File size: 10,650 Bytes
29928b8
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
60741b5
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
"""Tests for CompressionCache with LRU eviction."""

from __future__ import annotations

import pytest

from headroom.cache.compression_cache import CompressionCache


@pytest.fixture
def cache() -> CompressionCache:
    return CompressionCache()


@pytest.fixture
def small_cache() -> CompressionCache:
    return CompressionCache(max_entries=3)


class TestCompressionCache:
    def test_cache_miss_returns_none(self, cache: CompressionCache) -> None:
        h = CompressionCache.content_hash("some content")
        assert cache.get_compressed(h) is None

    def test_store_and_retrieve(self, cache: CompressionCache) -> None:
        content = "hello world this is a long message"
        h = CompressionCache.content_hash(content)
        cache.store_compressed(h, "hello world...compressed", tokens_saved=15)
        assert cache.get_compressed(h) == "hello world...compressed"

    def test_different_content_different_hash(self) -> None:
        h1 = CompressionCache.content_hash("content A")
        h2 = CompressionCache.content_hash("content B")
        assert h1 != h2

    def test_overwrite_same_hash(self, cache: CompressionCache) -> None:
        h = CompressionCache.content_hash("some content")
        cache.store_compressed(h, "v1", tokens_saved=10)
        cache.store_compressed(h, "v2", tokens_saved=20)
        assert cache.get_compressed(h) == "v2"

    def test_stats_tracking(self, cache: CompressionCache) -> None:
        h = CompressionCache.content_hash("content")
        cache.store_compressed(h, "compressed", tokens_saved=5)

        # One hit
        cache.get_compressed(h)
        # One miss
        cache.get_compressed("nonexistent")

        stats = cache.get_stats()
        assert stats["hits"] == 1
        assert stats["misses"] == 1
        assert stats["entries"] == 1
        assert stats["tokens_saved"] == 5

    def test_eviction_at_max_entries(self, small_cache: CompressionCache) -> None:
        h1 = CompressionCache.content_hash("a")
        h2 = CompressionCache.content_hash("b")
        h3 = CompressionCache.content_hash("c")
        h4 = CompressionCache.content_hash("d")

        small_cache.store_compressed(h1, "ca", tokens_saved=1)
        small_cache.store_compressed(h2, "cb", tokens_saved=1)
        small_cache.store_compressed(h3, "cc", tokens_saved=1)

        # Adding a 4th should evict the oldest (h1)
        small_cache.store_compressed(h4, "cd", tokens_saved=1)

        assert small_cache.get_compressed(h1) is None
        assert small_cache.get_compressed(h2) == "cb"
        assert small_cache.get_compressed(h4) == "cd"

    def test_access_refreshes_lru(self, small_cache: CompressionCache) -> None:
        h1 = CompressionCache.content_hash("a")
        h2 = CompressionCache.content_hash("b")
        h3 = CompressionCache.content_hash("c")
        h4 = CompressionCache.content_hash("d")

        small_cache.store_compressed(h1, "ca", tokens_saved=1)
        small_cache.store_compressed(h2, "cb", tokens_saved=1)
        small_cache.store_compressed(h3, "cc", tokens_saved=1)

        # Access h1 to refresh it
        small_cache.get_compressed(h1)

        # Adding h4 should evict h2 (oldest untouched), not h1
        small_cache.store_compressed(h4, "cd", tokens_saved=1)

        assert small_cache.get_compressed(h1) == "ca"
        assert small_cache.get_compressed(h2) is None
        assert small_cache.get_compressed(h4) == "cd"

    def test_content_hash_list_content(self) -> None:
        """content_hash handles Anthropic-format list content."""
        list_content = [
            {"type": "text", "text": "hello"},
            {"type": "text", "text": "world"},
        ]
        h = CompressionCache.content_hash(list_content)
        assert isinstance(h, str)
        assert len(h) == 16

        # Same content produces same hash
        assert CompressionCache.content_hash(list_content) == h

    def test_content_hash_string_length(self) -> None:
        h = CompressionCache.content_hash("test")
        assert len(h) == 16


class TestCompressionCacheFrozenCount:
    def test_empty_cache_returns_zero(self, cache: CompressionCache) -> None:
        assert cache.compute_frozen_count([]) == 0

    def test_user_assistant_always_stable(self, cache: CompressionCache) -> None:
        messages = [
            {"role": "user", "content": "hello"},
            {"role": "assistant", "content": "hi there"},
            {"role": "user", "content": "how are you"},
        ]
        assert cache.compute_frozen_count(messages) == 3

    def test_tool_result_with_cache_hit_is_stable(self, cache: CompressionCache) -> None:
        tool_content = "tool output data"
        h = CompressionCache.content_hash(tool_content)
        cache.store_compressed(h, "compressed tool output", tokens_saved=5)

