Support multiple artifact dependency aliases to the same package by npmccallum · Pull Request #17067 · rust-lang/cargo

npmccallum · 2026-06-03T14:55:35Z

This fills in behavior described by RFC 3176 that was not implemented when artifact dependencies were initially added.

The RFC allows a package to be declared multiple times under different dependency names, with each declaration selecting a different artifact target. This PR treats each artifact dependency declaration as an independent identity across unit construction, feature resolution, Cargo tree output, environment variables, unit graphs, and metadata. This allows one package to supply artifacts under multiple aliases or for multiple targets without collapsing their names or feature sets.

Design change: dependency-level `lib`

This expands the original scope by removing the dependency-level lib option. Preserving lib = true is not coherent once one package may appear through multiple artifact declarations: those declarations can have different aliases, targets, and feature selections, while the package library edge has only one dependency identity. Selecting one artifact declaration as the library dependency would be ambiguous, and combining them would collapse the independent feature graphs that this change requires.

Users that need both the library and an artifact must declare a normal dependency and a separately named artifact dependency. Those declarations retain independent feature graphs. Cargo rejects the dependency-level lib key whether it is true or false. This is a deliberate design change to the unstable artifact-dependencies feature. The historical registry-index field is retained for compatibility but no longer makes an artifact declaration double as a library dependency.

The series is split into reviewable commits:

Record the current artifact-alias conflict behavior.
Record the current metadata alias behavior.
Extract the dependency-name calculation in the resolver.
Implement independent artifact dependency identities and reject the dependency-level lib key.
Document independent aliases and the explicit normal-plus-artifact dependency form.

The rebased series was reviewed as a whole and every commit independently passed formatting, the Cargo build and test suite, the remaining workspace tests, and resolver-tests.

Closes #12374

FYI @joshtriplett @Byron

rustbot · 2026-06-03T14:55:41Z

r? @weihanglo

rustbot has assigned @weihanglo.
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

npmccallum · 2026-06-24T17:23:00Z

@weihanglo are you able to review this? If not, maybe @ehuss could?

weihanglo

Thanks for the contribution. Unfortunately, I may not have sufficient bandwidth to do meaningful reviews within this week or next week. Feel free to reassign.

View changes since this review

npmccallum · 2026-06-24T20:06:26Z

@rustbot reroll

epage · 2026-06-24T20:14:13Z

Note that in general, there is a lot less review bandwidth at this time. Weihang has the most bandwidth and he said he wouldn't be available. We'll see when/if I have time.

epage

Only had a quick glance at the tests.

The fixes seem to be independent. Can you split them up into smaller commits / reviewable unit?

View changes since this review

epage · 2026-06-24T20:17:49Z

+                authors = []
+                resolver = "2"
+
+                [dependencies.bar]


Is this mimicking existing tests? I at least would prefer we only use inline tables for deps, not standard tables.

epage · 2026-06-24T20:20:54Z

+
+    p.cargo("build --unit-graph -Zunstable-options -Z bindeps")
+        .masquerade_as_nightly_cargo(&["unit-graph", "bindeps"])
+        .with_stdout_contains(r#"[..]"index":0,"extern_crate_name":"bar"[..]"#)


Is this mimicking an existing pattern? We try to avoid the _contains functions

npmccallum · 2026-06-26T16:58:20Z

I split this into smaller commits:

current-behavior tests
resolver naming helper
build/unit-graph behavior
metadata behavior
docs

I also adjusted the tests from the inline comments to use inline dependency tables and a structured unit-graph stdout assertion.

weihanglo · 2026-06-26T20:02:44Z

May have time for this, though let me pick up the RFC first.

npmccallum · 2026-07-04T13:12:14Z

@joshtriplett This is my next highest priority if you have cycles. With this feature, arma can generate PMIs cross-architecture.

epage · 2026-07-22T18:09:28Z

+            .unwrap_or_else(|| (to_target.crate_name().into(), None))
+    }
+
+    pub fn lib_dependency_name<'a>(


