`cargo tree --duplicates --target all` non-deterministic output

[ratmice]

Problem

I have been working on reducing duplicate dependencies from a crate.
However I noticed that the output of cargo -tree --duplicates --target all
sometimes changes between printing a bunch of duplicates of the same version,
and printing warning: nothing to print

$ cargo tree --duplicates --target all

warning: nothing to print.

To find dependencies that require specific target platforms, try to use option `--target all` first, and then narrow your search scope accordingly.

I'm going to snip the output, but other times I see (what I believe different feature usage of a few crates between build-time and compile time)

$ cargo tree --duplicates --target all

syn v2.0.100
...

syn v2.0.100
...

windows-sys v0.59.0
...

windows-sys v0.59.0

I would expect cargo tree to print deterministic output when invoked.
The output seems to go through "phases", e.g. If you invoke it fast enough
you'll get a few instances of duplicates, followed by a few info instances of "warning: nothing to print".

But doesn't seem to be affected by whether rust-analyzer is running.

cargo 1.85.0 (d73d2ca 2024-12-31)

Steps

git clone -b unstable_cargo_tree https://github.com/ratmice/grmtools.git
cd grmtools/
cargo tree --duplicates --target all
cargo tree --duplicates --target all

repeat the last two steps until you see both output results.

Possible Solution(s)

No response

Notes

No response

Version

[epage]

I'm able to reproduce this

When we call from_reachables, we still get a graph back

cargo/src/cargo/ops/tree/mod.rs

Line 207 in 7220445

graph = graph.from_reachable(root_indexes.as_slice());

But when we call find_duplicates, it is empty

cargo/src/cargo/ops/tree/mod.rs

Line 208 in 7220445

graph.find_duplicates()

I've not yet narrowed down why that is.

We have some inherent non-determinism to our operations that makes comparing dumped state more difficult. While we sort_unstable_by for more operations, when we do it for deps, we sort by the name, which is usually the same, and get back different orders

cargo/src/cargo/ops/tree/graph.rs

Line 481 in 7220445

deps.sort_unstable_by_key(|dep| dep.name_in_toml());

[epage]

Another run of mine showed this non-determinism

[package]
name = "calc_actions"
version = "0.1.0"
authors = ["Laurence Tratt <http://tratt.net/laurie/>"]
edition = "2021"
license = "Apache-2.0/MIT"

[[bin]]
doc = false
name = "calc_actions"

[build-dependencies]
cfgrammar = { path="../../../cfgrammar" }
lrlex = { path="../../../lrlex" }
lrpar = { path="../.." }

[dependencies]
cfgrammar = { path="../../../cfgrammar" }
lrlex = { path="../../../lrlex" }
lrpar = { path="../.." }

In these two runs, we processed the build and normal cfgrammar in different orders.

When I added dep.kind() to the sort, I could no longer reproduce the problem which is annoying because that is unlikely the root cause.