|
| 1 | +- Feature Name: source_replacement_ambiguity |
| 2 | +- Start Date: 2022-07-05 |
| 3 | +- RFC PR: [rust-lang/rfcs#3289](https://github.com/rust-lang/rfcs/pull/3289) |
| 4 | +- Tracking Issue: [rust-lang/cargo#10894](https://github.com/rust-lang/cargo/issues/10894) |
| 5 | + |
| 6 | +# Summary |
| 7 | +[summary]: #summary |
| 8 | + |
| 9 | +When Cargo is performing an API operation (`yank`/`login`/`publish`/etc.) to a source-replaced `crates-io`, require the user to pass `--registry <NAME>` to specify exactly which registry to use. Additionally, ensure that the token for `crates-io` is never sent to a replacement registry. |
| 10 | + |
| 11 | +# Motivation |
| 12 | +[motivation]: #motivation |
| 13 | + |
| 14 | +There are multiple issues that this RFC attempts to resolve around source-replacement. |
| 15 | + |
| 16 | +* When Cargo is performing an API operation, source replacement is only respected for `crates-io`, not alternative registries. This is inconsistent. |
| 17 | +* The [error message](https://github.com/rust-lang/cargo/issues/6722) for attempting to publish to a replaced crates-io is confusing, and there is no workaround other than temporarily removing the source replacement configuration. |
| 18 | +* When performing an API operation other than `publish` with a replaced `crates-io` source, the `crates-io` credentials are sent to the replacement registry's API. This is a security risk. |
| 19 | +* It's unclear which credentials should be used when fetching a source-replaced authenticated alternate registry ([RFC 3139][3139]). |
| 20 | + |
| 21 | +# Guide-level explanation |
| 22 | +[guide-level-explanation]: #guide-level-explanation |
| 23 | + |
| 24 | +When the `crates-io` source is replaced, the user needs to specify `--registry <NAME>` when running an API operation to disambiguate which registry to use. Otherwise, `cargo` will issue an error. |
| 25 | + |
| 26 | +`cargo` only sends the token associated with a given registry to that registry and no other (even if source replacement is configured). |
| 27 | + |
| 28 | +When replacing a source with a registry, the `replace-with` key can reference the name of a registry in the `[registries]` table. |
| 29 | + |
| 30 | +## Example scenarios |
| 31 | + |
| 32 | +### Local source replacement (vendoring) |
| 33 | +A repository has a local `.cargo/config.toml` that vendors all dependencies from crates.io. Fetching and building within the repository would work as expected with the vendored sources. |
| 34 | + |
| 35 | +If the user decides to publish the crate, `cargo publish --registry crates-io` will ignore the source-replacement and publish to crates.io. |
| 36 | + |
| 37 | +### `crates-io` mirror registry |
| 38 | +A server has been set up that provides a complete mirror of crates.io. The user has configured a `~/.cargo/config.toml` that points to the mirror registry in the `[registries]` table. The mirror requires authentication to access (based on [RFC 3139][3139]). |
| 39 | + |
| 40 | +The user can log in to the mirror using `cargo login --registry mirror`. Fetching and building use the mirror. |
| 41 | + |
| 42 | +The user decides to publish the crate to crates.io, and does `cargo login --registry crates-io` to log in to crates.io. Source replacement is ignored, and the token is saved. |
| 43 | + |
| 44 | +Next, the user runs `cargo publish --registry crates-io` to publish to crates.io. Cargo ignores source replacement when building and publishing the crate to crates.io. |
| 45 | + |
| 46 | +# Reference-level explanation |
| 47 | +[reference-level-explanation]: #reference-level-explanation |
| 48 | + |
| 49 | +### Change 1: respect `--registry` |
| 50 | +When running an API operation (`login`, `logout`, `owner`, `publish`, `search`, `yank`), Cargo always uses the registry specified by `--registry <NAME>`, and never a source-replacement. |
| 51 | + |
| 52 | +### Change 2: error for replaced crates-io |
| 53 | +When running an API operation (as defined above) and ALL of the following are true: |
| 54 | +* `crates-io` has been replaced by a remote-registry source. |
| 55 | +* command line argument `--registry <NAME>` is not present. |
| 56 | +* command line argument `--index <URL>` is not present. |
