fix(tasks): restore parallel starts with poetry via list_paths cache and stable exec-env cache

[jdx]

Summary

Minimal fix for parallel task execution with poetry. Tasks now start concurrently again (~4ms spread vs ~6s).

Problem

With poetry installed, parallel tasks were serializing due to:

1. Repeated list_paths computations for each task
2. Cache invalidations from dirs::DATA freshness checks
3. Lock contention in external_plugin_cache

Solution

1. Added LIST_PATHS_CACHE to cache toolset list_paths results per project/toolset
2. Use Arc in ExternalPluginCache to release locks earlier
3. Removed dirs::DATA fresh check that caused spurious cache invalidations
4. Added trace logs for debugging task startup timing

Testing

Updated test_task_parallel_execution to demonstrate the bug with poetry installed. Test now passes reliably with all tasks starting within milliseconds.

Verification

• Before: Tasks start spread ~6000ms (sequential)
• After: Tasks start spread ~4ms (parallel)
• Total execution time: ~3s (vs ~8s+ when sequential)

[github-actions]

Hyperfine Performance

mise x -- echo

Command	Mean [ms]	Min [ms]	Max [ms]	Relative
mise-2025.9.6 x -- echo	18.3 ± 0.3	17.9	20.9	1.00
mise x -- echo	18.9 ± 0.2	18.5	19.7	1.03 ± 0.02

mise env

Command	Mean [ms]	Min [ms]	Max [ms]	Relative
mise-2025.9.6 env	17.8 ± 0.3	17.3	20.0	1.00
mise env	18.4 ± 0.4	17.8	21.8	1.03 ± 0.03

mise hook-env

Command	Mean [ms]	Min [ms]	Max [ms]	Relative
mise-2025.9.6 hook-env	17.4 ± 0.2	17.0	18.3	1.00
mise hook-env	18.0 ± 0.3	17.6	20.4	1.03 ± 0.02

mise ls

Command	Mean [ms]	Min [ms]	Max [ms]	Relative
mise-2025.9.6 ls	15.9 ± 0.2	15.5	16.9	1.00
mise ls	16.0 ± 0.2	15.5	17.5	1.01 ± 0.02

xtasks/test/perf

Command	mise-2025.9.6	mise	Variance
install (cached)	163ms	✅ 102ms	+59%
ls (cached)	62ms	62ms	+0%
bin-paths (cached)	65ms	68ms	-4%
task-ls (cached)	465ms	463ms	+0%

✅ Performance improvement: install cached is 59%

[Copilot]

Pull Request Overview

This PR fixes parallel task execution with poetry by implementing a comprehensive caching strategy to prevent serialization of concurrent tasks. The issue was that tasks were recomputing environment and path information repeatedly, causing them to run sequentially instead of in parallel.

• Introduces a global LIST_PATHS_CACHE to cache toolset path computations across identical configurations
• Removes spurious cache invalidations from poetry exec-env cache by eliminating the dirs::DATA freshness check
• Adds detailed tracing to identify performance bottlenecks in task execution

Reviewed Changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
src/toolset/mod.rs	Implements global LIST_PATHS_CACHE with stable cache keys and adds tracing for list_bin_paths operations
src/cli/run.rs	Adds comprehensive timing traces for task execution phases including semaphore acquisition and environment rendering
src/backend/external_plugin_cache.rs	Refactors cache managers to use Arc for thread safety and adds timing traces for cache operations
src/backend/asdf.rs	Removes dirs::DATA freshness check and adds exec_env timing traces
e2e/tasks/test_task_parallel_execution	Enhances test with poetry tooling and multiple runs for reliability verification

---

_{Tip: Customize your code reviews with copilot-instructions.md. Create the file or learn how to get started.}

[Copilot]

The cache key construction could lead to collisions if project paths contain the delimiter '|'. Consider using a more robust serialization approach like JSON or a structured hash.

[Copilot]

The nested Arc<CacheManager> pattern creates unnecessary indirection. Consider using DashMap directly for better performance and simpler code structure.