        messages = [
            {"role": "user", "content": "do something"},
            {
                "role": "assistant",
                "content": [{"type": "tool_use", "id": "t1", "name": "my_tool", "input": {}}],
            },
            {
                "role": "user",
                "content": [{"type": "tool_result", "tool_use_id": "t1", "content": tool_content}],
            },
        ]
        assert cache.compute_frozen_count(messages) == 3

    def test_tool_result_cache_miss_stops_frozen(self, cache: CompressionCache) -> None:
        messages = [
            {"role": "user", "content": "hello"},
            {
                "role": "user",
                "content": [
                    {"type": "tool_result", "tool_use_id": "t1", "content": "uncached stuff"}
                ],
            },
            {"role": "user", "content": "follow up"},
        ]
        assert cache.compute_frozen_count(messages) == 1

    def test_frozen_count_with_dropped_messages(self, cache: CompressionCache) -> None:
        cached_content = "cached tool output"
        h = CompressionCache.content_hash(cached_content)
        cache.store_compressed(h, "compressed", tokens_saved=3)

        messages = [
            {"role": "user", "content": "start"},
            {
                "role": "user",
                "content": [
                    {"type": "tool_result", "tool_use_id": "t1", "content": cached_content}
                ],
            },
            {
                "role": "user",
                "content": [{"type": "tool_result", "tool_use_id": "t2", "content": "not cached"}],
            },
        ]
        assert cache.compute_frozen_count(messages) == 2


class TestCompressionCacheApplyAndUpdate:
    def test_apply_cached_swaps_tool_results(self, cache: CompressionCache) -> None:
        original_content = "big tool output"
        h = CompressionCache.content_hash(original_content)
        cache.store_compressed(h, "small output", tokens_saved=5)

        messages = [
            {"role": "user", "content": "hi"},
            {
                "role": "user",
                "content": [
                    {"type": "tool_result", "tool_use_id": "t1", "content": original_content}
                ],
            },
        ]
        result = cache.apply_cached(messages)
        assert result[1]["content"][0]["content"] == "small output"

    def test_apply_cached_preserves_uncached_messages(self, cache: CompressionCache) -> None:
        messages = [
            {"role": "user", "content": "hello"},
            {"role": "assistant", "content": "world"},
        ]
        result = cache.apply_cached(messages)
        assert result[0] is messages[0]
        assert result[1] is messages[1]

    def test_apply_cached_never_adds_messages(self, cache: CompressionCache) -> None:
        # Store something in cache that doesn't correspond to any message
        cache.store_compressed("orphan_hash", "orphan_value", tokens_saved=1)

        messages = [
            {"role": "user", "content": "hello"},
            {"role": "assistant", "content": "hi"},
        ]
        result = cache.apply_cached(messages)
        assert len(result) == len(messages)

    def test_update_from_result_caches_changes(self, cache: CompressionCache) -> None:
        originals = [
            {
                "role": "user",
                "content": [
                    {"type": "tool_result", "tool_use_id": "t1", "content": "original output"}
                ],
            },
        ]
        compressed = [
            {
                "role": "user",
                "content": [
                    {"type": "tool_result", "tool_use_id": "t1", "content": "compressed output"}
                ],
            },
        ]
        cache.update_from_result(originals, compressed)

        h = CompressionCache.content_hash("original output")
        assert cache.get_compressed(h) == "compressed output"

    def test_update_from_result_ignores_unchanged(self, cache: CompressionCache) -> None:
        originals = [
            {
                "role": "user",
                "content": [
                    {"type": "tool_result", "tool_use_id": "t1", "content": "same content"}
                ],
            },
        ]
        compressed = [
            {
                "role": "user",
                "content": [
                    {"type": "tool_result", "tool_use_id": "t1", "content": "same content"}
                ],
            },
        ]
        cache.update_from_result(originals, compressed)
        h = CompressionCache.content_hash("same content")
        assert cache.get_compressed(h) is None

    def test_apply_does_not_modify_original_messages(self, cache: CompressionCache) -> None:
        original_content = "big tool output"
        h = CompressionCache.content_hash(original_content)
        cache.store_compressed(h, "small output", tokens_saved=5)

        msg = {
            "role": "user",
            "content": [{"type": "tool_result", "tool_use_id": "t1", "content": original_content}],
        }
        messages = [msg]
        cache.apply_cached(messages)

        # Original must be untouched
        assert msg["content"][0]["content"] == original_content

    def test_openai_format_tool_result(self, cache: CompressionCache) -> None:
        original_content = "openai tool output"
        h = CompressionCache.content_hash(original_content)
        cache.store_compressed(h, "compressed openai", tokens_saved=4)

        messages = [
            {"role": "tool", "tool_call_id": "tc1", "content": original_content},
        ]
        result = cache.apply_cached(messages)
        assert result[0]["content"] == "compressed openai"
        # Original untouched
        assert messages[0]["content"] == original_content