This appears to be dead code within this commit

View changes since the review

epage · 2026-07-22T18:09:40Z

+        Ok((extern_crate_name, dep_name))
+    }
+
+    pub fn dep_extern_crate_name_and_dep_name(


Doesn't look like this needs to be pub within this commit

View changes since the review

epage · 2026-07-22T18:12:31Z

-                [build-dependencies.bar]
-                path = "bar/"
-                artifact = "bin"
-                target = "{}"
-
-                [build-dependencies.bar-native]
-                package = "bar"
-                path = "bar/"
-                artifact = "bin"
-                target = "{}"
+                [build-dependencies]
+                bar = {{ path = "bar/", artifact = "bin", target = "{}" }}
+                bar-native = {{ package = "bar", path = "bar/", artifact = "bin", target = "{}" }}


Was this style change made in the wrong commit? It makes it harder to see what may be intentional changes fthat go with this fix

View changes since the review

epage · 2026-07-22T18:13:32Z

-        .with_status(101)
-        .with_stderr_data(str![[r#"
-[LOCKING] 1 package to latest compatible version
-[ERROR] the crate `foo v0.0.0 ([ROOT]/foo)` depends on crate `bar v0.5.0 ([ROOT]/foo/bar)` multiple times with different names
-
-"#]])
+        .with_stderr_contains(
+            "[RUNNING] `rustc --crate-name bar --edition=2015 bar/src/main.rs [..]--target [ALT_TARGET] [..]",
+        )


What is the goal with using with_stderr_contains with this line of verbose output?

For context, we try to minimize the use of _contains assertions because we can't do snapshotting with them

View changes since the review

epage · 2026-07-22T18:13:56Z

+#[cargo_test]
+fn multiple_artifact_aliases_no_op_rebuild() {


Why is a new test being added in a non-test commit?

View changes since the review

epage · 2026-07-22T18:15:17Z

+        .with_stdout_contains(r#"[..]"rename":"bar-lib"[..]"#)
+        .with_stdout_contains(r#"[..]"rename":"bar-bin"[..]"#)
+        .with_stdout_contains(r#"[..]"extern_name":"bar_lib"[..]"#)
+        .with_stdout_contains(r#"[..]"extern_name":"bar_bin"[..]"#)


The lack of context is another problem with _contains assertions: I was looking to the test to better understand a part of your change and I can't tell what is happening from this

View changes since the review

Record the duplicate-name diagnostics produced when one package is requested through multiple artifact aliases. Cover build and normal dependencies, binary and static-library artifacts, same- and cross-target aliases, and the `--unit-graph` entry point. The snapshots capture how dependency-name validation collapses distinct artifact requests onto one package edge.

Record how `cargo metadata` resolves multiple declarations for one package. A single artifact alias retains the package-edge name, while an artifact and an ordinary dependency to a bin-only package can coexist because there is no library edge to collide. Complete JSON snapshots keep manifest declarations, package-level resolve edges, and dep-kind extern names visible together. Rejection coverage captures multiple artifact aliases and mixed artifact and ordinary dependencies when the package provides a library.

Move the per-dependency extern-crate and manifest-name calculation into a private helper while preserving the existing duplicate-name validation and behavior.

Treat each artifact dependency declaration as a distinct dependency identity when constructing units, resolving features, and emitting metadata. This allows one package to supply artifacts under multiple aliases or for multiple targets without collapsing their names or feature sets. An artifact dependency no longer doubles as a Rust library dependency. Reject the dependency-level `lib` key; users that need both forms can declare a normal dependency and an artifact dependency. Keep those declarations separate so artifact features do not unify with each other or with the library dependency. Carry alias-specific names through artifact units, unit graphs, Cargo tree nodes, environment variables, and metadata. Normal dependency names continue to determine library edges, while artifact dep-kind entries report each alias.

rustbot · 2026-07-27T15:14:18Z

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 · 2026-07-27T15:14:36Z

This update expands the scope of the PR in one deliberate way: dependency-level lib = true is no longer supported.

