PRs screened

runtime

nodejs/node · #65985

Files changed →View PR ↗

perf_hooks: reuse buffer for uv metricsC++ · 35 + / 15

Reworks 33 lines of existing logic in src/node_perf.cc.

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

src/node_perf.cc · 6 files

@@ -265,17 +277,13 @@ void LoopIdleTime(const FunctionCallbackInfo<Value>& args) {  void UvMetricsInfo(const FunctionCallbackInfo<Value>& args) {   Environment* env = Environment::GetCurrent(args);-  Isolate* isolate = env->isolate();   uv_metrics_t metrics;   // uv_metrics_info always return 0   CHECK_EQ(uv_metrics_info(env->event_loop(), &metrics), 0);-  Local<Value> data[] = {-      Integer::New(isolate, metrics.loop_count),-      Integer::New(isolate, metrics.events),-      Integer::New(isolate, metrics.events_waiting),-  };-  Local<Array> arr = Array::New(env->isolate(), data, arraysize(data));-  args.GetReturnValue().Set(arr);+  AliasedInt32Array& buffer = env->performance_state()->uv_metrics;+  buffer[0] = static_cast<int32_t>(metrics.loop_count);+  buffer[1] = static_cast<int32_t>(metrics.events);+  buffer[2] = static_cast<int32_t>(metrics.events_waiting); }  void CreateELDHistogram(const FunctionCallbackInfo<Value>& args) {

core implementation

huggingface/accelerate · #4229

Files changed →View PR ↗

Fix CI Python · 32 + / 4

Introduces 1 new declaration in src/accelerate/utils/other.py.

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

src/accelerate/utils/other.py · 5 files

@@ -248,6 +248,29 @@ def model_has_dtensor(model: torch.nn.Module) -> bool:     return any(isinstance(p, DTensor) for p in model.parameters())  +def get_model_tp_size(model: torch.nn.Module) -> Optional[int]:+    """+    Get the tensor parallel degree a `transformers` model was sharded with, or `None` if it was not sharded.++    Args:+        model (`torch.nn.Module`):+            The model to inspect.++    Returns:+        `Optional[int]`: The model's tensor parallel size.+    """+    # `transformers<5` records it on the model itself, while `transformers>=5` moved it to the+    # `DistributedConfig` held by the model config and left `model.tp_size` behind as a `None` stub.+    tp_size = getattr(model, "tp_size", None)+    if tp_size is not None:+        return tp_size++    distributed_config = getattr(getattr(model, "config", None), "distributed_config", None)+    if isinstance(distributed_config, dict):+        return distributed_config.get("tp_size")+    return getattr(distributed_config, "tp_size", None)++ def extract_model_from_parallel(     model, keep_fp32_wrapper: bool = True, keep_torch_compile: bool = True, recursive: bool = False ):

language design

react/react · #37613

Files changed →View PR ↗

[DOM] Fix Fragment compareDocumentPosition for the root containerJavaScript · 19 + / 7

Introduces 1 new declaration in packages/react-dom-bindings/src/client/ReactFiberConfigDOM.js.

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

packages/react-dom-bindings/src/client/ReactFiberConfigDOM.js · 3 files

@@ -3715,13 +3716,13 @@ function validateDocumentPositionWithFiberTree(   }   if (documentPosition & Node.DOCUMENT_POSITION_CONTAINS) {     if (otherFiber === null) {-      // otherFiber could be null if its the document, documentElement, or body-      const ownerDocument = getOwnerDocumentFromRootContainer(otherNode);-      return (-        (otherNode as Instance | Document) === ownerDocument ||-        otherNode === ownerDocument.documentElement ||-        otherNode === ownerDocument.body-      );+      // otherFiber is null when otherNode is not part of a React tree. That+      // includes the document, documentElement and body, but also the root+      // container and any element above it. All of them contain the whole+      // React tree, so check containment of the root container rather than+      // enumerating the nodes above it.+      const rootContainer = getFragmentRootContainerInfo(fragmentFiber);+      return rootContainer !== null && otherNode.contains(rootContainer);     }     return isFragmentContainedByFiber(fragmentFiber, otherFiber);   }

core implementation

TheAlgorithms/Python · #15321

Files changed →View PR ↗

Surface overlapping PRs in pr_file_map.py to ease landing PRsPython · 47 + / 15

Reworks 59 lines of existing logic in scripts/pr_file_map.py.

