$ cat google/week/2026-11-09.log
the week in review · Nov 9 – Nov 15, 2026
Merchant API v1beta removed from 13 Google Cloud packages
By RepoJournal · composed from the cited sources · human-reviewed weekly · methodology
JAX spent the week hardening numerics and sharding while python-genai shipped 3.2.0 and 3.3.0; cloud-python also dropped Python 3.10.
feat(merchantapi)!: remove discontinued Merchant API v1beta googleapis/google-cloud-python
The service has rejected every v1beta call since 2026-06-01, and the generator no longer emits the google.shopping.merchant_<api>_v1beta protos into any of the 13 shopping-merchant packages; the change went out as v2.0.0 in google-shopping-merchant-products. Import paths and message types for v1beta are simply absent now, so anything pinned to them needs a v1 migration before its next dependenc...
v3.3.0 googleapis/python-genai
by release-please[bot]
python-genai 3.3.0 adds --flavor=interactions-eap packaging, scrubs private exports, and adds XHIGH and xhigh reasoning effort levels. It follows 3.2.0 the next day, which brought the Decisions API via client.decisions.create and environments support. Two minor releases in two days on a client SDK that moves this often means pinning a range, not a version.
Add `mode="high"` to `jax.random.truncated_normal` to fix tail sampling accuracy. google/jax
Tail accuracy in jax.random.truncated_normal was wrong because inverse-CDF sampling ran through erf and erf_inv; the new mode="high" path fixes it. The old path still runs, so existing keys and reproducibility survive by default, but a sample from a far tail was not the sample you asked for.
Upcast narrow signed integer indices before negative-index wrapping. google/jax
Indexing an axis longer than 127 elements with jnp.int8(-1) previously wrapped the axis length and computed the wrong index, returning a silent wrong result rather than an error. Narrow signed indices are now upcast before negative-index wrapping, so code that relied on the bug (or never noticed it) changes behavior quietly.
Fix non-termination in `jnp.gcd` and `jnp.lcm` for minimum signed integers. google/jax
by Jake VanderPlas
Negating iinfo(dtype).min overflows back to the negative minimum in two's complement, which left jnp.gcd and jnp.lcm looping on those inputs instead of terminating. The fix lands in the JAX kernels; if a computation hung on a minimum signed integer, that was this.
Add `out_sharding` to `jnp.{quantile,nanquantile,nanpercentile}` and remove `auto_axes` from `jnp.percentile`. google/jax
out_sharding now propagates natively through _quantile, unsharding reduced axes and resharding the output, and jnp.percentile routes straight through jnp.quantile. The auto_axes keyword is gone from jnp.percentile, so multi-device code calling it that way needs the out_sharding spelling.
jax.random.uniform: Add a new higher-resolution shift-and-scale conversion, protected by a new `jax_legacy_random_uniform` flag google/jax
The shift-and-scale conversion in jax.random.uniform gains one extra bit of randomness in the least-significant place, doubling grid resolution, with performance neutral to marginally different depending on shape. It sits behind the jax_legacy_random_uniform flag, so set it if you need bit-identical draws from existing keys.
Remove deprecated keyword arguments from `jnp.clip` type signature. google/jax
by Jake VanderPlas
The deprecated keyword arguments were dropped from the runtime in JAX v0.10.0 and lingered only in the type stubs; they are now gone from the jnp.clip signature, which matters if type-checking caught the mismatch rather than runtime.
$ ls google/week/ # the briefings behind this review
Keep up with Google in about 3 minutes: what actually shipped — the commits, pull requests, releases, and security advisories that matter.
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.