Skip to content

Release 0.10.0: promote the release candidate unchanged - #115

Merged
codetalcott merged 1 commit into
mainfrom
claude/release-0.10.0
Aug 26, 2026
Merged

Release 0.10.0: promote the release candidate unchanged#115
codetalcott merged 1 commit into
mainfrom
claude/release-0.10.0

Conversation

@codetalcott

Copy link
Copy Markdown
Owner

Version strings and changelog. No source changes from 0.10.0rc1.

The rc did what it was for: published through trusted publishing, installed from the real index on machines that never built it, and served. Nothing needed fixing afterwards, so this is the same artifact under a stable number.

$ pip install m0serve && m0serve pypiapp:application
served from PyPI

What the rc cost, and why the changelog keeps it

Three attempts, four defects, every one invisible from the machine that built the artifact:

defect found by
SIGILL — mojo build defaults --target-cpu to the build host wheel-consume-linux
otool is macOS-only, so Linux couldn't inspect half the wheels wheel-inspect
glob never expanded in the negative control the guard refusing a wrong-reason refusal
rc published as "Latest" above the stable release watching publish succeed

The first is the one that would have shipped. It isn't detectable by static inspection at all — the binary is correct, it just needs instructions the consumer's CPU doesn't implement. Only running it on different silicon finds it, which is exactly what those jobs do.

After merge

Tag v0.10.0. publish-pypi uploads on the same proven path; the pre-release flag now infers correctly from the tag, so v0.10.0 will publish as Latest and 0.10.0rc1 stays marked Pre-release.

Test

m0serve --version reports 0.10.0; smoke-wheel and smoke-serve green; check-docs agrees across all three version copies (pyproject.toml, M0SERVE_VERSION, uv.lock).

🤖 Generated with Claude Code

0.10.0rc1 did its job. It published through trusted publishing, installed
from the real index on machines that never built it, and served a WSGI app.
Nothing needed fixing afterwards, so this is the same artifact under a stable
number -- version strings only, no source changes.

The changelog keeps what the rc cost, because the reasoning outlives the
incident: three attempts, four defects, every one invisible from the machine
that built the artifact. The sharpest was SIGILL from `mojo build` defaulting
--target-cpu to the build host, which no static check could have found -- the
binary was correct, it just needed instructions the consumer's CPU did not
implement.

Verified at 0.10.0: builds, `m0serve --version` reports it, smoke-wheel and
smoke-serve green.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
@codetalcott
codetalcott merged commit eb23705 into main Aug 26, 2026
4 checks passed
@codetalcott
codetalcott deleted the claude/release-0.10.0 branch August 26, 2026 00:01
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