chore: fix repo link (closed, accidental PR) by eugencowie · Pull Request #37 · aptabase/tauri-plugin-aptabase

eugencowie · 2026-10-13T12:13:50Z

Opened against wrong repo, sorry!

Framework-independent Rust SDK for constructing an Aptabase client, enqueueing analytics events, managing sessions and metadata, selecting hosted or self-hosted endpoints, and delivering queued events through asynchronous Tokio-based batching and flushing.

## 🤖 New release * `aptabase-rs`: 0.1.0 <details><summary>Changelog</summary> </details> --- This PR was generated with [release-plz](https://github.com/release-plz/release-plz/).

## 🤖 New release * `aptabase-rs`: 0.1.0 -> 0.1.1 (✓ API compatible changes) <details><summary>Changelog</summary> </details> --- This PR was generated with [release-plz](https://github.com/release-plz/release-plz/).

## 🤖 New release * `aptabase-rs`: 0.1.1 -> 0.1.2 <details><summary>Changelog</summary> </details> --- This PR was generated with [release-plz](https://github.com/release-plz/release-plz/).

Define caller-owned session ID support for applications that need to persist or reuse Aptabase sessions across client construction.  ## Summary by CodeRabbit * **New Features** * Added support for supplying a custom session ID when initialising the client. * Exposed a helper to generate a reusable session ID for app-managed persistence. * Session IDs now continue to rotate automatically when not supplied. * **Documentation** * Expanded the README with guidance and an example for persisting and reusing session IDs across restarts. * **Bug Fixes** * Empty session IDs are now treated as missing, so the SDK falls back to automatic generation.

Bumps [tokio](https://github.com/tokio-rs/tokio) from 1.53.1 to 1.53.2. - [Release notes](https://github.com/tokio-rs/tokio/releases) - [Commits](tokio-rs/tokio@tokio-1.53.1...tokio-1.53.2) --- updated-dependencies: - dependency-name: tokio dependency-version: 1.53.2 dependency-type: direct:production update-type: version-update:semver-patch ... Signed-off-by: dependabot[bot] <support@github.com>

mise's native mr-boxington integration wraps cargo in this project, so builds and tests reuse compiler outputs across target directories and worktrees. The Rust MSRV pin is unchanged.

The action caches the same pruned target/ and registry as rust-cache, keyed on the rustc in use, and adds native link caching. It saves only from pushes to main, so PRs stop adding entries to an Actions cache that is already over its 10 GB limit. The example job gets its own cache generation so it never restores the root crate's target/.

## Summary Cargo runs through mbx (mr-boxington) in this project, so builds and tests reuse compiler outputs across target directories and worktrees. ```diff [tools] +mr-boxington = "latest" -rust = { version = "1.88", components = ["clippy", "rustfmt"] } # the MSRV +rust = { version = "1.88", components = ["clippy", "rustfmt"], mr_boxington = true } # the MSRV ``` ```text cargo build / mise run test mise command wrapper (cargo) mbx # cache lookup per compiler action cargo 1.88 # MSRV pin unchanged ``` CI swaps rust-cache for mr-boxington-action, which caches the pruned `target/` and registry keyed on the rustc in use, adds native link caching, and saves only from pushes to main. ```diff check / check-example mise-action - install_args: rust … + install_args: rust mr-boxington … - Swatinem/rust-cache + jdx/mr-boxington-action # check-example: own working-directory and cache-generation mise run check / example:check ``` ## Evidence Plain `cargo build` into two fresh target directories, same toolchain, profile and features: - **Before (fresh dir 1, cold):** 0 hits, 505 not looked up, 12.5s **After (fresh dir 2):** 324 hits, 179 misses, 6.1s (568 outputs restored) All 179 misses are C objects from aws-lc's build script that embed absolute target paths (`mbx explain --last`); they hit when the path is the same. `mbx doctor`: 0 failures. `mise run test`: 11 tests pass, 312 hits. `jactionlint` passes. CI on this PR starts cold under the new cache keys; the first warm run is the first PR after this merges to main. ## Merge Danger **Door:** two-way Revert the commits, or set `mr_boxington = false` and run `mise reshim`. Old rust-cache entries age out of the Actions cache on their own. **Blast Radius:** tooling Changes how `cargo` is invoked for anyone using this repo's mise config, and which Actions cache entries CI reads and writes. PRs no longer save cache entries. No crate code or published artifacts change.

