copy.deepcopy() takes 55% of runtime of a simple benchmark

[grigi]

Hi, we are working on Tortoise ORM and I did a series of simple benchmarks to start finding performance bottlenecks. In the one transactioned insert test:
https://github.com/tortoise/orm-benchmarks#interesting-profiling-results
I found that copy.deepcopy() takes ~55% of the runtime. Changing it to copy.copy() gives me an 83% speed up in that particular test.

Of course it is there for a reason as without it we can get unexpected side-effects, such as what the pypika.tests.test_joins.JoinBehaviorTests.test_immutable__queries_after_join test determines.

I'm just wondering what can we do to speed this up?

[twheys]

Deepcopy could be replaced with a more comprehensive implementation of __copy__ in the QueryBuilder class. Can you try using the changes in my PR #161 in your benchmark and let me know if you see a similar improvement?

[grigi]

Thanks @twheys
I did the test and got a 53% speedup in that particular benchmark.
Thank you, this is a big gain 👍

[twheys]

Is that sufficient? Could probably try and optimize a bit better. The copies are used to keep the queries immutable.