Performance overhead when importing modules using "code" instead of "globalCode"

[KasparMetsa]

Used Claude to find a problem with syntax like this:

functions:
  # Import inside code: (runs on every message - inefficient but works)
  log_action_count:
    type: forEach
    code: |
      from processor_functions import log_user_action_count
      log_user_action_count(key, value, log)

The Problem

When using code: for imports, the import statement executes on every message:

# Generated code when using code: for import
def log_action_count(key, value):
    from processor_functions import log_user_action_count  # RUNS EVERY CALL
    log_user_action_count(key, value, log)

When using globalCode:, the import executes once at startup:

# Generated code when using globalCode: for import
from processor_functions import log_user_action_count  # RUNS ONCE

def log_action_count(key, value):
    log_user_action_count(key, value, log)

Here is full description, proof and possible solutions by Claude

PR_422_IMPORT_OVERHEAD_ANALYSIS.md

Here is benchmark by Clade to prove the overhead:

benchmark_import.py

When I ran it, I got this output:

benchmark.txt