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

scripts/pr_file_map.py · 1 files

@@ -6,16 +6,23 @@ and, for each file touched by any open PR, which PR number(s) touch it.  Output is GitHub-flavored Markdown that includes this script's path, the-current UTC datetime, and summary counts for open PRs, file entries, and-existing/missing files. It then renders a sorted list of files that currently-exist in the working directory, each with its modifying PR numbers, followed-by a separate section for files referenced by open PRs but that do not exist-in the working directory (e.g. deleted, renamed, or on a branch not checked-out locally).+current UTC datetime, and summary counts for open PRs, file touches, distinct+files, and existing/missing files. It highlights files touched by more than one+open PR first (the likely merge-conflict hot spots when landing PRs), then+renders a sorted list of files that currently exist in the working directory,+each with its modifying PR numbers, followed by a separate section for files+referenced by open PRs but that do not exist in the working directory (e.g.+deleted, renamed, or on a branch not checked out locally).++Two file totals are reported because they answer different questions:+  - "file touches" counts every (PR, file) pair, so a file edited by three open+    PRs contributes three touches; and+  - "distinct files" counts each touched path once.+Only the distinct total equals `existing + missing`, since those are deduped.  Run status is also written to stderr with:   - Number of PRs from `get_open_prs()`-  - Number of files from `get_pr_files()`+  - Number of file touches from `get_pr_files()` and distinct files touched   - Number of existing and missing files  Requirements: gh (GitHub CLI), authenticated (`gh auth login`)

language design

denoland/deno · #36736

Files changed →View PR ↗

fix(cli): generate valid bash completionsRust · 13 + / 4

Introduces 1 new declaration in libs/cli_parser/src/completions.rs.

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

libs/cli_parser/src/completions.rs · 1 files

@@ -747,6 +746,16 @@ mod tests {     assert!(v.iter().any(|f| f.starts_with("--")), "{v:?}");   } +  #[test]+  fn generate_bash_uses_real_newlines() {+    let s = String::from_utf8(generate("bash", &DENO_ROOT)).unwrap();+    assert!(+      s.contains("\"deno,run\")\n                cmd=\"deno__run\""),+      "{s}"+    );+    assert!(s.contains("            *)\n                ;;"), "{s}");+  }+   #[test]   fn generate_zsh_short_flags_keep_their_dash() {     // Regression test for #36713: the brace expansion dropped the `-` on the

core implementation

NVIDIA-NeMo/Speech · #15454

Files changed →View PR ↗

fix: replace assert statements with explicit raise in ASR modelsPython · 9 + / 2

Reworks 6 lines of existing logic in nemo/collections/asr/models/hybrid_rnnt_ctc_models.py.

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

nemo/collections/asr/models/hybrid_rnnt_ctc_models.py · 2 files

@@ -340,7 +340,11 @@ def change_decoding_strategy(             self.cur_decoder = "rnnt"             return super().change_decoding_strategy(decoding_cfg=decoding_cfg, verbose=verbose) -        assert decoder_type == 'ctc' and hasattr(self, 'ctc_decoder')+        if decoder_type != 'ctc' or not hasattr(self, 'ctc_decoder'):+            raise ValueError(+                f"Unsupported decoder_type '{decoder_type}'. "+                f"Expected 'ctc' with a 'ctc_decoder' attribute on the model."+            )         if decoding_cfg is None:             # Assume same decoding config as before             logging.info("No `decoding_cfg` passed when changing decoding strategy, using internal config")

framework internals

ggml-org/llama.cpp · #28646

Files changed →View PR ↗

webui: stop re-probing disabled /tools endpoint on every messageTypeScript · 12 + / 9

Reworks 7 lines of existing logic in tools/ui/src/lib/stores/tools.svelte.ts.

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

tools/ui/src/lib/stores/tools.svelte.ts · 3 files

@@ -246,13 +246,10 @@ class ToolsStore { 				toolInfos.filter((info) => info.uses_cwd).map((info) => info.tool) 			); 		} catch (err) {-			const errorMessage = err instanceof Error ? err.message : String(err);--			this._error = errorMessage;+			this._error = err instanceof Error ? err.message : String(err);  			// 403 from /tools means the server was started without --tools-			// TODO: check status code instead of relying on message-			if (errorMessage.includes('this feature is disabled')) {+			if (err instanceof ApiError && err.status === 403) { 				this._toolsEndpointUnreachable = true; 				console.info('[ToolsStore] Server tools are disabled on the server'); 			} else {

Read today’s edition →