140 wires and counting

$ follow Google

Keep up with Google in about 3 minutes: what actually shipped — the commits, pull requests, releases, and security advisories that matter.

or

fair warning: these emails are deeply technical. diffs, version numbers, CVEs, benchmark deltas. if that's not your idea of a good read, this isn't your newsletter.

Folds into your digest — weekly by default, monthly if you prefer. Unsubscribe in one click.

$ status

wire 2026-09-19
stories 74

© 2026 RepoJournal Home Showcase How it works Privacy

$ the-wire · showcase

JAX drops NumPy 2.1 and stale version gates

By RepoJournal · Filed · About Google · Composed from the cited sources · methodology

JAX's build floor moves up while its numerical tests start measuring true ULP error instead of bit distance, and the Firestore Python SDK grows BSON timestamp and binary types.

Drop support for NumPy 2.1. google/jax

by Peter Hawkins

Support for NumPy 2.1 is gone, and the requirements lock files were regenerated alongside it. If your environment still pins NumPy 2.1, installing current jax now fails at resolve time rather than at runtime, so bump the pin before upgrading.

Measure real ULP errors in `numerics` tests (`elementary_test` and `special_test`). google/jax

by copybara-service

The numerics test helper in `numerics_test_util` now measures continuous floating-point ULP error instead of integer bit-distance between rounded output and rounded reference, computing a high-precision reference (float64 for narrower types, mpmath for float64) and measuring error against `ulp(ref)`. It also changes the default ULP error bound, so `elementary_test` and `special_test` assert aga...

Consolidate elementary and special function accuracy documentation. google/jax

by copybara-service

A new docs/accuracy.md collects maximum error bounds for elementary and special functions across CPU, GPU, and TPU for bfloat16, float16, float32, and float64, with cross-references added from docstrings throughout jax.numpy, jax.lax, and jax.scipy.special. Precision expectations that previously lived in scattered test tolerances are now written down per architecture and dtype.

Remove version checks that are now always false: google/jax

by Peter Hawkins

The dead branches came out: checks for jaxlib minimum version 0.11.2, ifrt_version 69, jaxlib_extension_version 495, and libtpu 0.0.48 are always false and were deleted. Anyone building jaxlib or libtpu from source should treat those as the effective floors.

feat(firestore): add BSONTimestamp support googleapis/google-cloud-python

by ohmayr

Firestore gains a `BSONTimestamp` container class in `google.cloud.firestore_v1.bson`, wrapping the 32-bit unsigned `seconds` and `increment` fields of an internal BSON operation log timestamp with spec-compliant unsigned conversion, alongside BSONBinary support for subtyped payloads such as UUIDs and MD5 hashes. The long tail was quiet: two `chore: Misc updates` commits in python-genai, a blac...

Quick answers

What shipped in Google on September 19, 2026?
JAX's build floor moves up while its numerical tests start measuring true ULP error instead of bit distance, and the Firestore Python SDK grows BSON timestamp and binary types. In total, 40 commits and 34 pull requests landed.
Who contributed to Google on September 19, 2026?
5 developers shipped this update, including Peter Hawkins, copybara-service, Ivan Lopez Hernandez, ohmayr, and daniel-sanche.
What were the notable Google updates?
Drop support for NumPy 2.1, Measure real ULP errors in `numerics` tests (`elementary_test` and `special_test`), and Consolidate elementary and special function accuracy documentation.