Add preloading for workspace packages in resolve_with_previous

[Muscraft]

Minor Performance Hit

When I was looking into concerns about the performance of workspace inheritance, I found that cargo would parse manifests twice for cargo metadata and cargo check.

I found this after @epage suggested looking into the concerns using dbg!(package.name) placed in different places. When placing that inside of TomlManifest::to_real_manifest and running against any workspace (inheritance or not) it would show that we are parsing each workspace member's Cargo.toml twice.

e.g. cargo run -- metadata --manifest-path ./temp/lunatic/Cargo.toml > /dev/null I would get the output of:

[src/cargo/util/toml/mod.rs:1581] package_name = "lunatic-runtime"
[src/cargo/util/toml/mod.rs:1581] package_name = "hash-map-id"
[src/cargo/util/toml/mod.rs:1581] package_name = "lunatic-stdout-capture"
[src/cargo/util/toml/mod.rs:1581] package_name = "lunatic-process"
[src/cargo/util/toml/mod.rs:1581] package_name = "lunatic-common-api"
[src/cargo/util/toml/mod.rs:1581] package_name = "lunatic-process-api"
[src/cargo/util/toml/mod.rs:1581] package_name = "lunatic-error-api"
[src/cargo/util/toml/mod.rs:1581] package_name = "lunatic-wasi-api"
[src/cargo/util/toml/mod.rs:1581] package_name = "lunatic-messaging-api"
[src/cargo/util/toml/mod.rs:1581] package_name = "lunatic-networking-api"
[src/cargo/util/toml/mod.rs:1581] package_name = "lunatic-timer-api"
[src/cargo/util/toml/mod.rs:1581] package_name = "lunatic-version-api"
[src/cargo/util/toml/mod.rs:1581] package_name = "lunatic-registry-api"
warning: please specify `--format-version` flag explicitly to avoid compatibility problems
[src/cargo/util/toml/mod.rs:1581] package_name = "hash-map-id"
[src/cargo/util/toml/mod.rs:1581] package_name = "lunatic-stdout-capture"
[src/cargo/util/toml/mod.rs:1581] package_name = "lunatic-process"
[src/cargo/util/toml/mod.rs:1581] package_name = "lunatic-common-api"
[src/cargo/util/toml/mod.rs:1581] package_name = "lunatic-process-api"
[src/cargo/util/toml/mod.rs:1581] package_name = "lunatic-error-api"
[src/cargo/util/toml/mod.rs:1581] package_name = "lunatic-wasi-api"
[src/cargo/util/toml/mod.rs:1581] package_name = "lunatic-messaging-api"
[src/cargo/util/toml/mod.rs:1581] package_name = "lunatic-networking-api"
[src/cargo/util/toml/mod.rs:1581] package_name = "lunatic-timer-api"
[src/cargo/util/toml/mod.rs:1581] package_name = "lunatic-version-api"
[src/cargo/util/toml/mod.rs:1581] package_name = "lunatic-registry-api"
[src/cargo/util/toml/mod.rs:1581] package_name = "lunatic-runtime"

Observed Cause

As I dug into this it looks like the manifests are parsed once when creating the Workspace and then again in resolve::resolve_with_previous. I dug deeper into why resolve::resolve_with_previous was parsing the manifests again and found the cause to be from these lines since they call into PackageRegistry::ensure_loaded. ensure_loaded will call self.load(namespace, kind)?; and self.block_until_ready()?; for each of the members. The problem with those lines comes from self.block_until_ready()?;. This is because the source for each member should be PathSource (from what I have seen) which calls self.update() and reparses the manifest.

Proposed solution

After discussing in the related zulip thread, Workspace::preload seemed to fit the needs. It solves the problem by creating a PathSource for all packages that have been found during the creation of a Workspace. Once they have been created they get added to the PackageRegistry that gets passed in.

The reason for keeping the loop over the workspace members is that there are cases where preload does not work (cargo install, cargo package, virtual packages, etc).

Performance gains

This change has notable performance gains if cargo is invoked in a large workspace. On my machine, there was a ~10ms drop-in time on benches/workspace/rust. This could be even bigger depending on the machine and its storage device. While this does help mitigate some of the performance hit brought on by workspace inheritance, it does not solve it entirely.

Before changes

Benchmark 1: cd rust; ../../../target/release/cargo metadata
  Time (mean ± σ):     149.4 ms ±   3.8 ms    [User: 105.9 ms, System: 31.7 ms]
  Range (min … max):   144.2 ms … 162.2 ms    19 runs
 
Benchmark 2: cd rust-ws-inherit; ../../../target/release/cargo metadata
  Time (mean ± σ):     191.6 ms ±   1.4 ms    [User: 145.9 ms, System: 34.2 ms]
  Range (min … max):   188.8 ms … 193.9 ms    15 runs

After changes

Benchmark 1: cd rust; ../../../target/release/cargo metadata
  Time (mean ± σ):     139.3 ms ±   1.7 ms    [User: 99.8 ms, System: 29.4 ms]
  Range (min … max):   137.1 ms … 144.5 ms    20 runs
 
Benchmark 2: cd rust-ws-inherit; ../../../target/release/cargo metadata
  Time (mean ± σ):     161.7 ms ±   1.4 ms    [User: 120.4 ms, System: 31.2 ms]
  Range (min … max):   158.0 ms … 164.6 ms    18 runs

[rust-highfive]

r? @ehuss

(rust-highfive has picked a reviewer for you, use r? to override)

[Muscraft]

Looks like my finger hit enter on accident before I could get the title and comment done. I'll update them to be correct

[joshtriplett]

(Sorry, this should have gone through bors.)