While implementing independent artifact aliases, it became clear that retaining it would leave the library edge ambiguously attached to one of several aliases, targets, and feature sets—or would require unifying declarations that must remain independent. The revised design requires an explicit normal dependency when the library is also needed.

I have made this design change explicit in the implementation, tests, documentation, commit message, and updated PR description rather than treating it as an incidental consequence. The rewritten five-commit series separates characterization tests, metadata tests, the resolver refactor, the behavior change, and documentation.

epage · 2026-07-27T16:03:20Z

+    p.cargo("metadata --format-version 1 -Z bindeps")
+        .masquerade_as_nightly_cargo(&["bindeps"])
+        .with_stdout_data(str![[r#"
+{"packages":[{"name":"bar","version":"0.5.0","id":"path+[ROOTURL]/foo/bar#0.5.0","license":null,"license_file":null,"description":null,"source":null,"dependencies":[],"targets":[{"kind":["lib"],"crate_types":["lib"],"name":"bar","src_path":"[ROOT]/foo/bar/src/lib.rs","edition":"2015","doc":true,"doctest":true,"test":true},{"kind":["bin"],"crate_types":["bin"],"name":"bar","src_path":"[ROOT]/foo/bar/src/main.rs","edition":"2015","doc":true,"doctest":false,"test":true}],"features":{},"manifest_path":"[ROOT]/foo/bar/Cargo.toml","metadata":null,"publish":null,"authors":[],"categories":[],"keywords":[],"readme":null,"repository":null,"homepage":null,"documentation":null,"edition":"2015","links":null,"default_run":null,"rust_version":null},{"name":"foo","version":"0.5.0","id":"path+[ROOTURL]/foo#0.5.0","license":null,"license_file":null,"description":null,"source":null,"dependencies":[{"name":"bar","source":null,"req":"*","kind":null,"rename":"bar-alias","optional":false,"uses_default_features":true,"features":[],"artifact":{"kinds":["bin"],"lib":false,"target":null},"target":null,"registry":null,"path":"[ROOT]/foo/bar"}],"targets":[{"kind":["lib"],"crate_types":["lib"],"name":"foo","src_path":"[ROOT]/foo/src/lib.rs","edition":"2015","doc":true,"doctest":true,"test":true}],"features":{},"manifest_path":"[ROOT]/foo/Cargo.toml","metadata":null,"publish":null,"authors":[],"categories":[],"keywords":[],"readme":null,"repository":null,"homepage":null,"documentation":null,"edition":"2015","links":null,"default_run":null,"rust_version":null}],"workspace_members":["path+[ROOTURL]/foo#0.5.0"],"workspace_default_members":["path+[ROOTURL]/foo#0.5.0"],"resolve":{"nodes":[{"id":"path+[ROOTURL]/foo/bar#0.5.0","dependencies":[],"deps":[],"features":[]},{"id":"path+[ROOTURL]/foo#0.5.0","dependencies":["path+[ROOTURL]/foo/bar#0.5.0"],"deps":[{"name":"bar_alias","pkg":"path+[ROOTURL]/foo/bar#0.5.0","dep_kinds":[{"kind":null,"target":null,"extern_name":"bar_alias","artifact":"bin","bin_name":"bar"}]}],"features":[]}],"root":"path+[ROOTURL]/foo#0.5.0"},"target_directory":"[ROOT]/foo/target","build_directory":"[ROOT]/foo/target","version":1,"workspace_root":"[ROOT]/foo","metadata":null}


There a reason this isn't being done as a json snapshot like our other metadata tests?

View changes since the review

epage · 2026-07-27T16:07:24Z

This update expands the scope of the PR in one deliberate way: dependency-level lib = true is no longer supported.

