From 3d64f1447737512199d09d839f490db3fc5a4063 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Du=C5=A1an=20Jolovi=C4=87?= <dusan.jolovic@proton.me>
Date: Wed, 6 May 2026 04:25:58 +0200
Subject: [PATCH] perf: use hash() as sort key in stable repr and canonicalize
 JSON keys

Replaces repr()-based sorting with hash()-based sorting for sets,
frozensets and dict keys in _stable_repr_to. hash() returns a
pre-computed integer for built-in types, avoiding the string
allocation overhead of repr() while preserving a total order
within the process lifetime.

Also adds sort_keys=True to all json.dumps calls in
_put_commit_row so that tags, dep_versions and input_refs are
stored in a canonical order on disk.

New tests lock in the guarantee that insertion order does not
affect the stable hash for dicts and sets.
---
 src/cashet/hashing.py |  6 +++---
 src/cashet/store.py   |  6 +++---
 tests/test_hashing.py | 20 ++++++++++++++++++++
 3 files changed, 26 insertions(+), 6 deletions(-)

diff --git a/src/cashet/hashing.py b/src/cashet/hashing.py
index fa00f4b..33795f5 100644
--- a/src/cashet/hashing.py
+++ b/src/cashet/hashing.py
@@ -347,7 +347,7 @@ def _stable_repr_to(
         _visited.add(obj_id)
         buf.write("{")
         first = True
-        for item in sorted(obj, key=repr):
+        for item in sorted(obj, key=hash):
             if not first:
                 buf.write(", ")
             first = False
@@ -362,7 +362,7 @@ def _stable_repr_to(
         _visited.add(obj_id)
         buf.write("frozenset({")
         first = True
-        for item in sorted(obj, key=repr):
+        for item in sorted(obj, key=hash):
             if not first:
                 buf.write(", ")
             first = False
@@ -377,7 +377,7 @@ def _stable_repr_to(
         _visited.add(obj_id)
         buf.write("{")
         first = True
-        for key, val in sorted(obj.items(), key=lambda p: repr(p[0])):
+        for key, val in sorted(obj.items(), key=lambda p: hash(p[0])):
             if not first:
                 buf.write(", ")
             first = False
diff --git a/src/cashet/store.py b/src/cashet/store.py
index a80c60a..69a5122 100644
--- a/src/cashet/store.py
+++ b/src/cashet/store.py
@@ -306,20 +306,20 @@ def _put_commit_row(
                 commit.task_def.args_hash,
                 commit.task_def.args_snapshot,
                 commit.task_def.func_source,
-                json.dumps(commit.task_def.dep_versions),
+                json.dumps(commit.task_def.dep_versions, sort_keys=True),
                 int(commit.task_def.cache),
                 commit.task_def.retries,
                 int(commit.task_def.force),
                 timeout_seconds,
                 commit.task_def.ttl.total_seconds() if commit.task_def.ttl else None,
-                json.dumps([r.hash for r in commit.input_refs]),
+                json.dumps([r.hash for r in commit.input_refs], sort_keys=True),
                 output_hash,
                 output_size,
                 output_tier,
                 commit.parent_hash,
                 commit.status.value,
                 commit.error,
-                json.dumps(commit.tags),
+                json.dumps(commit.tags, sort_keys=True),
                 commit.created_at.isoformat(),
                 commit.claimed_at.isoformat(),
                 accessed_at,
diff --git a/tests/test_hashing.py b/tests/test_hashing.py
index 5f2933d..8fb2f86 100644
--- a/tests/test_hashing.py
+++ b/tests/test_hashing.py
@@ -257,6 +257,26 @@ def identity(data: Any) -> Any:
         ref2 = client.submit(identity, {1: 2})
         assert ref1.hash != ref2.hash
 
+    def test_dict_insertion_order_invariant(self, client: Client) -> None:
+        def identity(data: Any) -> Any:
+            return data
+
+        d1 = {"x": 1, "y": 2, "z": 3}
+        d2 = {"z": 3, "y": 2, "x": 1}
+        ref1 = client.submit(identity, d1)
+        ref2 = client.submit(identity, d2)
+        assert ref1.hash == ref2.hash
+
+    def test_set_insertion_order_invariant(self, client: Client) -> None:
+        def identity(data: Any) -> Any:
+            return data
+
+        s1 = {"a", "b", "c"}
+        s2 = {"c", "b", "a"}
+        ref1 = client.submit(identity, s1)
+        ref2 = client.submit(identity, s2)
+        assert ref1.hash == ref2.hash
+
 
 class TestRecursiveStructures:
     def test_recursive_list_does_not_crash(self, client: Client) -> None: