Transitive Frontier


Querying Cargo Dependency DAGs with Guppy

dependency graph

guppy is a rust crate that provides tools for working with cargo dependency graphs using the petgraph graph data structure crate. It’s used by Facebook to audit a high-security subset of the cargo dependency graph for some of their more high-visibility projects. Treating the dependency graph resulting from a cargo build operation as a DAG lets us draw on the well-studied field of graph algorithms to answer complex questions about our build without resorting to ad-hoc traversals or re-implementation of common graph primitives.

For my first project using guppy, I decided to build a tool to produce machine-readable summaries describing why some target dependency is included in a cargo workspace’s build graph. My motivation was to support projects that are migrating from futures 0.1 to futures 0.3. Many rust projects started using futures 0.1 for their initial async implementation, and are still in the process of switching over to futures 0.3. If you’re interested in learning more about the differences between the two packages, this blog post by ncameron is a great resource. Being able to easily generate machine-readable reports opens up new possibilities - for example, you could use the output of this tool to build a linter that asserts that no new transitive dependencies on futures 0.1 are introduced into a workspace, to provide tooling-backed assurances that usage of futures 0.1 only ever decreases.

The technical name for this is reverse transitive dependency, which is just the name for the subset of the dependency graph that leads back to some target dependency. I decided to focus on the workspace frontier, which I’ve defined as the set of edges in a dependency graph that originate from within a cargo workspace but do not terminate in that cargo workspace. These are the edges in the dependency graph via which dependencies on external packages are introduced into the workspace, which makes them the natural focus for my use case. Since I’m specifically interested in the places where some dependencies on some specific package make their way into the workspace, I can narrow my search to the subset of the dependency graph that leads back to that package, otherwise known as the reverse transitive dependencies of that package.

I’ve called this tool transitive-frontier, because it finds the intersection of the workspace frontier and the reverse transitive dependency graph - that is, the places where dependencies on some target package are introduced into a cargo workspace. on some package are introduced. Building it has been a great way to familiarize myself with guppy. You can find it on github here, if you’d like to follow along or use it to audit dependencies in one of your projects.

Using Guppy

Before describing the transitive_frontier tool, I’m going to go over how I use guppy: these examples are run against some simple strawman workspaces, which you can find in the example_workspaces directory in the transitive_frontier repo. Here’s a simplified representation of the dependency graph for the first workspace:

dependency graph

The main entry point into guppy is the PackageGraph, which holds information about the dependency graph resulting from a cargo build operation (as generated via cargo metadata). Here’s how to construct a PackageGraph instance for your workspace:

let mut cmd = MetadataCommand::new();
if let Some(workspace) = &opt.workspace {
    cmd.current_dir(workspace);
}
let package_graph = PackageGraph::from_command(&mut cmd)?;

Next, let’s construct a query to find all the packages that depend on futures 0.1 in this package graph. You can find package ids in the output of cargo metadata, but the transitive_frontier tool supports providing a substring of the target package id for convenience.

let futures =
  "futures 0.1.30 (registry+https://github.com/rust-lang/crates.io-index)";
let target_dependency = PackageId::new(futures);
let package_query = package_graph.query_reverse(iter::once(id))?;

After building a query, we resolve it, yielding a PackageSet. It’s possible to add filters at this stage, to avoid traversing certain dependency edges.

let package_set = package_query.resolve();

After we’ve constructed the query, we need to resolve it - this step is where we actually traverse edges to build a package set. Here we’re just resolving every edge that the query matches, but guppy also supports resolving queries with a filter function, to allow for skipping some edges.

