Comparator support for tensorflow, dict views, slices, 0-d numpy arrays, torch dtypes

[misrasaurabh1]

PR Type

Enhancement, Tests

---

Description

• Add TensorFlow object comparisons

• Support dict view comparisons

• Handle numpy 0-d and datetime64/ timedelta64

• Compare torch.dtype and slice objects

---

Diagram Walkthrough

flowchart LR
  C["comparator(): core comparator"] --> TF["TensorFlow types support"]
  C --> DV["dict view (keys/values/items) support"]
  C --> NP["NumPy 0-d, datetime64/timedelta64"]
  C --> TORCHDT["torch.dtype equality"]
  C --> SL["slice handling"]
  TF -- "tensor/variable/sparse/ragged/shape/dtype" --> C

Loading

File Walkthrough

	Relevant files
Enhancement	comparator.py Expand comparator support for TF, dict views, numpy, torch, slice codeflash/verification/comparator.py Detect TensorFlow availability Add comparisons for tf Tensor, Variable, DType, TensorShape, SparseTensor, RaggedTensor Add dict view handling: keys, values, items Add numpy datetime64/timedelta64, 0-d array handling Add torch.dtype equality Include slice in simple comparable types +72/-0
Tests	test_comparator.py Add comprehensive tests for new comparator types                  tests/test_comparator.py Add tests for torch.dtype equality Add tests for dict_keys/values/items comparisons Add comprehensive TensorFlow tests: tensors, dtypes, variables, shapes, sparse, ragged Add tests for slice comparisons Add tests for numpy datetime64/timedelta64 and 0-d arrays +649/-2

---

[github-actions]

PR Reviewer Guide 🔍

Here are some key observations to aid the review process:

⏱️ Estimated effort to review: 3 🔵🔵🔵⚪⚪

🧪 PR contains tests

🔒 No security concerns identified

⚡ Recommended focus areas for review Possible Issue TensorFlow handling compares only based on the type of orig but doesn't validate that new is the same TF type before accessing attributes like .dtype, .shape, or calling .numpy(). If new is a non-TF object, this may raise AttributeError instead of returning False. Consider guarding with paired isinstance checks for new. if HAS_TENSORFLOW: import tensorflow as tf # type: ignore # noqa: PGH003 if isinstance(orig, tf.Tensor): if orig.dtype != new.dtype: return False if orig.shape != new.shape: return False # Use numpy conversion for proper NaN handling return comparator(orig.numpy(), new.numpy(), superset_obj) if isinstance(orig, tf.Variable): if orig.dtype != new.dtype: return False if orig.shape != new.shape: return False return comparator(orig.numpy(), new.numpy(), superset_obj) if isinstance(orig, tf.dtypes.DType): return orig == new if isinstance(orig, tf.TensorShape): return orig == new if isinstance(orig, tf.SparseTensor): if not comparator(orig.dense_shape.numpy(), new.dense_shape.numpy(), superset_obj): return False return ( comparator(orig.indices.numpy(), new.indices.numpy(), superset_obj) and comparator(orig.values.numpy(), new.values.numpy(), superset_obj) ) if isinstance(orig, tf.RaggedTensor): if orig.dtype != new.dtype: return False if orig.shape.rank != new.shape.rank: return False return comparator(orig.to_list(), new.to_list(), superset_obj) Consistency Dict view handling treats dict_items as list comparison (order-sensitive) despite comment stating set-like behavior. This may cause false negatives when order differs but items are equivalent. Consider comparing as sets of tuples, similar to keys. # Handle dict view types (dict_keys, dict_values, dict_items) # Use type name checking since these are not directly importable types type_name = type(orig).__name__ if type_name == "dict_keys": # dict_keys can be compared as sets (order doesn't matter) return comparator(set(orig), set(new)) if type_name == "dict_values": # dict_values need element-wise comparison (order matters) return comparator(list(orig), list(new)) if type_name == "dict_items": # dict_items can be compared as sets of tuples (order doesn't matter for items) return comparator(list(orig), list(new)) Edge Case NumPy datetime64/timedelta64 equality relies on direct ==, which can be unit-sensitive across versions. Consider normalizing units or converting to integers with a common unit when possible, or documenting behavior. if isinstance(orig, (np.datetime64, np.timedelta64)): # Handle NaT (Not a Time) - numpy's equivalent of NaN for datetime if np.isnat(orig) and np.isnat(new): return True if np.isnat(orig) or np.isnat(new): return False return orig == new

[github-actions]

PR Code Suggestions ✨

Explore these optional code suggestions:

Category	Suggestion	Impact
Possible issue	Guard TensorFlow type mismatches Add type checks to ensure new is the same TensorFlow type before accessing its attributes. Without this, comparing a TF object with a non-TF object can raise AttributeError. Return False early if types mismatch. codeflash/verification/comparator.py [103-141] if HAS_TENSORFLOW: import tensorflow as tf # type: ignore # noqa: PGH003 if isinstance(orig, tf.Tensor): + if not isinstance(new, tf.Tensor): + return False if orig.dtype != new.dtype: return False if orig.shape != new.shape: return False - # Use numpy conversion for proper NaN handling return comparator(orig.numpy(), new.numpy(), superset_obj) if isinstance(orig, tf.Variable): + if not isinstance(new, tf.Variable): + return False if orig.dtype != new.dtype: return False if orig.shape != new.shape: return False return comparator(orig.numpy(), new.numpy(), superset_obj) if isinstance(orig, tf.dtypes.DType): - return orig == new + return isinstance(new, tf.dtypes.DType) and (orig == new) if isinstance(orig, tf.TensorShape): - return orig == new + return isinstance(new, tf.TensorShape) and (orig == new) if isinstance(orig, tf.SparseTensor): + if not isinstance(new, tf.SparseTensor): + return False if not comparator(orig.dense_shape.numpy(), new.dense_shape.numpy(), superset_obj): return False return ( - comparator(orig.indices.numpy(), new.indices.numpy(), superset_obj) and - comparator(orig.values.numpy(), new.values.numpy(), superset_obj) + comparator(orig.indices.numpy(), new.indices.numpy(), superset_obj) + and comparator(orig.values.numpy(), new.values.numpy(), superset_obj) ) if isinstance(orig, tf.RaggedTensor): + if not isinstance(new, tf.RaggedTensor): + return False if orig.dtype != new.dtype: return False if orig.shape.rank != new.shape.rank: return False return comparator(orig.to_list(), new.to_list(), superset_obj) Suggestion importance[1-10]: 8 __ Why: Correctly adds symmetric type checks for TensorFlow objects to prevent attribute access on mismatched types and aligns with the comparator's strict-type behavior; impactful for correctness and safety.	Medium
	Guard numpy type mismatches Add type checks ensuring new is a compatible numpy type before attribute access. Accessing new.dtype/new.shape or using np.isnat(new) when new isn't a numpy object can raise AttributeError or TypeError; return False early on mismatch. codeflash/verification/comparator.py [189-214] if HAS_NUMPY: import numpy as np # type: ignore # noqa: PGH003 if isinstance(orig, (np.datetime64, np.timedelta64)): - # Handle NaT (Not a Time) - numpy's equivalent of NaN for datetime + if not isinstance(new, (np.datetime64, np.timedelta64)): + return False if np.isnat(orig) and np.isnat(new): return True if np.isnat(orig) or np.isnat(new): return False - return orig == new + return bool(orig == new) if isinstance(orig, np.ndarray): + if not isinstance(new, np.ndarray): + return False if orig.dtype != new.dtype: return False if orig.shape != new.shape: return False - # Handle 0-d arrays specially to avoid "iteration over a 0-d array" error if orig.ndim == 0: try: - return np.allclose(orig, new, equal_nan=True) + return bool(np.allclose(orig, new, equal_nan=True)) except Exception: return bool(orig == new) try: - return np.allclose(orig, new, equal_nan=True) + return bool(np.allclose(orig, new, equal_nan=True)) except Exception: # fails at "ufunc 'isfinite' not supported for the input types" + ... Suggestion importance[1-10]: 8 __ Why: Adds necessary type checks before numpy-specific attribute/method access, avoiding exceptions and enforcing strict comparability; accurate and materially improves robustness.	Medium
	Validate dict view type symmetry Ensure the right-hand operand is a matching dict view before converting. As written, comparing a dict view to an arbitrary iterable may incorrectly return True or raise unexpected errors. Return False when new is not the same view type. codeflash/verification/comparator.py [176-186] # Handle dict view types (dict_keys, dict_values, dict_items) -# Use type name checking since these are not directly importable types type_name = type(orig).__name__ if type_name == "dict_keys": - # dict_keys can be compared as sets (order doesn't matter) + if type(new).__name__ != "dict_keys": + return False return comparator(set(orig), set(new)) if type_name == "dict_values": - # dict_values need element-wise comparison (order matters) + if type(new).__name__ != "dict_values": + return False return comparator(list(orig), list(new)) if type_name == "dict_items": - # dict_items can be compared as sets of tuples (order doesn't matter for items) + if type(new).__name__ != "dict_items": + return False return comparator(list(orig), list(new)) Suggestion importance[1-10]: 7 __ Why: Ensures dict view comparisons don't silently coerce arbitrary iterables, preventing false positives and potential errors; a solid maintainability and correctness improvement.	Medium