fix(publish): make allow_public actually reach crates.io

[loispostula]

The bug

allow_public inserted a registry named crates.io. Cargo does not accept that as a registry name, so the public publish was unreachable for every repo relying on it.

publish/mod.rs:1155 passes the registry name verbatim to cargo publish --registry, and cargo rejects it:

error: invalid character `.` in registry name: `crates.io`,
characters must be Unicode XID characters

Real consequence: opd-parser shipped 0.6.0 to the private registry while crates.io stayed at 0.5.0.

The fix

Canonicalize on cargo's own spelling, crates-io.

That makes the allow_public entry collapse with any crates-io already contributed by package.publish, instead of sitting beside it as a second, unusable entry. It also makes the pre-existing filter(|r| *r != "crates-io") at check_workspace/mod.rs:876 correct for both sources — that line already assumed this spelling.

Also defaults the sparse index for crates-io, since it has no [registries] entry to read one from (cargo knows the index natively). This sits after the env and config merges, so an explicit argument, CARGO_REGISTRIES_CRATES_IO_INDEX, or a config entry still wins.

Verification

Against real cargo, not just unit tests:

cargo publish --dry-run --registry crates-io                  # reaches Uploading
CARGO_REGISTRIES_CRATES_IO_INDEX=... _TOKEN=... \
  cargo publish --dry-run --registry crates-io                # also reaches Uploading

The second matters because get_registry_env injects exactly those vars, and cargo could plausibly have rejected an index override on its built-in registry. It does not.

243 tests pass, cargo fmt --check clean. Also verified green with CARGO_REGISTRIES_CRATES_IO_INDEX set to the real index, matching how CI runs.

Correction to my first description of this PR

I originally claimed a second, independent failure: that the missing index made check_crate_exists fail in production, and its caller swallowed the error into publish = false.

The CI failure on the first push disproved that. My test asserted the default index by constructing a registry through CargoRegistry::new, and it failed on Prow — because Prow does set CARGO_REGISTRIES_CRATES_IO_INDEX from vault, so the env supplied an index and the default never applied. Which means the registry very likely resolved an index in the real publish job too, and the existence check was not necessarily the thing that broke.

The illegal registry name is fatal on its own and is proven, so the fix stands unchanged. The index default is now a robustness improvement for environments that do not set that var, not a claimed root cause. The test was rewritten to assert default_index() directly rather than read ambient env.

Consumers need one more thing

crates-io must be listed in package.publish. A publish list naming only the private registry still blocks the upload — cargo's own allowlist, not this code. For opd-parser that is fslabs/opd_parser#17.

Unrelated, not fixed here

cargo clippy --all-targets -- -D warnings fails on generate_wix/mod.rs:917 (iter_kv_map). Confirmed pre-existing by stashing this change and re-running. Toolchain drift: this repo pins 1.93, my local cargo is 1.98. Left alone as out of scope.

[fslabs-bot]

[APPROVALNOTIFIER] This PR is APPROVED

This pull-request has been approved by: JulianRuiseco

The full list of commands accepted by this bot can be found here.

The pull request process is described here

Details

Needs approval from an approver in each of these files:

• OWNERS [JulianRuiseco]

Approvers can indicate their approval by writing /approve in a comment
Approvers can cancel approval by writing /approve cancel in a comment