| 57 | +* `Cargo.toml` manifest key `publish = <NAME>` is not set (only applies for publishing). |
| 58 | + |
| 59 | +`cargo` issues an error: |
| 60 | +``` |
| 61 | +error: crates-io is replaced: use `--registry replacement` or `--registry crates-io` |
| 62 | +``` |
| 63 | + |
| 64 | +### Change 3: credentials are only sent to the same registry |
| 65 | +If the `crates-io` source is replaced with another remote registry, the credentials for |
| 66 | +`crates-io` are never sent to the replacement registry. This makes `crates-io` consistant |
| 67 | +with alternative registries and ensures credentials are only sent to the registry they are |
| 68 | +associated with. |
| 69 | + |
| 70 | +### Change 4: `[source]` table can reference `[registries]` table |
| 71 | +The `replace-with` key in the `[source]` table can reference a registry defined in the `[registries]` table. |
| 72 | + |
| 73 | +For example, the following configuration would be valid: |
| 74 | + |
| 75 | +``` |
| 76 | +[source.crates-io] |
| 77 | +replace-with = "my-registry" |
| 78 | +
|
| 79 | +[registries.my-registry] |
| 80 | +index = "https://my-registry-index/" |
| 81 | +``` |
| 82 | + |
| 83 | +This is necessary to allow the `--registry <NAME>` command-line argument to work with source-replaced registries. It also allows additional configuration (such as a token) to be specified for a source-replacement registry without duplicating configuration between `[registries]` and `[source]` tables. |
| 84 | + |
| 85 | +# Drawbacks |
| 86 | +[drawbacks]: #drawbacks |
| 87 | + |
| 88 | +Behavior is changed around where credentials are sent, which could break some workflows. |
| 89 | + |
| 90 | +If a mirror of crates.io is set up with `config.json` containing `"api": "https://crates.io"`, then the current system of sending the crates.io token to the replaced source would work correctly, and this RFC would break it. |
| 91 | + |
| 92 | +# Rationale and alternatives |
| 93 | +[rationale-and-alternatives]: #rationale-and-alternatives |
| 94 | + |
| 95 | +## Alternative: ignore source replacement for API operations |
| 96 | +When doing an API operation with a replaced `crates-io`, `cargo` would ignore source replacement without additional arguments. This is how alternative registries currently work. |
| 97 | + |
| 98 | +If the user wants to use the replacement, they could pass `--registry <NAME>`, but would not be required to do so. |
| 99 | + |
| 100 | +A new option `--respect-source-config` could be added to make cargo follow the source replacement for API operations (similar to what we already have for `cargo vendor`). |
| 101 | + |
| 102 | +This may be too confusing for users since it silently changes behavior. The RFC proposes a solution that requires the user to be explicit about which registry to use in the ambiguous situation (crates-io replacement). |
| 103 | + |
| 104 | +## Alternative: disallow source replacement for API operations |
| 105 | + |
| 106 | +Attempting an API operation on a replaced source would be an error. The user could use `--registry crates-io` to explicitly bypass the source replacement. |
| 107 | +``` |
| 108 | +Error: <operation> is not supported on replaced source `crates-io-mirror`; use `--registry crates-io` for the original source |
| 109 | +``` |
| 110 | + |
| 111 | +# Prior art |
| 112 | +[prior-art]: #prior-art |
| 113 | + |
| 114 | +Other package managers don't seem to have a source replacement feature. |
| 115 | + |
| 116 | +# Unresolved questions |
| 117 | +[unresolved-questions]: #unresolved-questions |
| 118 | + |
| 119 | +Should the `--registry <NAME>` command line argument be allowed to reference the name of a `source` from the `[source]` table as well? This makes it more flexible, but adds potentially unnecessary complexity. |
| 120 | + |
| 121 | +Cargo's tests rely on the ability to replace the crates.io source and have the crates.io credentials go to the replaced source. We need a way for these tests to continue working. |
| 122 | + |
| 123 | +# Future possibilities |
| 124 | +[future-possibilities]: #future-possibilities |
| 125 | + |
| 126 | +Can't think of anything. |
| 127 | + |
| 128 | +[3139]: https://rust-lang.github.io/rfcs/3139-cargo-alternative-registry-auth.html |
0 commit comments