Skip to content

Fix reproducibility issue with recent rosetta versions - #175

Open
alexhulbert wants to merge 1 commit into
mainfrom
ah/rosetta-jit-cache-fix
Open

Fix reproducibility issue with recent rosetta versions#175
alexhulbert wants to merge 1 commit into
mainfrom
ah/rosetta-jit-cache-fix

Conversation

@alexhulbert

Copy link
Copy Markdown
Member

There was a Apple Silicon reproducibility regression due to recent rosetta versions leaving files in /.cache, which this PR fixes by adding that directory to the debloat list.

@alexhulbert
alexhulbert requested review from a team as code owners July 21, 2026 04:05
samlaf added a commit to SeismicSystems/seismic-images that referenced this pull request Jul 22, 2026
)

Fixes #46.

The reprotest gate failed bimodally (issue #46): jemalloc's configure,
built via tikv-jemalloc-sys inside seismic-reth, probes the build host's
virtual-address width (48 vs 57-bit/LA57) and bakes it into LG_VADDR, so
a mixed-SKU runner pool yields two possible binaries. Pin LG_VADDR=48 --
safe (Linux never returns >48-bit mappings unless explicitly hinted) and
jemalloc's compact-rtree fast path. Document that -C target-cpu=generic
is likewise load-bearing for blst's ADX host detection.

Instrument CI so the next host-sniffing dependency gets localized
instead of guessed at: lscpu in both build jobs (records CPU model and
virtual address width), and per-file sha256 listings of the unpacked
initrd, diffed by reprotest-compare on mismatch to name the exact
divergent files.

Also debloat /.cache (Rosetta JIT droppings on Apple Silicon builds),
copied from upstream
flashbots/flashbots-images#175, so local lima
builds can match CI measurements.
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.

2 participants