Here’s a visualization of the package set this query produces, with the reverse transitive dependencies of futures 0.1 highlighted:

dependency graph

However, we don’t want a package set - we want the frontier via which dependencies futures 0.1 are introduced into our workspace, shown further highlighted here:

dependency graph

For this step, we need to traverse the set of edges included in the PackageSet. Since it’s a package set, it only tracks packages, so what we’re actually doing is iterating over every edge in the package graph for where to and from are members of the package set.

let mut frontier = HashMap::new();

for link in package_set.links(DependencyDirection::Reverse) {
    // != implements logical xor
    if link.to().in_workspace() != link.from().in_workspace() {
        let entry = frontier
            .entry(link.from().name().to_string())
            .or_insert_with(Vec::new);
        entry.push(format!("{} {}", link.to().name(), link.to().version()))
    }
}

This code block iterates over every edge in the PackageSet, taking edges with one end inside the workspace and the other outside it and storing them in a HashMap keyed by the name of the package introducing the dependency, resulting in this output:

target_dependency = "futures 0.1.30"

[frontier]
capabilities-old = ["library_old 0.1.0", "futures 0.1.30"]
server = ["futures 0.1.30"]

You can reproduce this by running the following command in the transitive_frontier repo: cargo run -- -p "futures 0.1" -- example_workspaces/project. This example, and those that will follow, highlight snippets of code similar to those used to implement the functionality of the transitive_frontier tool.

Ignoring Packages

In a real project that’s still migrating off futures 0.1 usage, futures 0.3 would likely be compiled with the compat feature enabled, to better support interop. example_packages/project_with_compat provides an example of this. Let’s take a look at the transitive frontier for futures 0.1 for that workspace, via cargo run -- -p "futures 0.1" -- example_workspaces/project_with_compat:

target_dependency = "futures 0.1.30"

[frontier]
capabilities-old = ["futures 0.1.30", "library_old 0.1.0"]
server = ["futures 0.3.8", "futures 0.1.30"]
capabilities-new = ["library_new 0.1.0", "futures 0.3.8"]

Unfortunately, the compat dependency in futures_util makes futures 0.1 a transitive dependency of futures 0.3, which causes every edge that introduces a dependency on futures 0.3 to be included in the report. This is not the desired behavior. Here’s a visualization of the problem, showing the path via which futures 0.3 depends on futures 0.1:

dependency graph

To solve this issue, we’re going to have to skip some edges:

let package_set = package_query
      .resolve_with_fn(|_, link| !opt.skip.iter().any(|s| link.to().id().repr().contains(s)));

This code block takes a provided list of substrings to ignore and skips edges to any package with an id containing an ignored substring. Let’s try regenerating the transitive frontier with that enabled, via cargo run -- -p "futures 0.1" --skip "futures-util" -- example_workspaces/project_with_compat:

target_dependency = "futures 0.1.30 (registry+https://github.com/rust-lang/crates.io-index)"

[frontier]
capabilities-old = ["library_old 0.1.0", "futures 0.1.30"]
server = ["futures 0.1.30"]

That fixes the problem by skipping futures-util, as you can see in this visualization:

dependency graph

Now we have the ability to generate machine-readable reports on where transitive dependencies on futures 0.1 enter the workspace. For this workspace, that doesn’t mean that much, but for larger workspaces with tens or even hundreds of crates automating processes like this can be critical.

Summary

Guppy has been really great to work with. It provides a great set of tools for querying cargo dependency graphs, the docs are really comprehensive, and the maintainers are really responsive and helpful (thanks @rain!). I particularly appreciated being able to explore the problem by writing code to query the graph instead of greppingCargo.toml files. I was able to leverage all this to build the transitive_frontier tool really quickly. The abstractions provided by guppy let me spend most of my development time exploring the problem space and thinking about output formats.

transitive_frontier supports outputting two machine readable formats (JSON, TOML) and one human readable one (HTML). You can fetch it via cargo install transitive_frontier.

Example Invocations

  • Find the transitive frontier for futures 0.1 for a workspace and output TOML:
transitive_frontier -p "futures 0.1" --skip "futures-util" \
                    -f toml -- ~/target/dir > summary.toml
  • Find the transitive frontier for futures 0.1 for a workspace and output JSON:
transitive_frontier -p "futures 0.1" --skip "futures-util" \
                    -f json -- ~/target/dir > summary.json
  • Find the transitive frontier for futures 0.1 for a workspace and output HTML:
transitive_frontier -p "futures 0.1" --skip "futures-util" \
                    -f html -- ~/target/dir > summary.html
guppy  rust 

See also