I have been wondering about simplifying that down. I have also wondered about going a step further and only allowing one artifact per dependency specification.

This seems like a big design change and it feels buried within a commit for allowing multiple edges to the same package. Is there a way t we can split that out into its own PR, even if there are some hacky things along the way that would then get cleaned up?

And does this mean we should remove lib completely from dependency specifications? I didn't see that change in this PR.

Artifact dependencies no longer also act as Rust library dependencies. Remove `lib` from the dependency manifest schema and artifact metadata. Retain the legacy registry fields solely for wire compatibility. Ignore the flag when reading index entries, record `false` in generated index entries, and omit it from publish uploads. Report a targeted migration error when a manifest uses the removed key.

npmccallum · 2026-07-27T19:01:13Z

Thanks—you’re right that this is a design change and that it needs to be explicit. However, I believe that this PR represents the only reasonable design.

Each bindep declaration defines a distinct dependency tree: its identity, compile target, and feature selection. Its artifact list only selects which outputs Cargo builds from that tree.

lib = true breaks this model by asking one declaration to be both an independent artifact tree and a normal library edge. Those roles require separate declarations so each can have its own identity and features. A normal dependency declaration is therefore the only clear way to provide the library.

If “one artifact per dependency specification” means one declaration per binary, I think that puts the boundary in the wrong place. Multiple outputs can legitimately share one dependency tree and feature set, which is exactly what an artifact list expresses. Conversely, outputs needing different features can use the same package under separate aliases with disjoint artifact lists:

[build-dependencies.tools-stable]
package = "tools"
artifact = ["bin:fmt", "bin:lint"]
features = ["stable"]

[build-dependencies.tools-experimental]
package = "tools"
artifact = ["bin:pack", "bin:ship"]
features = ["experimental"]

That is why the removal belongs in this PR. I’ve made the complete surface removal explicit in its own commit, covering the manifest schema, artifact model, and metadata; the historical registry field remains only for compatibility and is ignored. Moving this to another PR would leave Cargo in a knowingly incoherent intermediate state.

Removing the dependency-level `lib` field is a breaking change to the public manifest schema API. Advance the pre-1.0 minor version accordingly.

Show how separate aliases request artifacts from the same package for multiple target platforms. Replace the removed dependency-level `lib` option with explicit normal and artifact dependency declarations. Explain that their feature graphs remain independent and show how to migrate existing manifests.

epage · 2026-07-27T20:34:47Z

If “one artifact per dependency specification” means one declaration per binary, I think that puts the boundary in the wrong place. Multiple outputs can legitimately share one dependency tree and feature set, which is exactly what an artifact list expresses. Conversely, outputs needing different features can use the same package under separate aliases with disjoint artifact lists:

For context, there is some regret on even allowing multiple addressable crates within one package.

I also wonder how much using multiple bins will be done in practice.

Each bindep declaration defines a distinct dependency tree: its identity, compile target, and feature selection. Its artifact list only selects which outputs Cargo builds from that tree.

Within the same target, are features unified or independent today, including no-target with rlibs?

I’ve made the complete surface removal explicit in its own commit, covering the manifest schema, artifact model, and metadata; the historical registry field remains only for compatibility and is ignored. Moving this to another PR would leave Cargo in a knowingly incoherent intermediate state.

I appreciate it being its own commit. What isn't too clear is why this can only be done where it is an incoherent state? Another option would be to remove it before this PR though that would require also updating any tests that set both a lib and an artifact. Might be better this way.

rustbot assigned weihanglo Jun 3, 2026

rustbot added A-dependency-resolution Area: dependency resolution and the resolver A-documenting-cargo-itself Area: Cargo's documentation Command-metadata S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. labels Jun 3, 2026

npmccallum mentioned this pull request Jun 4, 2026

feat(tree): show artifact dependency details #17075

Open

weihanglo reviewed Jun 24, 2026

View reviewed changes

rustbot assigned epage and unassigned weihanglo Jun 24, 2026

epage reviewed Jun 24, 2026

View reviewed changes

npmccallum force-pushed the master branch from 0294a83 to df3e262 Compare June 26, 2026 16:58

This comment has been minimized.

