Use lazy imports on Python 3.15 to improve startup speed

[hugovk]

Replaces and closes #608.

The upcoming Python 3.15 (now in beta) introduces a new lazy keyword for imports:

lazy import json

With this, the import doesn't actually happen until the first use:

json.dumps(thing)

So there's no need for the deferred method used in #608.

This is new syntax in 3.15, if you use it on an older version you get a syntax error. But there's a transitional mechanism: if you declare the imports in __lazy_modules__, they'll be treated as if they're lazy in 3.15+, and ignored for older versions.

• https://docs.python.org/3.15/whatsnew/3.15.html#whatsnew315-lazy-imports
• https://peps.python.org/pep-0810/

Running with Python 3.15:

python3 -X importtime -c 'import tablib' 2> import.log
pip install tuna
tuna import.log

Before: 8 ms

After: ~0 ms

Even better than #608!

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 93.19%. Comparing base (b8c6606) to head (4c09d9b).

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #645      +/-   ##
==========================================
+ Coverage   93.16%   93.19%   +0.03%     
==========================================
  Files          29       29              
  Lines        3231     3247      +16     
==========================================
+ Hits         3010     3026      +16     
  Misses        221      221              

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[claudep]

Interesting, really! I wonder however what's the cost of "hard" importing the Python's own modules, like io, decimal, operator. Aren't essentially external modules that are costly to import?

[hugovk]

The light blue boxes in the top chart are stdlib modules, and we can see many of them are contributing real cost to the import tree.

As linked in #608, we've been deferring imports inside the stdlib of other stdlib modules to help reduce this cost, and in 3.15+ using the lazy keyword instead (for example, python/cpython#147992, python/cpython#149318, python/cpython#149416, python/cpython#146327). re is a common target.

The stdlib has some benefit: they're usually byte-compiled to pyc at install time, but that's a cheap one-time cost for external modules, and doesn't affect costly execution time. Some modules are always imported as part of site (but that can be disabled). Some bootstrap modules are "frozen" into the binary to avoid file lookup.

Looking the top chart again, most of the lower edge underneath tablib is light blue, so most of the import time comes from importing stdlib.

[claudep]

Thanks for the detailed explanations 😍