fix: deduplicate Windows Conda casing aliases

[Karthik Nadig (karthiknadig)]

Summary

• preserve the canonical executable path on cache hits
• keep caller-facing aliases valid in the current working context
• invalidate stale in-memory entries when tracked executables disappear
• deduplicate Windows Conda casing aliases while preserving on-disk spelling
• version performance inventory semantics so the intentional v1-to-v2 deduplication transition is explicit and same-schema count mismatches remain blocking

Fixes #518
Related: microsoft/vscode-python-environments#1703

Validation

• cargo test -p pet-conda --test environment_locations_test
• cargo test --features ci-perf --test e2e_performance test_performance_summary --no-run
• python -B -m unittest discover -s scripts/tests -p 'test_*.py' -v (47 passed)
• ./scripts/rust-precommit.ps1
• replayed the failed Windows snapshot against its exact base: legacy comparison fails at 8/1 vs 10/2; inventory schema v2 passes with all latency metrics within budget

[github-actions]

Performance Report (Linux)

Result: ✅ Within regression budgets

Metric	PR	Baseline	Delta	Change	Blocking budget	Status
Server startup P50	1ms	1ms	+0ms	+0.0%	>5ms and >100%	➖
Server startup P95	1ms	1ms	+0ms	+0.0%	>50ms and >200%	➖
Full refresh P50	55ms	56ms	-1ms	-1.8%	>25ms and >30%	✅
Full refresh P95	58ms	60ms	-2ms	-3.3%	>50ms and >50%	✅
Time to first environment P50	12ms	14ms	-2ms	-14.3%	>20ms and >100%	✅
Time to first environment P95	17ms	21ms	-4ms	-19.0%	>25ms and >100%	✅
Cold refresh P50	140ms	150ms	-10ms	-6.7%	>100ms and >50%	✅

Workload	PR	Baseline
Environments	5	5
Managers	1	1

Inventory schema transition

• Inventory schema transitioned from v1 to v2; exact count matching is skipped for this comparison.

A regression must exceed both the documented absolute and relative budget. Environment and manager inventories must match exactly within the same inventory schema.

[github-actions]

Test Coverage Report (Linux)

Result: ✅ Within regression budget

Metric	PR	Baseline	Delta
Lines	80.815%	80.809%	+0.005pp
Functions	84.806%	84.806%	+0.000pp

Allowed numerical tolerance: 0.01 percentage points.

[github-actions]

Performance Report (macOS)

Result: ✅ Within regression budgets

Metric	PR	Baseline	Delta	Change	Blocking budget	Status
Server startup P50	100ms	105ms	-5ms	-4.8%	>100ms and >50%	✅
Server startup P95	681ms	746ms	-65ms	-8.7%	>750ms and >100%	✅
Full refresh P50	180ms	194ms	-14ms	-7.2%	>100ms and >50%	✅
Full refresh P95	217ms	211ms	+6ms	+2.8%	>300ms and >100%	🔺
Time to first environment P50	122ms	126ms	-4ms	-3.2%	>150ms and >50%	✅
Time to first environment P95	142ms	171ms	-29ms	-17.0%	>250ms and >100%	✅
Cold refresh P50	397ms	383ms	+14ms	+3.7%	>250ms and >50%	🔺

Workload	PR	Baseline
Environments	10	10
Managers	1	1

Inventory schema transition

• Inventory schema transitioned from v1 to v2; exact count matching is skipped for this comparison.

A regression must exceed both the documented absolute and relative budget. Environment and manager inventories must match exactly within the same inventory schema.

[github-actions]

Performance Report (Windows)

Result: ✅ Within regression budgets

Metric	PR	Baseline	Delta	Change	Blocking budget	Status
Server startup P50	9ms	8ms	+1ms	+12.5%	>10ms and >50%	🔺
Server startup P95	12ms	12ms	+0ms	+0.0%	>50ms and >100%	➖
Full refresh P50	149ms	151ms	-2ms	-1.3%	>150ms and >50%	✅
Full refresh P95	160ms	159ms	+1ms	+0.6%	>250ms and >100%	🔺
Time to first environment P50	23ms	20ms	+3ms	+15.0%	>25ms and >50%	🔺
Time to first environment P95	40ms	29ms	+11ms	+37.9%	>100ms and >100%	🔺
Cold refresh P50	149ms	153ms	-4ms	-2.6%	>150ms and >50%	✅

Workload	PR	Baseline
Environments	8	10
Managers	1	2

Inventory schema transition

• Inventory schema transitioned from v1 to v2; exact count matching is skipped for this comparison.

A regression must exceed both the documented absolute and relative budget. Environment and manager inventories must match exactly within the same inventory schema.

[github-actions]

Test Coverage Report (Windows)

Result: ✅ Within regression budget

Metric	PR	Baseline	Delta
Lines	78.229%	78.207%	+0.022pp
Functions	81.573%	81.533%	+0.041pp

Allowed numerical tolerance: 0.01 percentage points.

[Copilot]

Pull request overview

This PR updates PET’s snapshot/performance gating and Windows Conda discovery to support a deliberate “inventory schema” transition (v1 → v2) while eliminating duplicate Conda installs discovered under casing aliases on Windows.

Changes:

• Add an inventory_schema_version field to performance snapshots and gate inventory count comparisons on matching schema versions (with explicit v1→v2 transition reporting).
• Normalize and deduplicate Windows known Conda install locations after case normalization, preserving on-disk leaf casing.
• Add regression tests for the inventory-schema transition logic and Windows Conda casing-alias deduplication; update docs accordingly.

Show a summary per file

File	Description
scripts/tests/test_quality_snapshot.py	Adds snapshot helper support + unit tests covering inventory schema transition behavior.
scripts/quality_snapshot.py	Introduces inventory schema versioning, transition reporting, and conditional inventory count matching.
docs/QUALITY_SNAPSHOTS.md	Documents the new inventory schema semantics and the one-time v1→v2 transition behavior.
crates/pet/tests/e2e_performance.rs	Emits inventory_schema_version in the performance JSON output.
crates/pet-conda/tests/environment_locations_test.rs	Adds a Windows-only regression test for casing-alias deduplication/preserved casing.
crates/pet-conda/src/environment_locations.rs	Implements Windows leaf-casing restoration + post-normalization deduplication for known Conda install paths.
crates/pet-conda/Cargo.toml	Adds tempfile for the new Windows-only test.
Cargo.lock	Records the new tempfile dependency in the lockfile.

Review details

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

• Files reviewed: 7/8 changed files
• Comments generated: 0
• Review effort level: Lite