perf(install): use Kahn's algorithm for dependency scheduling

[jdx]

Summary

• Replace batch-based tool installation with channel-based Kahn's algorithm for better parallelization of dependency chains
• Add new tool_deps.rs module with petgraph-based dependency graph
• Replace batch loop in install_all_versions with tokio::select! loop
• Remove unused get_leaf_dependencies and install_some_versions functions

Before

Install all leaves in a batch → wait for entire batch to complete → recalculate leaves → repeat

For chains A → B → C and D → E:

[Install A,D] → wait for both → [Install B,E] → wait for both → [Install C]

After

As soon as a tool completes, immediately emit any newly-ready dependents

For chains A → B → C and D → E:

Start A,D; when A finishes, start B immediately (don't wait for D)

Test plan

• Unit tests pass (mise run test:unit)
• Build succeeds
• Lint passes
• E2E tests pass (tested locally with test_install suite)
• e2e/backend/test_pipx_direct_dependencies tests 2-level dependency chain
• e2e/backend/test_pipx_deep_dependencies_slow tests 3-level dependency chain

🤖 Generated with Claude Code

---

Note

Medium Risk
Moderate risk because it rewrites core install orchestration, changing concurrency, dependency ordering, and failure propagation (including blocking dependents), which could cause subtle deadlocks/order regressions or different error surfaces.

Overview
Switches Toolset::install_all_versions from a batch “install leaf set, recompute” loop to a streaming scheduler: tools are emitted as soon as their dependencies complete, improving parallelism across independent chains.

Introduces tool_deps.rs (ToolDeps) using a petgraph-backed dependency graph to dedupe requests, detect cycles (marking them blocked), and block transitive dependents on failures; install execution is refactored into install_with_deps/install_single_tool with a tokio::select! loop and panic recovery.

Updates plugin-install error handling to be collected per-tool and merged into the final InstallFailed result, and adjusts the E2E error-display expectation for “tool not found” to include the Failed to install … prefix.

^{Written by Cursor Bugbot for commit f9e3ce7. This will update automatically on new commits. Configure here.}

[gemini-code-assist]

Warning

You have reached your daily quota limit. Please wait up to 24 hours and I will start processing your requests again!

[Copilot]

Pull request overview

This PR replaces the batch-based tool installation with a channel-based implementation using Kahn's algorithm for better parallelization of dependency chains. The new approach allows tools to be installed immediately when their dependencies complete, rather than waiting for entire batches to finish.

Changes:

• Introduced tool_deps.rs module implementing petgraph-based dependency graph with Kahn's algorithm
• Refactored install_all_versions to use tokio::select! loop instead of batch processing
• Removed unused get_leaf_dependencies and install_some_versions helper functions

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 4 comments.

File	Description
src/toolset/toolset_install.rs	Replaced batch installation loop with dependency graph-based scheduling using channels
src/toolset/tool_deps.rs	New module implementing Kahn's algorithm for dependency ordering with petgraph
src/toolset/mod.rs	Added tool_deps module declaration
src/toolset/helpers.rs	Removed unused get_leaf_dependencies function

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

The receiver from this channel is immediately dropped. This creates a channel that cannot receive messages until subscribe() is called. Consider moving channel creation to subscribe() or documenting why this initial channel is necessary.

[Copilot]

This nested loop creates O(n²) complexity when collecting plugin errors. Since you've already filtered by backend in the outer loop (line 230-234), consider building a map of backend to tools beforehand to avoid this nested iteration.

[Copilot]

The biased annotation ensures completed installations are handled before new ones are started. Consider adding a comment explaining this priority ordering is intentional for correctness.

Suggested change

	tokio::select! {
	tokio::select! {
	// Use `biased` so completed installations are handled before starting new ones.
	// This priority ordering is intentional and required for correctness of dependency tracking.

[Copilot]

Missing import for trace! macro. Ensure use tracing::trace; or equivalent is added to support this logging statement.

[jdx]

bugbot run

[jdx]

bugbot run

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{Bugbot Autofix is ON, but a Cloud Agent failed to start.}

[github-actions]

Hyperfine Performance

mise x -- echo

Command	Mean [ms]	Min [ms]	Max [ms]	Relative
mise-2026.1.12 x -- echo	19.6 ± 0.4	19.1	25.0	1.00
mise x -- echo	19.8 ± 0.5	19.1	22.3	1.01 ± 0.03

mise env

Command	Mean [ms]	Min [ms]	Max [ms]	Relative
mise-2026.1.12 env	19.1 ± 0.4	18.4	23.5	1.00
mise env	19.3 ± 0.6	18.7	25.7	1.01 ± 0.04

mise hook-env

Command	Mean [ms]	Min [ms]	Max [ms]	Relative
mise-2026.1.12 hook-env	19.9 ± 0.5	19.1	22.6	1.00
mise hook-env	20.8 ± 1.0	19.3	31.0	1.04 ± 0.06

mise ls

Command	Mean [ms]	Min [ms]	Max [ms]	Relative
mise-2026.1.12 ls	17.8 ± 0.5	16.9	19.2	1.00 ± 0.03
mise ls	17.7 ± 0.3	17.0	19.6	1.00

xtasks/test/perf

Command	mise-2026.1.12	mise	Variance
install (cached)	109ms	110ms	+0%
ls (cached)	69ms	69ms	+0%
bin-paths (cached)	73ms	73ms	+0%
task-ls (cached)	533ms	531ms	+0%