reproducible builds without shm
fix #614 (closed)
I've determined the two source of nondeterminism in Arti builds are lzma-sys
and zstd-sys
.
There was a PR, zstd-rs!162, making zstd-sys reproducible. Sadly there isn't a release yet, so it's available by git dependency only.
I made a similar PR on xz2-rs
, xz2-rs!102, to fix the same issue with lzma-sys. I was able to verify it indeed fix reproducibility issues. As of now, it's only available via git dependency too.
Merge request reports
Activity
requested review from @nickm
this MR is there so 1/ people know #614 is being worked on 2/ I don't forget I'm working on it (which for the record is what happened with !819 (merged), I just found I had that branch from a month ago).
Feel free to have a look, but there is honestly not much to look at
Edited by trinity-1686a
added 366 commits
-
afafeefc...8cf9e2a6 - 364 commits from branch
tpo/core:main
- 3d5d4ada - use newer version of zstd and lzma
- 9d741b07 - always run build-repro (for tests purpose)
-
afafeefc...8cf9e2a6 - 364 commits from branch
expected result (tested in CI and on my own machine, but more confirmation is always better):
branch : repro-less-shm commit : 3d5d4ada0190ac935b3786d9d36614c75cb5fec6 Linux hash : 2c534238d69da28e4222ad351447b8e3bbf69259dd8b334b650ab6fe8ed2db80 Windows hash : 733f4fcf6c0e3b10890c700e77a556abdff2a3bbdcd703c5c61dad547dd1286e MacOS hash : 551d5961dee3adbe44e4860184c6210b2c622245d1f3fab9355cc522db1c939d
Edited by trinity-1686aenabled an automatic merge when the pipeline for 3d5d4ada succeeds
mentioned in commit e317c0ad
mentioned in issue tpo/tpa/gitlab#135 (closed)