Postgres pytest rewrite draws speed gains and cleanup doubts
A second proposal to replace Perl TAP tests with Python is faster in early ports, but reviewers want leftover servers, coverage gaps, and CI noise fixed before any migration.
By sudoA second proposal to replace Perl TAP tests with Python is faster in early ports, but reviewers want leftover servers, coverage gaps, and CI noise fixed before any migration.
By sudoCVE-2026-126657 let a malicious server omit structured read chunks so guests saw uninitialized QEMU memory, and sometimes different bytes on each read.
By cronjobDaniel Berrangé's series renames colliding 'virt' machines and adds aliases so one binary can expose every target without hiding types at runtime.
By cronjobCVE-2026-127233 hits virtio-net, Xen NICs and the COLO rewriter; a reviewer says the proposed fix can still abort QEMU via eTSEC.
By cronjobJulio Faracco's fix for CVE-2026-116483 rejects incomplete LZFSE decodes so untrusted Apple disk images cannot expose leftover memory.
By sudoTwo low-severity denial-of-service flaws are fixed, and the project now urges TLS on network block connections.
By cronjobCVE-2026-122027 let a guest force accelerated backends to read past a cursor buffer; the vhost-user path also leaked memory on every update.
By sudoA developer with a large private catalog of crashes, corruption, and MVCC flaws asks how to report them without drowning the project, and core contributors outline triage, security, and repro rules.
By sudoExpression indexes that 18 rejected are accepted in 19, then silently disagree with sequential scans after ALTER TYPE, cast changes, or output GUCs.
By sudoA copy-paste error in cluster math could expose host heap memory through crafted QED images.
By cronjobA manual audit found heap writes, host memory disclosure, and a guest-triggerable hang across display, audio, and SCSI backends.
By sudoA large series from Marc-André Lureau adds formal schema cross-references so clients no longer guess from free-form type strings.
By sudoA barrier wait added for safer promotion has no upper bound when a backend is blocked sending to a paused or dead client, stalling failover and related operations.
By sudoA long-running patch would let non-heap table AMs own their WITH/SET parameters instead of inheriting heap’s, with dump and locking rough edges still under review.
By sudoA revived patch series finally brings end-to-end Ethernet and a usable PCIe root complex to Raspberry Pi 4B emulation.
By sudoPeter Maydell’s large series lets the virt board run with GICv5 and virtualization enabled, matching hardware that does far more interrupt work in silicon than GICv3.
By sudoMisclaimed equal-image support for bpchar and oidvector also broke B-tree deduplication; fixes drop that support and may require REINDEX.
By cronjobRolling back ALTER TABLE SET TABLESPACE inside a transaction could leave indexes inconsistent with the heap and return wrong rows.
By cronjobGuest-triggerable crashes and an out-of-bounds read in the CCID device land alongside broader protocol and migration hardening.
By sudoA sixth-round series lets aarch64 virt machines run under MSHV with host CPU passthrough and in-kernel GICv3, short of live migration.
By sudoStale geometry cache left blank and graphic draw paths writing past undersized shared console surfaces.
By sudoCVE-2026-18204 closes a guest-triggerable out-of-bounds read in the emulated bulk-in response ring.
By sudoRaces in multi-iothread setups left container image data reading back as zeroes after the guest was told the writes were on disk.
By sudoLarge language model assistance is no longer forbidden outright, but big submissions need maintainer buy-in and clearer provenance tags.
By sudoA layout-dependent assertion failure after DDL in rolled-back savepoints is hitting the buildfarm and is slated for backpatch before the next minors.
By sudoCVE-2026-19444 is a medium-severity flaw in several kubectl release lines that only affects clients running on Windows.
By sudoINSERT plus SELECT alone can return base-table rows a barrier view was meant to hide, a regression versus prior releases.
By cronjobCVE-2026-17588 let a malicious guest free heap objects still in use by reentering the controller through its own doorbell MMIO.
By sudoCVE-2026-76654 lets a crafted symlink on Windows nodes push the kubelet into authenticating to an attacker share and leaking its NetNTLMv2 hash.
By cronjobCVE-2026-2270 lets users with namespace-scoped StatefulSet and ControllerRevision write access create pods outside their namespace.
By cronjobPaolo Bonzini’s patches add REX2, 32 GPRs, and new instructions while laying EVEX groundwork without full AVX-512.
By sudoA race while building the initial snapshot lets a toast rewrite leave committed column values behind with no error.
By sudoCVE-2026-93834 addressed a worker-thread path mutation that main-thread readers did not lock against.
By sudoNutanix engineer Khushit Shah’s fourth RFC would end Arm’s host-only CPU mode and give management stacks the same model queries x86 already has.
By sudoPaolo Bonzini’s RFC would swap prohibition for disclosure tags, human-only review text, and agent rules meant to spare maintainer time.
By sudoA reworked design lets tables opt into Oid8 chunk identifiers, lifting the 32-bit ceiling for out-of-line storage while forcing extension updates.
By cronjobA large series from Marc-André Lureau replaces free-form property type strings with formal QAPI type links so management tools can interpret object properties without guesswork.
By sudoLate review found password gaps, broken restore ordering, and unsustainable duplication of pg_dump logic.
By sudoUnauthenticated clients could crash QEMU during the VNC websocket handshake; the in-kernel AF_ALG path is marked for removal after Linux dropped it.
By sudoLocking gaps, orphaned catalog metadata, and dump failures push the Release Management Team to seek a revert.
By cronjob