Skip to content

chore(deps): bump approx-chol 0.4.0 -> 0.5.0 - #253

Merged
schroedk merged 1 commit into
mainfrom
chore/approx-chol-0.5.0
Aug 12, 2026
Merged

chore(deps): bump approx-chol 0.4.0 -> 0.5.0#253
schroedk merged 1 commit into
mainfrom
chore/approx-chol-0.5.0

Conversation

@schroedk

Copy link
Copy Markdown
Collaborator

Bumps approx-chol 0.4.0 → 0.5.0. No API changes at within's call sites; MSRV stays 1.85.

Breaking: the serialized Preconditioner wire format shifts v12 → v13. approx-chol dropped
original_n from its serialized Factor and bumped its own FACTOR_FORMAT_VERSION, so 0.3.0
bytes no longer decode. Fixture regenerated; v12 becomes the pre-bump negative.

The sampler is faster and the sampled Schur is unchanged. Standalone A/B with both versions
linked into one binary, interleaved arms, 0.4.0 re-run as a null control, n=25:

case 0.4.0 0.5.0 Δ noise floor
deg-10 stars, split=1 290.5 ns 276.8 ns −4.70% (21/25) −0.64%
deg-40 stars, split=1 1431 ns 1400 ns −2.19% (22/25) +0.27%
deg-10 stars, split=2 386.3 ns 386.6 ns +0.09% (null) +1.02%

Fill-edge checksums identical in all three. 0.5.0 also replaced the uniform-draw mapping with a
53-bit one that cannot return 1.0; it moves a draw by at most 2^-53, too little to change a CDF
bucket (zero divergences in 1.7M draws), so the preconditioner is bitwise unchanged.

The dependency dropped `original_n` from the serialized `Factor` and bumped its
own FACTOR_FORMAT_VERSION, so within's `Preconditioner` wire format shifts
v12 -> v13 transitively. Fixture regenerated; v12 becomes the pre-bump negative.
@schroedk
schroedk merged commit 244e9a6 into main Aug 12, 2026
5 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant