visitor: deepcopy literal value so each search returns a fresh object (fixes #318)

[qianbkk]

Fixes #318.

What

Parser caches the parsed AST per expression, so the literal node's already-decoded Python value is shared across every call to search(). Returning that object directly meant that two searches of the same literal expression ([], {}, [1, 2], ...) would hand back the same list/dict object -- any in-place mutation by the caller would leak into every subsequent call, including unrelated searches on completely different inputs.

visit_literal now wraps the cached value with copy.deepcopy on every visit. Immutable literals (str / int / float / bool / None) are unaffected; mutable JSON literals become independent per call.

Repro (before)

import jmespath

a = jmespath.search('`[]`', {})
a.extend([1, 2, 3])

b = jmespath.search('`[]`', {})  # expects [], got [1, 2, 3]
b.extend([9, 8, 7])

print(a)  # [1, 2, 3, 9, 8, 7] -- leaked from b

After: a and b are independent lists.

Diff

 def visit_literal(self, node, value):
-    return node['value']
+    return copy.deepcopy(node['value'])

Tests

Four regression tests in tests/test_search.py (new TestLiteralFreshness class):

• test_list_literal_is_independent_per_search
• test_object_literal_is_independent_per_search
• test_nested_list_literal_is_independent (covers [[]] etc.)
• test_string_literal_remains_a_string (control)

Full suite: 995 passed, 1 skipped (excluding tests/legacy/ which is data-only). No existing tests changed.

Performance

copy.deepcopy is only invoked for literal nodes in expressions that use them, and only mutable JSON literals allocate new containers. The common case -- read-only results, immutable literals, expressions that don't use literals at all -- has the same cost as before.

AI disclosure

Prepared with AI assistance (kilo assistant); the diagnosis, fix, and tests were developed and verified locally before submission.

[Copilot]

Copilot review overview

🟡 Changes recommended

Avoid unnecessary immutable-scalar copying and clarify the misleading comment before approval.

Get a fresh assessment by requesting another Copilot review.

Review effort: Lite
Findings: 1

Open (1)

• Correct comment describing deep-copy of decoded literals · New

What changed in this PR

Fixes shared mutable literal results by returning independent values for each search.

Changes:

• Deep-copy cached literal values during visitation.
• Add regression tests for list, object, nested, and string literals.

File	Summary	Review notes
jmespath/​visitor.py	Returns fresh literal values per visit.	Moderate (1 vote): avoid deep-copying immutable scalars. Nit (2 votes): clarify the implementation comment.
tests/​test_search.py	Adds literal freshness regression coverage.	No findings.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.