Skip to content

The record of the round, and the verification it rests on - #46

Merged
Sunrisepeak merged 1 commit into
mainfrom
openkal-0.15.0-record
Oct 2, 2026
Merged

Sunrisepeak merged 1 commit into
mainfrom
openkal-0.15.0-record

Conversation

@Sunrisepeak

Copy link
Copy Markdown
Member

This is the record of openkal 0.15.0 and the six repositories that moved with it: what was measured and where, the two defects the round exposed in openkal-linux and their repairs, what the round could not measure, the transcript of the ecosystem verification against the published index with the CN mirror selected, and a review at the scale of the ecosystem.

The verification script beside it is the instrument that produced the transcript: xlings subos use okl015 --sandbox --cmd "MCPP_VERIFY_VERSION=… bash /tmp/v.sh", with both probes resolved from the published index — the specification with openkal-linux for one, and openkal-musl for the other — and every step that cannot run on this machine recorded as not run rather than passed.

It is a separate pull request because a record written before the releases is a plan, and this one could not be written until every row it names was published.

…s it exposed

The design and the execution plan are already in this tree; this is what
happened. It carries the ecosystem verification script beside it --- run inside
an xlings sandbox against the published index with the CN mirror selected ---
and the transcript of that run, which is 0 assertions failed.

The round delivered one declaration and five implementations that answer it, and
it exposed two defects in the implementation beneath:

  1. The thread-local image was copied eight bytes below the variables that name
     it, because the segment's alignment was clamped to sixteen where the block's
     SIZE is computed. Found by openkal-llvm-runtime's own probe against the
     released packages, not by anything here.
  2. A context in the hosted arrangement was given storage laid out from an
     undescribed image, because only this implementation's own entry point ever
     described it.

Both are repaired, the repair is released (0.16.1, 0.20.1, 0.15.4), and the
record states what the round could not measure as plainly as what it did.
@Sunrisepeak
Sunrisepeak merged commit f5bc061 into main Oct 2, 2026
12 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