Preserve saved globals when copying DelayEvaluator

[Hashim1999164]

PR Type

[x] Bug fix
[ ] New feature
[ ] Core Runtime change (higher bar; see CONTRIBUTING.md)
[ ] Docs / tooling
[ ] Refactoring

Summary

Attribute and item access on DelayEvaluator copy the object and previously dropped the caller globals captured by config_expr. Custom functions then raised NameError after any chained access.

Issue

Fixes #3345

Reproduction

Runtime: local

Commands to run:

PYTHONPATH=. pytest test/unit/test_delay_evaluator.py

Where evidence shows up: unit test output

Before (error / log snippet)

After copy or attribute access, DelayEvaluator._globals became None, so eval of config_expr("my_func()") lost my_func and raised NameError.

After (evidence that fix works)

test_copy_preserves_saved_globals PASSED
test_deepcopy_preserves_saved_globals PASSED
test_getattr_preserves_saved_globals PASSED
test_getitem_preserves_saved_globals PASSED
test_chained_call_uses_saved_globals PASSED
test_copied_call_uses_saved_globals PASSED

Root Cause

copy and deepcopy constructed DelayEvaluator(self._config_expr) without saved_globals. getattr and getitem call copy, so chaining dropped the globals config_expr had stored.

Why This Fix Is Correct

Pass saved_globals=self._globals into both copy paths so the reference is kept, matching the previous comment that globals stay as a reference. Scope is limited to those two methods plus unit tests.

Failure Modes Considered

1. Deep copy still keeps globals by reference (intentional; same object identity as before for globals).
2. Evaluators created without saved_globals still copy None correctly.

Tests

[x] Unit tests added/updated
[ ] Reproduction script provided (required for Core Runtime)
[ ] CI passes
[ ] If tests are impractical: explain why below and provide manual evidence above

Non Goals

No change to eval semantics beyond preserving the globals reference across copy.

AI Tool Usage

[x] No AI tools were used in this contribution
[ ] AI tools were used (describe below)

[greptile-apps]

Greptile Summary

The PR preserves caller globals by reference when copying or deep-copying a DelayEvaluator, fixing chained attribute and item evaluation of custom expressions.

• Passes saved_globals through both copy implementations.
• Adds regression tests for copying, deep-copying, attribute and item access, and result-level chained evaluation.

Confidence Score: 5/5

The PR appears safe to merge.

No blocking failure remains; the previously requested chained result-level evaluation is now covered and would fail under the original globals-dropping behavior.

Important Files Changed

Filename	Overview
metaflow/user_configs/config_parameters.py	Copy and deep-copy operations now retain the saved globals reference while preserving the existing access-copy behavior.
test/unit/test_delay_evaluator.py	Adds focused regression coverage, including a chained call that reaches evaluation and asserts the resolved result.

_{Reviews (2): Last reviewed commit: "Cover chained DelayEvaluator evaluation ..." | Re-trigger Greptile}

[Shriprasad-P]

I independently reproduced the namespace-loss issue and verified that the implementation fixes it.

Preserving saved_globals by reference in both __copy__ and __deepcopy__ matches the existing DelayEvaluator semantics. Chained attribute/item evaluation that fails with a NameError on the base revision succeeds on this PR, while the copied access state remains independent.

One non-blocking coverage gap remains: the new tests verify _globals / _access directly but never call the copied or chained DelayEvaluator. Adding one regression test that evaluates the chained expression and asserts its returned value would cover the actual user-visible failure path rather than only the internal state.

Small comment nit as well: the new __copy__ comment uses config_expr("my_func()") as the example for attribute/item access, but non-identifier expressions have _access=None and don't take that chaining path. An identifier-based example such as config_expr("my_func").project would describe the affected path more accurately.

The implementation itself looks correct.

[Hashim1999164]

@Shriprasad-P yeah fair those tests were only poking at _globals. added one that actually calls the chained evaluator and checks the return value. also swapped the comment to config_expr("my_func").project, my_func() was the wrong example there

[SahilKumar75]

Thanks for preparing the implementation. I originally reported this issue with the same fix and had a local patch and tests ready, but I was waiting for maintainer acknowledgement as required by the contribution guidelines. I am happy to help review or test this change if useful.

[Hashim1999164]

@SahilKumar75 ah i didnt know you already had a patch sitting around. this was just copy dropping saved_globals. if you wanna poke at the tests go for it.