Cargo runs through mise's mbx wrapper, so without mr-boxington in install_args mise installs it on demand partway through cargo publish. Install it with the other tools instead.

Cargo runs through mise's mbx wrapper, so the release jobs already build through mbx. Without `mr-boxington` in `install_args`, mise installs it on demand partway through `cargo publish`. This installs it in the mise-action setup step alongside the other tools, so it's installed before release-plz runs and mise-action caches it. Follow-up to #31.

The release-pr job rebuilt the changelog with git-cliff-action and set the version from its output, while `mise run release` ran `release-plz update`. That leaves CHANGELOG.md alone (changelog_update = false) and picks the version with release-plz's own rules over every commit, upstream ones included, so a local run could pick a different version than the release PR. And `mise run changelog` alone left CHANGELOG.md a version ahead of Cargo.toml. - release:prepare replaces the changelog, changelog:build-ignore-list and release tasks. It writes .cliffignore, updates CHANGELOG.md with `git cliff --bump`, and sets Cargo.toml to git-cliff's bumped version - the release-pr job runs it in place of the ignore-list step, git-cliff-action and `set-version`, so CI uses mise.lock's git-cliff rather than the action's own download - cliff.toml no longer sets `output`, so git-cliff writes CHANGELOG.md only when told to with `-o`. Before, every git-cliff run overwrote it, even `--bumped-version`

release-plz/action reinstalled release-plz with cargo-binstall on every run, though mise had already installed mise.lock's version. Both jobs now call it directly, with GIT_TOKEN for the token and the repo URL and forge inferred from the checkout's remote. Of what else the action did: - its git identity step stays, as release-plz/git-config with gh from mise.lock, but only in release-pr: release-plz commits locally when it updates an existing release PR, while release creates tags and releases through the API - release-pr no longer gets CARGO_REGISTRY_TOKEN, as it never publishes The action also installed cargo-semver-checks, which isn't carried over as it never did anything here: release-plz skips its check because the release-pr job has already set the new version, and cargo-semver-checks 0.51 needs Rust 1.93 to run, above our pinned 1.88.

Release PR and GitHub release bodies have always had an empty changelog. git-cliff writes CHANGELOG.md, so release-plz runs with changelog_update = false, and with that off it builds no changelog text for either body. Both now get the release's `## <version>` section of CHANGELOG.md, printed by a new release:notes task, which fails if the section is missing rather than posting an empty changelog. - release-pr: release-plz's default PR body renders an empty Changelog <details>, as its `{% if changes %}` sees leftover whitespace. The job reads the PR number and version from release-plz's JSON and puts the changelog before the block's closing </details>, failing if a release-plz upgrade drops it - release-notes.yml: a new workflow runs on each published release and sets its notes, so a failure can be retried with a rerun, which the release job couldn't do once release-plz had published. It uses GitHub's own token with contents: write, and only fires because release-plz creates releases with RELEASE_PLZ_TOKEN, as its comment notes. It takes the version from the tag, so needn't install Rust

