fix(bindeps): load targets before downloading artifacts

[npmccallum]

Artifact dependencies can introduce additional compile targets through manifest target fields. Cargo loaded those targets during feature resolution for some paths, but PackageSet::download_accessible could encounter a transitive artifact build-dependency before the target data had been populated.

When that happened, dependency platform filtering asked for cfg data for an unknown target and panicked. Load artifact dependency targets during the same traversal that follows them for downloads, so target cfg data is available before filtering recurses into that target.

The regression covers a git dependency whose build script has a path artifact build-dependency with an explicit target, matching the topology reported in the issue.

Closes #16881.

[rustbot]

r? @epage

rustbot has assigned @epage.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

Why was this reviewer chosen?

The reviewer was selected based on:

• Owners of files modified in this PR: @ehuss, @epage, @weihanglo
• @ehuss, @epage, @weihanglo expanded to ehuss, epage, weihanglo
• Random selection from ehuss, epage, weihanglo

[epage]

Only looked at the PR structure. I'll need to look into this on a desktop. Other reviewers are free to look at this if they get to it before me.

View changes since this review

[epage]

See https://epage.github.io/dev/pr-style/#c-test

[rustbot]

This PR was rebased onto a different master commit. Here's a range-diff highlighting what actually changed.

Rebasing is a normal part of keeping PRs up to date, so no action is needed—this note is just to help reviewers.

[npmccallum]

Thanks, addressed the structure feedback by splitting this into two commits:

1. test(bindeps): cover transitive artifact target download
2. fix(bindeps): load targets before downloading artifacts

The test-only commit passes locally, and the final commit keeps the production change separate.

Verified:

cargo test --test testsuite artifact_dep::transitive_build_script_artifact_dependency_with_valid_target_does_not_panic -- --nocapture
cargo fmt --check
git diff --check upstream/master..issue-16881-bindeps-panic

[epage]

The test-only commit passes locally, and the final commit keeps the production change separate.

If that is the case, then why does the fix not change the tests?

Also, what does git diff --checked have to do with this?