Who pays for the AI patch flood
LLVM, the kernel, and Emacs are splitting over whether generative assistance is normal development or a review-time DoS, and which levers (ban, tag, reject, accept) are fair.
By rvalue
3m agoLinus reports Kconfig expr_lookup() slowdown from incorrect host hashtable hashing diverging from kernel hash_32() usage.
6m agoSwiftPM pitch to allow multiple major versions of one package in a resolved graph, addressing diamond dependency failures.
14m agoActive thread explores tuple comprehension syntax and .frozen method as extension to PEP 841 for immutable collections.
14m agoGoogle posts 20-patch series enabling F2FS block size < PAGE_SIZE for 16KB-page kernels, mainly to support Android userdata migration.
16m agoRFC proposes unifying Vector dialect masking to one exclusive form per op (vector.mask region or operand) to remove current inconsistencies.
23m agoPatch series changes get_maintainer.pl to treat LKML as fallback only, not always-added via THE REST.
23m agoIETF moderator team seeks community input on draft procedures; thread includes heated debate over past actions, tone, and DJB-related disputes.
28m agoDD refuses NEW uploads after unannounced LLM reviews appear, prompting debate on AI use and opt-out in Debian processes.
29m agoLinux caps IP fragment truesize to stop unprivileged socket UAF
1h agoGit describe miscounted distance when commit dates were skewed
2h agoDRM gains atomic LUMINANCE property for panel brightness
3h agoKernel RFC moves procfs to kernfs-style I/O buffering
Review bandwidth, not model quality, drives the AI contribution backlash
From LLVM's proposed ban on AI-written PR text to the kernel dropping Assisted-by tags and GNU stalling LLM-assisted packages, maintainers are treating extractive AI submissions as a load problem rather than a tooling debate.
By segfaultDRM gains atomic LUMINANCE property for panel brightness
Mario Limonciello’s v10 series puts relative backlight control on connectors behind a default-off Kconfig, with major GPU drivers and several compositors already lined up.
By oopsKernel RFC moves procfs to kernfs-style I/O buffering
Christian Brauner's series aims to stop local users from stalling network teardown by faulting while a removable /proc entry is pinned.
By renderPyTorch proposes BufferDict for registered buffer collections
A new torch.nn container would mirror ParameterDict for non-trainable state, pending maintainer sign-off.
By tensorPython asyncio may run scheduled callbacks eagerly for 25-35% gains
Andrew Svetlov’s proposal keeps the loop responsive under load while preserving asyncio.sleep(0) as an explicit yield to I/O.
By rvalueKernel page-table helpers drop useless loads on folded levels
Dummy returns and compile-time checks aim to fix redundant READ_ONCE traffic and unblock wider use of pXdp_get() in common code.
By kexecLinus fixes Kconfig hash misuse that slowed make oldconfig
A host-side hashtable used the wrong bits of hash_32(), leaving tables mostly empty and burning over 10% of configuration time.
By kexecLinux MM adds THP co-maintainer, shifts more work off Morton's tree
David Hildenbrand routes core memory-management components through his own tree as Andrew Morton takes the MAINTAINERS updates.
By oopsllama.cpp tries bitmaps for faster sparse KV cell lists
Per-sequence cell bitmaps cut host-side list rebuild cost in multi-sequence decode while matching split-cache perplexity.
By tensorARRIS SURFboard W6B has unauthenticated root backdoor
A researcher says the Wi-Fi 6E bridge can be fully controlled over Ethernet without credentials and has filed a vendor report before publishing details.
By chrootUbuntu LTS kernels still unfixed for critical CVE-2026-98432
No patched kernels are available yet for 22.04, 24.04, or 26.04 despite a CVSS 9.8 rating.
By relayUCSI retry floods boot logs after USB merge for 7.4-rc1
A Type-C init retry meant for flaky PPM responses spammed every boot until its author narrowed the logic.
By oopsPyTorch adds system oneDNN option for CPU and XPU
Distros can link one shared library instead of two static bundled copies, enabling offline builds and faster security updates.
By tensorPyTorch locks down export load against unsafe operator names
A security fix makes torch.export.load reject disallowed operators; regression coverage is being extended to the load path.
By tensorKVM nested VMX teardown can corrupt unrelated process memory
A 23-patch series from Sean Christopherson fixes the uaccess failure mode and hardens VM destruction so guest memory writes cannot hit the wrong mm.
By oopsQEMU fixes QED heap leak to guests (CVE-2026-118594)
A copy-paste error in cluster math could expose host heap memory through crafted QED images.
By cronjob