CI: Optimize benchmarks by cloning template environments

[jezdez]

Summary

This PR optimizes benchmark runtime by cloning a pre-created template environment instead of running conda create for each test.

Problem

The benchmark tests call conda create 5 times (3x in test_convert_tree, 2x in test_build_conda). Each conda create invokes the solver and extracts packages, taking ~10 seconds each.

Solution

1. Session-scoped template: Create a Python environment once at the start of the test session
2. Fast cloning: Clone the template using conda create --clone instead of raw file copying

Why conda create --clone instead of raw file copy?

Raw file copying (cp, shutil.copytree, etc.) doesn't properly handle conda's prefix relocation:

• conda-meta/*.json files contain hardcoded prefix paths
• Scripts have shebang lines pointing to the original environment
• Some packages store paths in config files

conda create --clone properly handles all of this while still being faster than a full create because it:

• Skips the solver (no SAT solving needed)
• Skips downloading (packages already cached)
• Only needs to relocate and link packages

Measured Performance Improvement

Branch	Benchmark Duration	Job Link
main (baseline)	17m 39s	benchmarks py312
This PR	10m 47s	benchmarks py312
Improvement	6m 52s faster (39%)

Changes

• tests/conftest.py:
  • Add python_template_env session-scoped fixture using session_conda_cli
  • Template creation failures propagate immediately (fail fast)
• tests/test_benchmarks.py:
  • Update tests to use python_template_env fixture
  • Clone from template using conda create --clone
  • Use idiomatic nonlocal for setup counters instead of list workaround

Test plan

• All benchmark tests pass locally
• CI passes on all platforms (Linux, macOS, Windows)
• Benchmarks complete successfully with CodSpeed
• Verified ~39% benchmark runtime improvement

[codspeed-hq]

CodSpeed Performance Report

Merging this PR will not alter performance

_{Comparing optimize-benchmarks-clone-env (de83b98) with main (3ff94a4)}

Summary

✅ 5 untouched benchmarks

[jezdez]

Validation Results ✅

The conda create --clone optimization has been validated. Here's the comparison:

Benchmark Job Runtime

Branch	Duration	Job Link
main (baseline)	17m 56s	benchmarks py312
This PR	10m 20s	benchmarks py312
Improvement	7m 36s faster (42%)

Full Workflow Comparison

Workflow	main	This PR
Test	18m 8s	15m 43s

Why this works

The original approach used raw file copying (cp, shutil.copytree) which doesn't handle conda's prefix relocation. This was fixed to use conda create --clone which:

1. Skips the solver - No SAT solving needed since we're copying an existing solution
2. Skips downloading - Packages are already cached from template creation
3. Properly relocates prefixes - Updates conda-meta/*.json and fixes shebang lines in scripts

The improvement is larger than initially estimated because:

• Template creation cost is amortized across all benchmark iterations
• Solver avoidance saves more time than package extraction alone
• Each benchmark runs multiple setup/teardown cycles

All tests pass on all platforms (Linux, macOS, Windows).

[jezdez]

This PR is more of a nice-to-have, than a hard requirement for 0.4.0

[soapy1]

Nice speed up ⏩

[jezdez]

Addressed all review feedback:

• Removed try/except fallback from python_template_env fixture — if template creation fails, the test session now fails fast with a clear error instead of silently yielding None and deferring the failure to each individual test.
• Replaced setup_counter = [0] with idiomatic nonlocal setup_counter in both test_convert_tree and test_build_conda.
• Removed conditional clone-or-create logic — since the fixture now always provides a valid path (or raises), the tests always use conda create --clone directly.

Also merged latest main and updated the PR description with fresh benchmark numbers (17m 39s → 10m 47s, ~39% improvement).

[soapy1]

⏩