Oct 9, 2026

ENH: sniff sep=None with the C engine in read_csv (GH#9645)Python · 105 + / 9 −

Introduces 5 new declarations in pandas/io/parsers/readers.py.

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

pandas/io/parsers/readers.py ↗ · 7 files

@@ -3643,3 +3689,53 @@ def _validate_skipfooter(kwds: dict[str, Any]) -> None:             raise ValueError("'skipfooter' not supported for iteration")         if kwds.get("nrows"):             raise ValueError("'skipfooter' not supported with 'nrows'")+++def _iter_physical_lines(+    handle: IO[Any], chunks: list[bytes | str], lineterminator: str | None+) -> Iterator[bytes | str]:+    """+    Yield each line of ``handle`` without its terminator, appending every chunk+    read to ``chunks``.+    """+    eol: re.Pattern[Any] | None = None+    partial: list[Any] = []+    skip_lf = False+    while True:+        chunk = handle.read(_SNIFF_CHUNK_SIZE)+        if not chunk:+            tail = partial[0][:0].join(partial) if partial else ""+            if tail:+                yield tail+            return+        chunks.append(chunk)+        if eol is None:+            pattern = re.escape(lineterminator) if lineterminator else r"\r\n|\r|\n"+            eol = re.compile(pattern.encode() if isinstance(chunk, bytes) else pattern)+        start = 0+        if skip_lf and chunk[:1] in (b"\n", "\n"):+            # the second half of a "\r\n" split across chunks+            start = 1+        for match in eol.finditer(chunk, start):+            partial.append(chunk[start : match.start()])+            yield chunk[:0].join(partial)

MIR move elimination [3/6]: PreciseLivenessRust · 646 + / 1 −

Introduces 41 new declarations in compiler/rustc_mir_dataflow/src/impls/precise_liveness.rs.

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

compiler/rustc_mir_dataflow/src/impls/precise_liveness.rs ↗ · 21 files

@@ -0,0 +1,596 @@+//! Computes the points where each local must have a distinct allocation.+//!+//! The result is a [`SparseIntervalMatrix`] with one row per local. Two locals+//! may share the same address only if their rows are disjoint. To model MIR+//! statements where a source operand and destination place may share an+//! address, each statement and terminator is split into an early point, where+//! operands are read, and a late point, where destinations are written.+//!+//! A local live range starts at the late point of any statement or terminator+//! that writes to it without a `Deref` projection. It ends at `StorageLive` or+//! `StorageDead`, or at the early point of a whole-local move operand or the+//! last use of that local on a control-flow path (only for locals whose address+//! is never observed). This happens at `StorageLive` because it resets a local+//! to an uninitialized state.+//!+//! `Call` terminators are handled specially: move operands are kept live+//! through the late point of the terminator so they conflict with each other+//! and with the destination place. This matches the runtime behavior where the+//! place is donated to the callee for the duration of the call.+//!+//! The analysis consists of three passes:+//!+//! 1. A backward value-liveness analysis which identifies the last point at+//!    which each local is directly used in each control flow path. This is used+//!    to shorten the lifetime of unborrowed locals to their last use.+//! 2. A forward allocation-liveness analysis which uses that information+//!    together with writes, moves, and storage markers to compute which+//!    allocations may be live on entry to each block.+//! 3. A final walk using those entry states which constructs the matrix and+//!    distinguishes whether each effect happens at the early or late point of a+//!    statement or terminator.++use rustc_index::IndexVec;

turbopack: report an actionable error when Node.js child processes cannot connectRust · 168 + / 6 −

Introduces 18 new declarations in turbopack/crates/turbopack-node/src/evaluate.rs.

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

turbopack/crates/turbopack-node/src/evaluate.rs ↗ · 11 files

@@ -769,3 +826,76 @@ impl Issue for EvaluationIssue {         Some(self.source)     } }++/// Node.js child processes could not connect back to the parent process, which+/// makes it impossible to run any Node.js code (webpack loaders, PostCSS, ...).+#[turbo_tasks::value(shared)]+pub struct NodeJsConnectIssue {+    /// The source being transformed when the connection failed. The failure is+    /// unrelated to its contents, but this tells the user which transform hit it.+    pub source: IssueSource,+    /// What was being evaluated, e.g. the loader chain.+    pub crash_context: Option<RcStr>,+    /// Error output of the child process, describing why the connection failed.+    pub stderr: RcStr,+}++#[async_trait]+#[turbo_tasks::value_impl]+impl Issue for NodeJsConnectIssue {+    fn severity(&self) -> IssueSeverity {+        IssueSeverity::Error+    }++    async fn title(&self) -> Result<StyledString> {+        Ok(StyledString::Text(rcstr!(+            "Node.js worker processes could not connect to Turbopack"+        )))+    }++    fn stage(&self) -> IssueStage {+        IssueStage::Transform+    }

Refactor apiserver_resource_size_estimate_bytes metric and apiserver_resource_objects to have a timestampGo · 134 + / 25 −

Introduces 9 new declarations in staging/src/k8s.io/apiserver/pkg/storage/etcd3/metrics/timestamped_resource_gauge_collector.go.

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

staging/src/k8s.io/apiserver/pkg/storage/etcd3/metrics/timestamped_resource_gauge_collector.go ↗ · 4 files

@@ -0,0 +1,110 @@+/*+Copyright The Kubernetes Authors.++Licensed under the Apache License, Version 2.0 (the "License");+you may not use this file except in compliance with the License.+You may obtain a copy of the License at++    http://www.apache.org/licenses/LICENSE-2.0++Unless required by applicable law or agreed to in writing, software+distributed under the License is distributed on an "AS IS" BASIS,+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.+See the License for the specific language governing permissions and+limitations under the License.+*/++package metrics++import (+	"sync"+	"time"++	"k8s.io/apimachinery/pkg/runtime/schema"+	compbasemetrics "k8s.io/component-base/metrics"+	"k8s.io/utils/clock"+)++type timestampedResourceGaugeCollector struct {+	compbasemetrics.BaseStableCollector++	desc  *compbasemetrics.Desc+	clock clock.PassiveClock+

Second phase of torch 2.6 cleanupsPython · 19 + / 108 −

Reworks 41 lines of existing logic in src/transformers/masking_utils.py.

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

src/transformers/masking_utils.py ↗ · 20 files

@@ -523,21 +517,14 @@ def sdpa_mask(         # Expand the mask to match batch size and query length if they weren't used in the mask function         attention_mask = attention_mask.expand(batch_size, -1, q_length, kv_length) -    # Option 2: Vmap mask creation (torch>=2.6 and custom patterns)-    elif _is_torch_greater_or_equal_than_2_6:+    # Option 2: Vmap mask creation (custom patterns)+    else:         # This creates the 4D mask easily. Note that we need this context manager as vmap cannot handle slicing a tensor from         # scalar tensor (it internally calls `.item()` which vmap does not allow, but this context works around it         # We don't need to add an offset to the mask_function either, as we vmap directly the correct indices for k and kv indices         with TransformGetItemToIndex():             attention_mask = _vmap_expansion_sdpa(mask_function)(batch_arange, head_arange, q_arange, kv_arange) -    # Option 3: Error out since it indicates that the user did something custom, which they shouldn't have (torch<2.6)-    else:-        raise ValueError(-            "The vmap functionality for mask creation is only supported from torch>=2.6. "-            "Please update your torch version or use `use_vmap=False` with index-based masks."-        )-     return attention_mask