## Summary Release PR and GitHub release bodies had an empty changelog: git-cliff writes `CHANGELOG.md`, so release-plz runs with `changelog_update = false` and has no changelog text for either. Both now get the release's `## <version>` section of `CHANGELOG.md`. ```diff mise.toml tasks -changelog # git cliff --bump (left Cargo.toml a version behind) -changelog:build-ignore-list -release # release-plz update (its own version rules) +release:prepare # .cliffignore + git cliff --bump + set-version +release:notes <version> # prints a version's CHANGELOG.md section, fails if missing release.yml release # on push to main - release-plz/action release + release-plz release # mise.lock's release-plz, no reinstall release-pr - mise run changelog:build-ignore-list - git-cliff-action --bump - release-plz set-version - release-plz/action release-pr + mise run release:prepare + release-plz release-pr -o json + gh pr edit # notes into the default body's empty Changelog <details> +release-notes.yml # on release published, so a failure can be rerun + gh release edit --notes "$(mise run release:notes <tag>)" ``` - `cliff.toml` no longer sets `output`, so git-cliff only writes `CHANGELOG.md` when told to with `-o`. Before, even `git cliff --bumped-version` overwrote it. - release-plz/action's cargo-semver-checks isn't carried over: release-plz skips the check once `release:prepare` has set the version, and 0.51 needs Rust 1.93 against our pinned 1.88. ## Evidence - **Before:** the 0.3.0 release PR (#13) has an empty `<details><summary>Changelog</summary>` block, and `gh release view v0.3.0` has an empty body. **After:** filling #13's real body with `mise run release:notes 0.3.0`, rendered through GitHub's Markdown API, shows the Features / Documentation / Miscellaneous Tasks lists inside the Changelog block (neither the PR nor the release was edited). - **Before:** `mise run release` set the version by release-plz's rules, and `mise run changelog` alone left `CHANGELOG.md` at a version Cargo.toml didn't have. **After:** in a scratch clone of main, `mise run release:prepare` writes `## 0.3.1` and sets Cargo.toml to 0.3.1, matching `git cliff --bumped-version`. - `release:notes 9.9.9` and `release:notes` with no version both exit 1. `release-plz release --dry-run -o json` against this repo infers the forge from the remote and returns `{"releases":[]}`. - `mise run check:workflows` (jactionlint + shellcheck) passes on each commit. ## Merge Danger **Door:** two-way CI and task config only. Reverting restores the old workflow and tasks. **Blast Radius:** releases - Merging runs release-pr straight away, which will open or update the 0.3.1 release PR with the changelog filled in. That's the first real test of `release-plz release-pr -o json`, the git-config identity, and the body fill. - `release-notes.yml` first runs on the next published release. It only fires because release-plz creates releases with `RELEASE_PLZ_TOKEN`; GitHub's own token wouldn't trigger it. - 0.3.0 and earlier keep their empty notes. Backfill by hand: `gh release edit v0.3.0 --notes "$(mise run release:notes 0.3.0)"`. - `mise run release` and `mise run changelog` no longer exist locally; use `mise run release:prepare`.

## 🤖 New release * `aptabase-rs`: 0.3.0 -> 0.3.1 <details><summary>Changelog</summary> ### 🚜 Refactor - Move fork-only code out of upstream-shared files ([#25](#25)) ### 🧪 Testing - Test event delivery against a local stand-in server ([#28](#28)) ### ⚙️ Miscellaneous Tasks - Stop tracking the example's lockfile ([#23](#23)) - Match upstream's Rust edition and require Rust 1.88 ([#21](#21)) - Stop mise run from installing unused tools in CI ([#26](#26)) - Pin release-plz and git-cliff to mise.lock versions ([#27](#27)) - Switch to jactionlint and harden the workflows ([#29](#29)) - Cache builds with mbx ([#31](#31)) - Install mbx for release jobs ([#32](#32)) - Add the changelog to release PR and GitHub release bodies ([#33](#33)) </details> --- This PR was generated with [release-plz](https://github.com/release-plz/release-plz/).

Both release jobs shared one mise-action step through an anchor, so the release job installed git-cliff and gh it never runs. Each job now has its own step: prepare-release keeps all five tools, and release installs only rust, mr-boxington and release-plz, which cargo publish needs. release-plz creates the GitHub release through the API, and with changelog_update = false it doesn't use git-cliff. - the release-pr job is now prepare-release, and runs before release - the checkout anchor goes too, as the two jobs no longer share steps - one-line with and env blocks use flow style, and env follows run, to match ci.yml and release-notes.yml - the PR-body script checks for the Changelog <details> before reading the notes

