Snapshots

Build locally without publishing

Snapshot mode runs the full build and archive pipeline but skips all publishing stages.

Classification

Not applicable — this is a workflow page, not a publisher. Snapshot mode disables every publisher in the pipeline (including Submitters); the only outputs are local artifacts under dist/.

Minimal config

anodizer release --snapshot

No YAML changes required for the default behavior.

Full config reference

snapshot:
  version_template: "{{ Version }}-SNAPSHOT-{{ ShortCommit }}"  # optional; version suffix (alias: name_template)

Authentication

Not applicable — snapshot mode never contacts external services. No tokens are read or required.

Common gotchas

The default template appends -SNAPSHOT to the version; override via snapshot.version_template (or its deprecated alias name_template).
--auto-snapshot engages snapshot mode whenever the git repo has uncommitted changes — useful for safety in CI.
Required publishers are silently skipped; snapshots never publish regardless of the required flag.

Auto-snapshot

Automatically enable snapshot mode when the git repo has uncommitted changes:

anodizer release --auto-snapshot

Emission validation

Some publishers don't push an asset — they emit a reference to one: the binstall [package.metadata.binstall] block names a download URL, the Nix derivation maps each packages.<system> to an asset, version-sync writes a crate version. A real release mutates source files or pushes to a remote; a snapshot/dry-run skips those side effects. Historically that meant a broken emission — a binstall URL pointing at an asset the release never produces, a nix system mapped to a missing asset — passed every local check and only blew up later at cargo binstall / nix build time on a consumer's machine.

Snapshot/dry-run closes that blindspot. anodizer renders the would-be emission in-memory (never mutating source, never cloning, never pushing) and cross-checks it against the asset set the run actually produced. A mismatch fails the snapshot loud, naming the crate, the emission, and what's wrong:

$ anodizer release --snapshot --host-targets
error: binstall emission for crate 'myapp' references an asset the release does
       not produce: myapp-1.2.3-linux-amd64.tar.gz (produced: myapp-1.2.3-x86_64-unknown-linux-gnu.tar.gz)

This runs per in-scope crate with that crate's own version/name/tag scope, so it's correct in single-crate, workspace-lockstep, per-crate, and --all modes. It is the local gate that catches the "release succeeds but is silently wrong" class before you push.

Two local gates: slim `snapshot` vs full `prepush`

anodizer's own Taskfile.yml wires two snapshot gates with different cost/coverage trade-offs. A consumer project can mirror the same shape:

Task	What it runs	When
`task snapshot`	dry-run pipeline, `--single-target`, no real build — fast	wired into the commit hook (via `task lint` → `task commit`)
`task prepush`	real build of every host-buildable target (`--host-targets`) + archive/sign/checksum + emission validation	before pushing

# Slim, fast — compiles nothing, single target. Runs on every commit.
$ anodizer release --snapshot --single-target --clean --dry-run

# Full — real host-scoped build of everything this host can compile, plus the
# emission-validate cross-check above. Slower; run before pushing.
$ anodizer release --snapshot --clean --host-targets

The slim gate keeps commits fast; the full gate is the one that actually compiles artifacts and validates binstall/nix/version-sync emissions against them. See --host-targets for how the host-scoped target set is computed (apple targets are skipped off a non-macOS host, logged loudly).