Respect exclusions before reporting DeepSearch path matches

[FanWu-ai]

DeepSearch reports matching dictionary keys and attribute names before the recursive search checks exclusions. For example, DeepSearch({'needle': 'x'}, 'needle', exclude_paths=["root['needle']"]) currently reports the excluded path in matched_paths.

Check the child object and its path with the existing exclusion predicate before reporting a path match. Add ten regression cases covering exact, wildcard and regex path exclusions, attribute names, excluded child types, retained siblings, and both verbosity levels.

This complements #629: that change fixes the recursive current-object type check, while this fixes the earlier path-reporting step. All eight initial path-exclusion regressions still fail with #629's production change alone.

Validation on Linux, CPython 3.12.14:

• Full suite including slow tests: pytest --cov=deepdiff --runslow --benchmark-disable -q: 1,319 passed, 8 skipped; 95% aggregate coverage.
• Search suite: 72 passed; 100% coverage of deepdiff/search.py.
• pyright deepdiff/search.py: 0 errors, 0 warnings.
• CI-blocking Flake8 rules E9,F63,F7,F82 and git diff --check: passed.
• Full Flake8 on the touched files reports seven pre-existing diagnostics, compared against the unchanged baseline; no new diagnostics.

Other Python versions, operating systems, and hosted CI have not been validated locally.

[seperman]

Static review: #638 Respect exclusions before reporting DeepSearch path matches

Summary: Adds self.__skip_this(obj_child, new_parent) to DeepSearch.__search_dict, before a dictionary key or attribute name is compared with the search term. Excluded paths (exact, glob and regex) and children of excluded types no longer show up in matched_paths. This fixes the reported problem.

Overlap with other PRs

• #629 (open) changes the type check at the top of __search from __skip_this(item, ...) to __skip_this(obj, ...). As the description says, the two PRs complement each other:
  • This PR alone also skips dictionary values and attributes of excluded types, because the new check is given obj_child. But the root object is never type-checked, and __search (deepdiff/search.py:324) still checks the type of the search item. For example, DeepSearch([1], 1, exclude_types=[list]) still reports root[0], and #629's test_skip_root_object_type cases would still fail.
  • #629 alone still reports excluded keys in matched_paths, so the new tests here would fail with only #629.
  • The two merge without a textual conflict (checked with git merge-tree).
• #632 (open) changes the line right after the new check (new_parent_cased = ..., deepdiff/search.py:241 here), and git merge-tree reports a textual conflict there. The fix is to keep both: this check, then #632's or self.use_regexp condition.
• Nothing equivalent is on dev.

Security

No security concerns found in static review.

Correctness

• deepdiff/search.py:239-240: on dev, __search_dict reports a path match (lines 244-250 here) before any exclusion check, because exclusions are only checked inside __search. So DeepSearch({'needle': 'x'}, 'needle', exclude_paths=["root['needle']"]) returns matched_paths: {"root['needle']"}. Running the check before the report fixes this. The fix also covers attributes and namedtuples, since __search_obj goes through __search_dict(print_as_attribute=True).
• Because the check is given obj_child, a key whose value has an excluded type is no longer reported in matched_paths either.
• Minor: each dictionary child now goes through __skip_this twice, once here and once at the top of __search (line 324). Glob parsing (_path_to_elements) is lru_cached, and __search_iterable already checks twice in the same way, so the extra cost should be small.
• Optional: parents_ids_added (line 236) could be computed after the new check, which avoids a frozenset union for every skipped child.

Tests

• The new tests at tests/test_search.py:263, :272 and :281 cover:

  • exact, glob (root[*]['needle']) and regex path exclusions
  • attribute names
  • an excluded value type

  Each runs at verbosity levels 1 and 2. Following dev's code, each case should produce an extra matched_paths entry there, so the tests fail on dev and pass with this change.

Suggestions

• Land this together with, or after, #629 so the root-object case and the item-type check in __search are fixed as well.
• If both this and #632 are merged, the conflict on the neighbouring line needs resolving (see Overlap above).

---

AI-assisted static review; the code was not executed.