jun 24, 2026
8 links from the engineering internet.
polars 1.42.0 adds out-of-core spilling and a strict mode
the dataframe engine ships py-1.42.0 with naive out-of-core spilling so queries can exceed ram, an experimental strict mode, is_sorted() on expr and dataframe, sql implicit join syntax, and bytes-based concurrency control for cloud io.
ruff 0.15.19 ships its crates to crates.io and fixes a notebook panic
astral's python linter and formatter now publishes its crates to crates.io, fixes a panic when inserting text at a notebook cell boundary, handles invalid editor-only settings, and improves pylint dunder method fixes and formatter performance.
mastodon 4.6.1 patches ldap login and remote key-fetch errors
the fediverse server ships a 4.6.1 patch fixing login failures under certain ldap configs, remote key fetching and unavailable-key handling, several advanced web ui regressions, and adds avatar and header description fields to the update_credentials api.
grafana loki 3.6.12 fixes a chunk-delete panic and possible data loss
the log aggregation system patches a panic when a whole chunk is deleted by line filters, possible data loss in dataobj-consumers, an s3 client region regression, and rolls up cves across the 3.6.x line on go 1.25.8.
strapi 5.49.0 adds mcp tool builders and fixes upload buffering
the headless cms ships definetool, defineresource and defineprompt builders for its mcp integration, stops buffering large uploads during mime detection, fixes premature admin logout on token expiry, and reworks save and publish keyboard shortcuts.
pulumi 3.248.0 adds snippets and output.recover to the engine
the iac tool adds snippets, blocks of pcl kept in state to track ad-hoc resources, and output.recover to catch exceptions in node.js outputs. it passes schema-loader and package-resolver addresses to providers in the handshake and fixes a get-schema panic.
n8n 2.27.4 lets allowlisted python packages use relative imports
the workflow automation tool patches its python code node so allowlisted packages can import their own submodules via relative imports, fixes building incorrect chained nodes, and bumps the google ads node to api v21.
vllm cuts the first v0.24.0 release candidate
the llm inference engine tags v0.24.0rc1, opening the release-candidate cycle for its next minor. the build fixes topk histogram compilation on sm75 turing gpus, restoring builds for older nvidia hardware.