Fix generics creation time and allow model name reusing

[Bobronium]

Change Summary

Creation time:

• Use sys._getframe() instead of slow inspect.stack()
• Use frame.f_globals instead of slow inspect.getmodule() to retrieve caller module name
• Combine get_caller_module_name() with is_call_from_module() into more than 2000x 🙈 faster get_caller_frame_info()

Concrete names reusing:

• Append '_' to global reference name if it's already in use instead of raising TypeError

Related issue number

fix #2070
fix #2069

(updated: added "fix" so I don't forget to close both issues, @samuelcolvin)

Checklist

• Unit tests for the changes exist
• Tests pass on CI and coverage remains at 100%
• Documentation reflects the changes where applicable
• changes/<pull request or issue id>-<github username>.md file added describing change
  (see changes/README.md for details)

[codecov]

Codecov Report

Merging #2078 into master will not change coverage.
The diff coverage is 100.00%.

@@            Coverage Diff            @@
##            master     #2078   +/-   ##
=========================================
  Coverage   100.00%   100.00%           
=========================================
  Files           21        21           
  Lines         4094      4091    -3     
  Branches       821       822    +1     
=========================================
- Hits          4094      4091    -3     

Impacted Files	Coverage Δ
pydantic/generics.py	100.00% <100.00%> (ø)
pydantic/fields.py	100.00% <0.00%> (ø)
pydantic/decorator.py	100.00% <0.00%> (ø)
pydantic/dataclasses.py	100.00% <0.00%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 95435de...fe88e7f. Read the comment docs.

[PrettyWood]

Thanks for also taking care of this issue too besides performance one! 🎉

[PrettyWood]

This probably needs to be added in v1.7.2 once last tests to keep coverage at 100% are added and @samuelcolvin approves

[Bobronium]

Coverage should be fixed now

Here's a benchmark showcasing the difference in speed between 1.7.1, 1.6.x and current PR:
https://repl.it/repls/HorizontalConcernedWatchdog#main.py

@bogdandm, could you run this one in your environment as well?

[Bobronium]

Just decided to compare speed of calling get_caller_frame_info() vs get_caller_module_name() with is_call_from_module():

import sys
from timeit import timeit

def get_caller_frame_info_new():
    previous_caller_frame = sys._getframe(2)
    frame_globals = previous_caller_frame.f_globals
    return frame_globals.get('__name__'), previous_caller_frame.f_locals is frame_globals

def get_caller_frame_info_old():
    # code of two functions placed in one and results are combined
    import inspect

    previous_caller_frame = inspect.stack()[2].frame
    previous_caller_module = inspect.getmodule(previous_caller_frame, previous_caller_frame.f_code.co_filename)
    name = previous_caller_module.__name__ if previous_caller_module is not None else None

    import inspect
    previous_caller_frame = inspect.stack()[2].frame
    called_globally = previous_caller_frame.f_locals is previous_caller_frame.f_globals

    return name, called_globally

def check():
    assert get_caller_frame_info_new() == get_caller_frame_info_old()

check()
old = timeit(get_caller_frame_info_old, number=1000)
new = timeit(get_caller_frame_info_new, number=1000)
print('new: ', new)
print('old: ', old)
print('old / new: ', old / new)

I'm getting this results

new:  0.00040132100000001003
old:  1.176067545
old / new:  2930.490916249014

Three. Thousand. Times. Faster. ON SSD. 🙈

I don't know what to say. Previous implementation is the definition of something unreasonably slow...

[samuelcolvin]

This is looking great, I'm going to play with it myself but it looks good.

Would also be good to wait for confirmation from @bogdandm.

[samuelcolvin]

Once this is merged, I'll release v1.7.2. If there's anything else we should include, let me know.

[bogdandm]

@MrMrRobat Could confirm that it is working (similar HDD setup, but a little bit faster disk I think)

new:  0.0005787028931081295
old:  1.042893337085843
old / new:  1802.1222107334108

And 2000 times faster on SSD as well.
Also just checked it by installing via pip from commit hash - it is as fast as 1.5.1 with my code example from the issue.

[samuelcolvin]

Awesome. I'll make a new release in the morning.