## Summary The two release jobs shared one mise-action step through a YAML anchor, so `release` installed tools it never runs. Each job now has its own step. ```diff release.yml - release-pr # "Release PR" + prepare-release # "Prepare release", now listed first install_args: rust mr-boxington release-plz git-cliff gh release - install_args: rust mr-boxington release-plz git-cliff gh # *install-tools + install_args: rust mr-boxington release-plz ``` - `release` runs only `release-plz release`, which needs cargo for `cargo publish`, so it needs `rust` and `mr-boxington`. release-plz creates the GitHub release through the API itself, so it doesn't need `gh`. With `changelog_update = false` it doesn't need `git-cliff` either. - `prepare-release` uses all five tools. `release:prepare` runs `git cliff` and `release-plz set-version`, and `release-plz release-pr` runs cargo. The PR-body fill uses `gh pr view` and `gh pr edit`. - The `checkout` anchor is gone too, and both checkout steps are written out. - Style now matches `ci.yml`: single-line `with`/`env` blocks use flow style, and `env` comes after `run` in all three workflows. - The PR-body script now checks for the Changelog `</details>` before it reads the notes. - Three comments are removed: - `release-notes.yml`: why this workflow needs `RELEASE_PLZ_TOKEN` - `release.yml`: why the `release-plz/git-config` step is there - `release.yml`: the "Fill the empty Changelog" note above the PR-body script ## Evidence - `mise run check:workflows` (jactionlint) passes. - No `&install-tools` or `*install-tools` remains in the repo. - The `release-plz.toml` comment now refers to the `prepare-release` job. ## Merge Danger **Door:** two-way Only workflow config changes. Reverting restores the shared step. **Blast Radius:** releases - If `release-plz release` needs a tool that's no longer installed, `release` fails on the next merge that publishes. `MISE_TASK_RUN_AUTO_INSTALL: false` only covers `mise run`, and `mise run` isn't used in that job. A missing tool would show up as an error or an on-demand install. - The check name changes from "Release PR" to "Prepare release". It only runs on push to `main`, so no PR status check refers to it.

Steps pinned to a SHA showed up in the Actions UI as "Run jdx/mise-action@2d8d4caf..." or as the first line of their script. Each step now has a name, and with/env blocks use block style so the ${{ }} expressions no longer need quoting.

## Summary Steps pinned to a SHA showed in the Actions UI as "Run jdx/mise-action@2d8d4caf…", and `run:` steps showed their first script line. Every step now has a name, and `with`/`env` use block style. ```diff steps: - - uses: jdx/mise-action@2d8d4caf… # v5.1.1 - with: { install_args: rust mr-boxington } + - name: Install tools + uses: jdx/mise-action@2d8d4caf… # v5.1.1 + with: + install_args: rust mr-boxington + ``` | Step | Name | |---|---| | `actions/checkout` | Checkout | | `jdx/mise-action` | Install tools | | `jdx/mr-boxington-action` | Set up build cache | | `mise run check` / `example:check` | Run checks | | `release-plz/git-config` | Configure git | | `mise run release:prepare` | Update changelog and version | | `release-plz release-pr` and changelog fill | Create or update release PR | | `release-plz release` | Publish release | | `gh release edit` | Set release notes | - This reverses the flow-style rule from c751fd7. Named steps are multi-line anyway, so the single-line `with`/`env` no longer saved much. - Block style means `${{ }}` values no longer need quoting. - A blank line now separates steps. ## Evidence - **Before:** steps listed by action SHA or by the first script line. **After:** every step named. `yq '[.jobs[].steps[] | select(has("name") | not)] | length'` gives `0` for all three workflows. - With step names removed, each workflow parses to identical JSON on `main` and on this branch (`yq -o=json 'del(.jobs[].steps[].name)'`). So actions, SHAs, inputs, env and scripts are unchanged. - `mise run check:workflows` (jactionlint) passes. ## Merge Danger **Door:** two-way Only formatting and step names change. Reverting restores the previous files. **Blast Radius:** CI - The workflows behave the same, as the parse comparison shows. Only the step labels in the Actions UI change. - Job names are unchanged, so required status checks are unaffected.

mise 2026.10 writes lockfile_version = 3, which records each tool's version specifier from mise.toml and the GitHub repository ID of mr-boxington's signed releases. Tool versions, URLs and checksums are unchanged.

