Skip to content

release: 0.4.1; create the release as a draft, then publish - #14

Merged
HarryR merged 1 commit into
mainfrom
release-draft-then-publish
Nov 15, 2026
Merged

HarryR merged 1 commit into
mainfrom
release-draft-then-publish

Conversation

@HarryR

@HarryR HarryR commented Nov 15, 2026

Copy link
Copy Markdown
Contributor

v0.4.0's release job published the release before uploading its assets (softprops/action-gh-release with draft: false). An immutable release accepts no uploads once published, so v0.4.0 has no binaries; its notes now say so and point to v0.4.1.

  • The release job creates the release as a draft with the four binaries, then publishes it with gh release edit --draft=false. It is immutable from that point, with the assets attached.
  • Both crates to 0.4.1: the same code as 0.4.0. Cargo.lock's two member entries follow (the only lock change).

Tested: make ci; local make release-x86 / release-aarch64 compared against this PR's CI binaries. The draft-then-publish path only runs on a tag, so v0.4.1 is its test.

The same fix goes to stage0, stage1 and stage2 (their release jobs have the same draft: false) on their open PRs.

🤖 Generated with Claude Code

v0.4.0's release job published the release before uploading its assets
(action-gh-release, draft: false). An immutable release accepts no
uploads once published, so v0.4.0 is empty (its notes now say so). The
release is now created as a draft with the four binaries and published
by `gh release edit --draft=false`, after which it is immutable with
them attached.

Both crates to 0.4.1, same code as 0.4.0; Cargo.lock's member entries
follow.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@HarryR
HarryR marked this pull request as ready for review November 15, 2026 12:23
@HarryR
HarryR merged commit 157ec04 into main Nov 15, 2026
4 checks passed
@HarryR
HarryR deleted the release-draft-then-publish branch November 15, 2026 12:23
HarryR added a commit to lockboot/stage0 that referenced this pull request Nov 18, 2026
…ithout a debug record; publish scripts verify the release tag (#14)

Part of the GCP release changeset: with lockboot/stage1#22,
lockboot/stage2#5 and lockboot/workspace#6, on top of vaportpm v0.4.1
(lockboot/vaportpm#12).

**PCR replay**
- `predict-pcrs.py` marks stage0's EV_EFI_GPT_EVENT `"from_attestation":
"gpt"`, so workspace pcr-replay takes that event's digest from the
attestation and reproduces PCR5 (PCR2 on GCE TDX) on every boot, not
only before stage2 rewrites the GPT. `pcrs.json` stays the first-boot
reference.

**Reproducible build**
- The UEFI targets link with `/DEBUG:NONE`. CI's `stage0.efi`, `ena.efi`
and `gvnic.efi` each differed from a local build of the same commit in 8
bytes: the GUID of the linker's debug record, a hash of the `.pdb`,
which records where the toolchain is installed (`/tmp/.rustup` in CI,
`/src/.rustup` locally). The Authenticode hash and PCR4 differed with
it. No `.pdb` is written now and the image carries no debug record. This
changes the measured stage0 binary.
- `os-release`'s `yy.mm` version is the committer date (UTC) of the
commit being built, not the build date.

**Release and publish tooling**
- CI: the release job creates the release as a draft with its assets,
then publishes it. Publishing first leaves an immutable release empty,
as vaportpm v0.4.0 was (lockboot/vaportpm#14).
- `tools/publish/{ec2,gcp}/create.sh`, release path: `gh attestation
verify` gains `--source-ref refs/tags/<version>` (the provenance must
come from a build of that tag), and `gh release verify-asset` checks the
zip is an asset of the immutable release.
- Both scripts read `os-release` as data instead of sourcing it.

**Tested**
- Workspace `make selftest`, x86_64 and aarch64.
- CI's build equals the local build on both arches: PCR4 in `pcrs.json`
is identical, and the body of CI's signed `BOOT*.EFI` is byte-identical
to the local unsigned `stage0.efi` (PCR7 differs: the Secure Boot keys
are minted per build).
- GCP release gate (workspace `tools/cloudtest`, 2026-11-18, run
be236c1f6330) on n2d SEV, n2d SEV-SNP and c3 TDX, with the image built
from this branch: stage1's attestation verifies and every PCR replays
against the golden platform files on both payload boots. The gate ran
before the debug record was removed and is not re-run.
- The two `gh` checks against lockboot/vaportpm v0.4.1 (its tag passes;
another tag and refs/heads/main fail); the os-release parser on a good,
a hostile and an incomplete file.

**Not tested**
- `create.sh <arch> <tag>` end to end and the draft-then-publish release
job: both need this repo's first tag.

🤖 Generated with [Claude Code](https://claude.com/claude-code)
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