From post by @AlexWaygood in #117372 (comment)_:

I'm a little dismayed that dataclasses adds so much overhead. Does that imply that dataclasses shouldn't be used in the stdlib?

dataclasses is an unfortunately heavy import, yes. Although it imports relatively few modules itself, one of those modules is inspect, which has a huge import time (inspect imports half the standard library). In the past, I've looked into removing the dataclasses dependency on inspect, but couldn't see a way to do so without making the code significantly more ugly; the same goes when it comes to improving the import time for inspect itself. (In hindsight, I'd argue inspect should probably have been a package rather than a huge single-file Python module... but hindsight is 20/20.)

Perhaps inspect could be refactored to limit the import-time cost.

Currently on my m3 mac pro, on the second attempt, the import times are 1.2ms and 7.9ms (cumulative):

 cpython main @ ./python.exe -X importtime -c 'import inspect' 2> /dev/null ; ./python.exe -X importtime -c 'import inspect' 2>1 | grep -E '(cumul|inspect)'
import time: self [us] | cumulative | imported package
import time:      1184 |       7892 | inspect

I've searched the issue tracker and didn't see anything tracking this concern, so right now it's a naïve feature request.

Related: #108901

Linked PRs

• gh-117865: Defer several imports in inspect.py #119526
• gh-117865: Defer import of re in ast #119546
• gh-117865: Speedup import of inspect module #144756