## Summary mise 2026.10 writes `lockfile_version = 3`. It adds metadata only. Tool versions, URLs and checksums stay the same. ```diff +lockfile_version = 3 + [[tools.gh]] version = "2.96.0" backend = "aqua:cli/cli" +specifiers = ["latest"] # every tool: its version from mise.toml [tools.mr-boxington."platforms.linux-x64"] checksum = "sha256:4177…" signer = "sigstore-oidc:…/release.yml" +repository_ids = { repository = "1340721957" } # mr-boxington platforms only ``` ## Evidence - The diff is 15 insertions and 0 deletions, so no `version`, `url` or `checksum` line changes. - Running `mise lock` (mise 2026.10.3) on this branch leaves `mise.lock` byte-identical. ## Merge Danger **Door:** two-way Reverting restores the old format. mise rewrites it on the next `mise lock` anyway. **Blast Radius:** tooling - Older mise versions may not read lockfile v3. CI's `mise-action` installs a current mise, so CI is unaffected. Contributors on an older mise may need to update it.

The mise local config, agent/editor, Windows, macOS and Linux sections ignore files that come from each contributor's setup rather than from the project. Those belong in a global gitignore. The Rust, dotenv, examples and git-cliff patterns stay.

## Summary `.gitignore` now covers only what the project itself generates. Patterns for each contributor's own setup belong in their global gitignore. ```diff .gitignore Rust # target, *.rs.bk, *.pdb, mutants.out, rustc-ice dotenv # .env Examples # examples/*/Cargo.lock git-cliff # .cliffignore - mise # mise.local.toml and variants - Agents # .claude/, CLAUDE.md, .cursor/, .codex/, .aider*, … - Windows # Thumbs.db, desktop.ini, … - macOS # .DS_Store, ._*, … - Linux # *~, .directory, .Trash-*, … ``` ## Evidence - `git status --untracked-files=all` in this checkout shows no new untracked files after the change. ## Merge Danger **Door:** two-way Reverting restores the removed sections. **Blast Radius:** contributors - Contributors without a global gitignore may see local files such as `.DS_Store`, `mise.local.toml` or `.claude/` as untracked and could commit them by accident.

.cliffignore already removes branch commits and main only takes PR merges, so the PR-number filter is redundant. Dropping it fixes scoped commits being silently skipped and unrecognised types landing under Security, and adds an Other group so unknown types still show up.

## Summary Uses the same `cliff.toml` as my other projects (only the repo name in the PR links differs). Dropping the PR-number filter fixes two bugs where commits were silently dropped or put under the wrong heading. ```diff [changelog] -postprocessors = [ (#N) -> PR link ] [git] -filter_commits = true tag_pattern = "^v[0-9]" commit_parsers = [ - "^feat!?: .+ $#[0-9]+$" -> Features # no scope allowed + "^feat" -> Features ...same for fix/doc/perf/refactor/style/test/chore|ci/revert - message "^[a-z]+!?: .+ $#N$" + body ".*security" -> Security + body ".*security" -> Security + ".*" -> Other ] +commit_preprocessors = [ (#N) -> PR link ] [bump] breaking_always_bump_major = false ``` The PR-number filter isn't needed. `.cliffignore` (from `mise run release:prepare`) already removes branch commits, and main's ruleset only takes PR merges, so every commit left already ends in `(#N)`. ## Evidence Test commits run through `git cliff --unreleased` with the old and new config: | Commit | Before | After | |---|---|---| | `feat(cli): Scoped feature (#1000)` | dropped | Features, `*(cli)* Scoped feature` | | `build: Unknown type (#1001)` | 🛡️ Security | 💼 Other | - **Before:** the scope regex doesn't allow `(cli)`, and git-cliff ignored the Security rule's `body` condition once `message` matched. **After:** both are handled as shown above. - Regenerating the whole changelog with `.cliffignore` leaves every released section of `CHANGELOG.md` unchanged. ## Merge Danger **Door:** two-way Changelog config only. Reverting restores the old parsers. **Blast Radius:** changelog - The next release PR's changelog will include PRs with scoped or non-standard types that the old config dropped. None exist in the history so far. - Running plain `git cliff` without regenerating `.cliffignore` would now list branch commits. `mise run release:prepare` always regenerates it.

