Skip to content

fix(release): swap publish-ask / publish-crate ordering#85

Merged
joaoh82 merged 1 commit intomainfrom
fix/release-publish-order
May 3, 2026
Merged

fix(release): swap publish-ask / publish-crate ordering#85
joaoh82 merged 1 commit intomainfrom
fix/release-publish-order

Conversation

@joaoh82
Copy link
Copy Markdown
Owner

@joaoh82 joaoh82 commented May 3, 2026

Summary

The v0.2.0 release (Action run 25290833060) failed at `Publish sqlrite crate to crates.io`:

```
failed to select a version for the requirement `sqlrite-ask = "^0.2"`
candidate versions found which didn't match: 0.1.25, 0.1.24, ...
```

Root cause: the publish ordering reflected the pre-v0.1.19 dep direction (`sqlrite-ask → sqlrite-engine`) — but the v0.1.19 dep-direction-flip retrospective (roadmap.md L488) records that sqlrite-ask became pure-string-in/string-out and the engine took on the sqlrite-ask dependency (gated by the `ask` feature). Through 0.1.x this latent bug was masked because the engine's published version requirement was `^0.1` and old `sqlrite-ask 0.1.x` versions were already on crates.io. The v0.2.0 bump (`^0.2`) had no matching crate yet, so the publish failed loudly.

Fix

  • `publish-ask` now runs first (depends only on `detect` + `tag-all`). No internal SQLRite path-deps after the dep-direction flip.
  • `publish-crate` (engine) now waits on `publish-ask`. Comment updated to flag the latent bug + the retrospective.
  • `publish-mcp`'s `needs: [..., publish-crate, publish-ask]` is unchanged — sqlrite-mcp still depends on the engine.

Resulting chain: `detect → tag-all → publish-ask → publish-crate → publish-mcp`.

Path forward for v0.2.0

After this merges, two options to finish the v0.2.0 cut:

Option What Tradeoff
A. Fire `workflow_dispatch` with `version=0.2.0` Re-runs the whole release flow under the fixed ordering. The 3 Rust crate publishes succeed cleanly. Already-shipped publishes (Python wheels, npm packages, Go SDK Release) re-run; PyPI/npm reject duplicate-version uploads, so those jobs fail noisily but don't undo anything.
B. Local `cargo publish` From your machine: `cargo publish -p sqlrite-ask --no-verify && cargo publish -p sqlrite-engine --no-verify && cargo publish -p sqlrite-mcp --no-verify`. Manually upload the missing GitHub Release assets if needed. Fastest path to "v0.2.0 fully shipped". Skips the workflow's GitHub Release creation steps for those three crates; you'd add the tags manually.
C. Cut v0.2.1 Ship a "release plumbing fix" v0.2.1 right after this merges. The fixed workflow runs cleanly end-to-end. Wastes a version; pins users to v0.2.1 unnecessarily.

I'd recommend A — the duplicate-publish failures are noisy but harmless, and it keeps the release pipeline as the single source of truth for v0.2.0's ship state.

🤖 Generated with Claude Code

The v0.2.0 release failed at the `Publish sqlrite crate to crates.io`
step:

  failed to select a version for the requirement
  `sqlrite-ask = "^0.2"`
  candidate versions found which didn't match: 0.1.25, 0.1.24, ...

Root cause: the publish ordering reflected the **pre-v0.1.19** dep
direction (`sqlrite-ask → sqlrite-engine`) — but the v0.1.19
dep-direction-flip retrospective records that sqlrite-ask became a
pure-string-in/string-out adapter, and the engine took on the
sqlrite-ask dependency (gated by the engine's `ask` feature).

Through 0.1.x this latent bug was masked: the engine's published
version requirement was `^0.1`, and old `sqlrite-ask 0.1.x` versions
were already on crates.io, so the engine publish resolved cleanly.

At v0.2.0 the bumped requirement (`^0.2`) had no matching crate and
the publish failed loudly.

Fix:

- publish-ask now runs FIRST (depends only on detect + tag-all). No
  internal SQLRite path-deps after the dep-direction flip.
- publish-crate (engine) now waits on publish-ask. Comment updated
  to flag the latent bug + the v0.1.19 retrospective.
- publish-mcp's `needs: [..., publish-crate, publish-ask]` is
  unchanged — sqlrite-mcp still depends on the engine.

Resulting chain: detect → tag-all → publish-ask → publish-crate →
publish-mcp.

This unblocks the v0.2.0 cut once re-triggered via workflow_dispatch
(or via cutting v0.2.1 if the partial v0.2.0 state warrants a fresh
release rather than catching up).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
@joaoh82 joaoh82 merged commit 80664b8 into main May 3, 2026
15 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