Sign in to view

npmccallum force-pushed the master branch 2 times, most recently from 4546761 to 55cb078 Compare July 3, 2026 14:16

This comment has been minimized.

Sign in to view

npmccallum force-pushed the master branch from 55cb078 to a67fa8b Compare July 20, 2026 00:09

epage reviewed Jul 22, 2026

View reviewed changes

npmccallum added 4 commits July 27, 2026 10:21

refactor(resolve): extract dependency name calculation

731ecb7

Move the per-dependency extern-crate and manifest-name calculation into a private helper while preserving the existing duplicate-name validation and behavior.

npmccallum force-pushed the master branch from a67fa8b to d6ae9a0 Compare July 27, 2026 15:14

epage reviewed Jul 27, 2026

View reviewed changes

npmccallum force-pushed the master branch from d6ae9a0 to d99914f Compare July 27, 2026 18:57

rustbot added A-interacts-with-crates.io Area: interaction with registries A-registries Area: registries Command-package Command-publish labels Jul 27, 2026

npmccallum added 2 commits July 27, 2026 15:17

chore(schema): bump cargo-util-schemas to 0.15.0

432aa91

Removing the dependency-level `lib` field is a breaking change to the public manifest schema API. Advance the pre-1.0 minor version accordingly.

npmccallum force-pushed the master branch from d99914f to f2a3650 Compare July 27, 2026 19:18

		#[cargo_test]
		fn multiple_artifact_aliases_no_op_rebuild() {

Uh oh!

Uh oh!

Conversation

npmccallum commented Jun 3, 2026 • edited by rustbot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Design change: dependency-level lib

Uh oh!

rustbot commented Jun 3, 2026

Uh oh!

npmccallum commented Jun 24, 2026

Uh oh!

weihanglo left a comment • edited by rustbot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

npmccallum commented Jun 24, 2026

Uh oh!

epage commented Jun 24, 2026

Uh oh!

epage left a comment • edited by rustbot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

epage Jun 24, 2026

Choose a reason for hiding this comment

Uh oh!

epage Jun 24, 2026

Choose a reason for hiding this comment

Uh oh!

This comment has been minimized.

npmccallum commented Jun 26, 2026

Uh oh!

weihanglo commented Jun 26, 2026

Uh oh!

This comment has been minimized.

npmccallum commented Jul 4, 2026

Uh oh!

This comment has been minimized.

epage Jul 22, 2026 • edited by rustbot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

epage Jul 22, 2026 • edited by rustbot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

epage Jul 22, 2026 • edited by rustbot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

epage Jul 22, 2026 • edited by rustbot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

epage Jul 22, 2026 • edited by rustbot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

epage Jul 22, 2026 • edited by rustbot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

rustbot commented Jul 27, 2026

Uh oh!

npmccallum commented Jul 27, 2026

Uh oh!

epage Jul 27, 2026 • edited by rustbot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

epage commented Jul 27, 2026

Uh oh!

npmccallum commented Jul 27, 2026

Uh oh!

epage commented Jul 27, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

npmccallum commented Jun 3, 2026 •

edited by rustbot

Loading

Design change: dependency-level `lib`

weihanglo left a comment •

edited by rustbot

Loading

epage left a comment •

edited by rustbot

Loading

epage Jul 22, 2026 •

edited by rustbot

Loading

epage Jul 22, 2026 •

edited by rustbot

Loading

epage Jul 22, 2026 •

edited by rustbot

Loading

epage Jul 22, 2026 •

edited by rustbot

Loading

epage Jul 22, 2026 •

edited by rustbot

Loading

epage Jul 22, 2026 •

edited by rustbot

Loading

epage Jul 27, 2026 •

edited by rustbot

Loading