eugencowie added 30 commits June 11, 2026 20:24

Configure mise

953669d

Add lock file

89faa7f

Configure openspec

8426cd0

Implement generic client

3b3c8bf

Tidy up changes

3013e13

Add AI disclosure

1053255

Configure Dependabot

f72dd6b

Configure CI workflow

3fe41a2

Configure release workflow

9fe211f

ci: Configure release workflow (#2)

20ae243

Update readme

8a3e01c

Fix changelog

aa22723

docs: Update readme (#4)

1bd8cc4

chore(release): Release 0.1.0

1db10f5

chore(release): Release 0.1.0 (#3)

e4237ea

## 🤖 New release * `aptabase-rs`: 0.1.0 <details><summary>Changelog</summary> </details> --- This PR was generated with [release-plz](https://github.com/release-plz/release-plz/).

Fix release workflow

bba2e9a

Update readme

83eab90

ci: Fix release workflow (#5)

78a0f52

chore(release): Release 0.1.1

8ec1972

chore(release): Release 0.1.1 (#6)

fe744f3

## 🤖 New release * `aptabase-rs`: 0.1.0 -> 0.1.1 (✓ API compatible changes) <details><summary>Changelog</summary> </details> --- This PR was generated with [release-plz](https://github.com/release-plz/release-plz/).

Fix build error in example

043a788

fix: Fix build error in example (#7)

27b07b4

Tidy up readme

eb3d50e

docs: Tidy up readme (#9)

0c6a03b

chore(release): Release 0.1.2

acad9fd

chore(release): Release 0.1.2 (#8)

107446a

## 🤖 New release * `aptabase-rs`: 0.1.1 -> 0.1.2 <details><summary>Changelog</summary> </details> --- This PR was generated with [release-plz](https://github.com/release-plz/release-plz/).

Add custom session ID support

0dd0427

Fix failing test

5ee8904

dependabot Bot and others added 24 commits October 11, 2026 15:26

Cache Rust builds with mbx

8b71e97

mise's native mr-boxington integration wraps cargo in this project, so builds and tests reuse compiler outputs across target directories and worktrees. The Rust MSRV pin is unchanged.

Install mbx up front in release jobs

da7f65b

Cargo runs through mise's mbx wrapper, so without mr-boxington in install_args mise installs it on demand partway through cargo publish. Install it with the other tools instead.

chore(release): Release 0.3.1

381d5d8

Regenerate mise.lock in lockfile format v3

a900ec6

mise 2026.10 writes lockfile_version = 3, which records each tool's version specifier from mise.toml and the GitHub repository ID of mr-boxington's signed releases. Tool versions, URLs and checksums are unchanged.

Fix repo link

5a332a1

eugencowie closed this Oct 13, 2026

eugencowie changed the title ~~chore: Fix repo link~~ chore: fix repo link (closed, wrong repo) Oct 13, 2026

eugencowie deleted the chore/fix-repo-link branch October 13, 2026 12:15

eugencowie restored the chore/fix-repo-link branch October 13, 2026 12:18

eugencowie deleted the chore/fix-repo-link branch October 13, 2026 12:19

eugencowie changed the title ~~chore: fix repo link (closed, wrong repo)~~ chore: fix repo link (closed, accidental PR) Oct 13, 2026

Navigation Menu

Repository navigation

chore: fix repo link (closed, accidental PR) - #37

chore: fix repo link (closed, accidental PR)#37
eugencowie wants to merge 122 commits into
aptabase:mainfrom
eugencowie:chore/fix-repo-link

eugencowie commented Oct 13, 2026 •

edited

Loading

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

eugencowie commented Oct 13, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

eugencowie commented Oct 13, 2026 •

edited

Loading