ENH: Add py-rattler

[HaoZeke]

As it says on the tin. Some notes:

• This is much faster than both libmambapy and conda
• It is more featureful than virtualenv (supports installing python versions)
• Has a well documented API
• Is supported by Python >= 3.8

Regarding the version bump; asv itself is no longer installed into the project, only asv-runner, which is used to actually drive the benchmarks, and that is still pure Python 3.7 as I recall, so it should be fine.

[HaoZeke]

I'll be splitting this PR into issues and PRs. However, for the section remaining (implementing rattler):

mkdir tmp_asv; cd tmp_asv
pipx run asv quickstart # Choose the first defaults
pipx run pdm init # Also the defaults

We'll need to add the channels necessary for mamba, i.e.

diff --git c/asv.conf.json w/asv.conf.json
index 32946ee..2933f54 100644
--- c/asv.conf.json
+++ w/asv.conf.json
@@ -69,7 +69,7 @@
 
     // The list of conda channel names to be searched for benchmark
     // dependency packages in the specified order
-    // "conda_channels": ["conda-forge", "defaults"],
+    "conda_channels": ["conda-forge"],
 
     // A conda environment file that is used for environment creation.
     // "conda_environment_file": "environment.yml",

Timing data

This is a "pseudo-cold" cache, in that I've not removed local caches (e.g. ~, so it isn't as accurate as it ought to be.

hyperfine 'asv run -E rattler' 'asv run -E mamba' 'asv run -E virtualenv' 'asv run -E conda'
Benchmark 1: asv run -E rattler
  Time (mean ± σ):      3.668 s ±  5.100 s    [User: 3.057 s, System: 0.493 s]
  Range (min … max):    1.794 s … 18.177 s    10 runs
 
  Warning: The first benchmarking run for this command was significantly slower than the rest (18.177 s). This could be caused by (filesystem) caches that were not filled until after the first run. You should consider using the '--warmup' option to fill those caches before the actual benchmark. Alternatively, use the '--prepare' option to clear the caches before each timing run.
 
Benchmark 2: asv run -E mamba
  Time (mean ± σ):      6.739 s ± 14.599 s    [User: 5.273 s, System: 0.632 s]
  Range (min … max):    1.870 s … 48.286 s    10 runs
 
  Warning: The first benchmarking run for this command was significantly slower than the rest (48.286 s). This could be caused by (filesystem) caches that were not filled until after the first run. You should consider using the '--warmup' option to fill those caches before the actual benchmark. Alternatively, use the '--prepare' option to clear the caches before each timing run.
 
Benchmark 3: asv run -E virtualenv
  Time (mean ± σ):      3.495 s ±  4.561 s    [User: 2.895 s, System: 0.341 s]
  Range (min … max):    1.878 s … 16.470 s    10 runs
 
  Warning: The first benchmarking run for this command was significantly slower than the rest (16.470 s). This could be caused by (filesystem) caches that were not filled until after the first run. You should consider using the '--warmup' option to fill those caches before the actual benchmark. Alternatively, use the '--prepare' option to clear the caches before each timing run.
 
Benchmark 4: asv run -E conda
  Time (mean ± σ):      6.268 s ± 13.187 s    [User: 5.160 s, System: 0.529 s]
  Range (min … max):    1.944 s … 43.799 s    10 runs
 
  Warning: The first benchmarking run for this command was significantly slower than the rest (43.799 s). This could be caused by (filesystem) caches that were not filled until after the first run. You should consider using the '--warmup' option to fill those caches before the actual benchmark. Alternatively, use the '--prepare' option to clear the caches before each timing run.
 
Summary
  asv run -E virtualenv ran
    1.05 ± 2.00 times faster than asv run -E rattler
    1.79 ± 4.44 times faster than asv run -E conda
    1.93 ± 4.88 times faster than asv run -E mamba

Once the environments are present, as expected, there's no real difference:

hyperfine 'asv run -E rattler' 'asv run -E mamba' 'asv run -E virtualenv' 'asv run -E conda'
Benchmark 1: asv run -E rattler
  Time (mean ± σ):      2.121 s ±  0.161 s    [User: 1.838 s, System: 0.190 s]
  Range (min … max):    1.900 s …  2.336 s    10 runs
 
Benchmark 2: asv run -E mamba
  Time (mean ± σ):      2.082 s ±  0.145 s    [User: 1.765 s, System: 0.191 s]
  Range (min … max):    1.849 s …  2.342 s    10 runs
 
Benchmark 3: asv run -E virtualenv
  Time (mean ± σ):      2.122 s ±  0.095 s    [User: 1.842 s, System: 0.188 s]
  Range (min … max):    1.975 s …  2.268 s    10 runs
 
Benchmark 4: asv run -E conda
  Time (mean ± σ):      2.089 s ±  0.109 s    [User: 1.805 s, System: 0.184 s]
  Range (min … max):    1.928 s …  2.266 s    10 runs
 
Summary
  asv run -E mamba ran
    1.00 ± 0.09 times faster than asv run -E conda
    1.02 ± 0.10 times faster than asv run -E rattler
    1.02 ± 0.08 times faster than asv run -E virtualenv

virtualenv is quicker by a hair (not really statistically significant) but rattler also provides more functionality (compared to the equivalent conda and mamba helpers).

On another benchmark (asv_samples):

hyperfine 'rm -rf .asv && asv run -E rattler' 'rm -rf .asv && asv run -E mamba' 'rm -rf .asv && asv run -E virtualenv' 'rm -rf .asv && asv run -E conda' 
Benchmark 1: rm -rf .asv && asv run -E rattler
  Time (mean ± σ):     40.837 s ±  2.582 s    [User: 31.366 s, System: 7.414 s]
  Range (min … max):   38.861 s … 47.478 s    10 runs
 
Benchmark 2: rm -rf .asv && asv run -E mamba
 ⠇ Initial time measurement       ░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░ ETA 00:00: ⠦ ⠙ Initial time measurement       ░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░ ETA 00:0  Time (mean ± σ):     66.351 s ±  8.670 s    [User: 44.504 s, System: 7.894 s]
  Range (min … max):   59.911 s … 89.377 s    10 runs
 
Benchmark 3: rm -rf .asv && asv run -E virtualenv
  Time (mean ± σ):     37.522 s ±  0.880 s    [User: 28.702 s, System: 3.686 s]
  Range (min … max):   36.334 s … 39.181 s    10 runs
 
Benchmark 4: rm -rf .asv && asv run -E conda
  Time (mean ± σ):     53.831 s ±  8.633 s    [User: 41.895 s, System: 5.632 s]
  Range (min … max):   49.172 s … 77.869 s    10 runs
 
  Warning: The first benchmarking run for this command was significantly slower than the rest (77.869 s). This could be caused by (filesystem) caches that were not filled until after the first run. You should consider using the '--warmup' option to fill those caches before the actual benchmark. Alternatively, use the '--prepare' option to clear the caches before each timing run.
 
Summary
  rm -rf .asv && asv run -E virtualenv ran
    1.09 ± 0.07 times faster than rm -rf .asv && asv run -E rattler
    1.43 ± 0.23 times faster than rm -rf .asv && asv run -E conda
    1.77 ± 0.23 times faster than rm -rf .asv && asv run -E mamba

Note that these are all extremely trivial packages, with only a single python version and numpy as requirements, larger, more complex projects are expected to benefit from this more.

[mattip]

Is CI expected to fail on macOS (python3.8, pypy3.10)?