mirror of
https://github.com/headroomlabs-ai/headroom.git
synced 2026-11-22 18:41:07 +08:00
## Description Several small helpers exist twice or more across the proxy handler modules, with the same body. This keeps one copy of each and points the handlers at it. The method names the tests and benchmarks call (`AnthropicHandlerMixin._restore_frozen_prefix`, `_extract_anthropic_cache_ttl_metrics`, `_thinking_tokens_for`) stay, as thin delegates. Refs #4136 ## Type of Change - [ ] Bug fix (non-breaking change that fixes an issue) - [ ] New feature (non-breaking change that adds functionality) - [ ] Breaking change (fix or feature that would cause existing functionality to change) - [ ] Documentation update - [ ] Performance improvement - [x] Code refactoring (no functional changes) ## Changes Made - **`_restore_frozen_prefix`** was byte-identical in `anthropic.py` and `openai.py`. - It is now `restore_frozen_prefix` in `headroom/cache/prefix_tracker.py`, next to `extract_cache_stable_delta`, which the handler already delegates to the same way. Both staticmethods delegate. - `_strict_previous_turn_frozen_count` is **not** merged: the OpenAI copy deliberately treats `tool`/`function` tails as mutable. - **`_extract_anthropic_cache_ttl_metrics`** was identical in `anthropic.py` and `streaming.py`. - The body is now `provider_usage.anthropic_cache_ttl_buckets` (same `int(x or 0)` semantics). - Both methods delegate, so tests that override the method on an instance still work. - **`_has_headroom_retrieve_tool`** (`anthropic.py`) and **`_has_headroom_retrieve_tool_responses`** (`openai.py`) had the same loop. - They are now one `has_ccr_retrieve_tool` in `headroom/ccr/tool_calls.py`, beside `has_ccr_tool_calls`. It uses `CCR_TOOL_NAME`, which equals the literal the Anthropic copy hard-coded. - **`_thinking_for_stream`** (`streaming.py`) re-implemented `anthropic._thinking_tokens_for`. The stream finaliser now imports and calls that function lazily, as before. - **The usage-count coercion `max(int(x), 0)`** was copied six times (four times in `openai.py`, once in `streaming.py`, and once in `anthropic.py`). - Those sites now use `provider_usage.usage_int(value, default=0)`. This is the module's existing private `_int`, renamed so the handlers can import it, with a `default` argument for the one site that falls back to the pre-request count. - Gemini's `_usage_int` has different semantics (it raises on garbage), so it is left alone. - **Tests:** - `test_thinking_tokens.py::test_helper_never_raises_on_junk` asserted `x is None or True`, which can never fail. It now checks that the result is a `ThinkingTokens` with no count. - Added a parametrized `usage_int` test in `tests/test_provider_billed_input.py`. ## Testing - [x] Unit tests pass (`pytest`) - [x] Linting passes (`ruff check .`) - [x] Type checking passes (`mypy headroom`) - [x] New tests added for new functionality - [ ] Manual testing performed ### Test Output ```text $ uvx ruff@0.16.8 check headroom/proxy/handlers/ headroom/proxy/provider_usage.py headroom/cache/prefix_tracker.py headroom/ccr/tool_calls.py tests/test_thinking_tokens.py tests/test_provider_billed_input.py All checks passed! $ python -m mypy headroom/proxy/provider_usage.py headroom/cache/prefix_tracker.py headroom/ccr/tool_calls.py headroom/proxy/handlers/anthropic.py headroom/proxy/handlers/openai.py headroom/proxy/handlers/streaming.py --ignore-missing-imports Success: no issues found in 6 source files $ pytest -q (41 files: every test file that references a touched symbol, plus the audit's verify list, incl. tests/test_proxy_handler_helpers.py tests/test_proxy_cache_ttl_metrics.py tests/test_thinking_tokens.py tests/test_provider_billed_input.py tests/test_openai_chat_turn_hooks.py tests/test_route_advice.py tests/test_anthropic_pre_upstream_backpressure.py tests/test_codex_ws_savings_deferral.py tests/test_gemini_ccr_continuation_usage.py tests/test_proxy_anthropic_cache_stability.py tests/test_claude_session_mode_benchmark.py tests/test_cache/test_prefix_tracker.py tests/test_ccr_tool_calls.py tests/test_turn_hook_usage.py tests/test_backend_streaming_cache_metrics.py tests/test_backend_nonstreaming_cache_metrics.py tests/test_streaming_sse_malformed_events.py) 1102 passed in 93.82s $ pytest -q tests/test_thinking_tokens.py tests/test_provider_billed_input.py 48 passed in 4.06s $ pytest -q tests/test_provider_billed_input.py tests/test_proxy_cache_ttl_metrics.py tests/test_turn_hook_usage.py tests/test_proxy_handler_helpers.py # after the __all__ amend 125 passed in 5.33s ``` ## Real Behavior Proof - Environment: macOS, Python 3.12, repo venv, branch `cleanup/proxy-handlers-shared-helpers` @e1c0bbe6, temp `HOME`/`HEADROOM_WORKSPACE_DIR`/`CODEX_HOME`, `HEADROOM_OFFLINE=1` - Exact command / steps: (1) the shared smoke harness: a real `headroom proxy` from this checkout against a local fake upstream, with one Anthropic and one OpenAI chat turn (both non-streaming paths go through `usage_int` / `_AnthropicTurnHookUsage` / the TTL-bucket helper); (2) a direct run of the moved helpers through the handler classes, using upstream-shaped usage blocks (a null count, a string count, a negative count, TTL buckets, a frozen prefix a transform changed, flat and nested retrieve tools) - Observed result: shown below. (1) Both turns return 200 with savings, and there are no new warnings. (2) The outputs match the old per-site behaviour: null becomes 0, `"7"` becomes 7, negative becomes 0, the TTL buckets are (4, 5), the frozen system message is restored (1 restored), and the retrieve tool is detected in both shapes but not for another name. - Not tested: a real hosted upstream ```text $ bash smoke.sh <worktree> branch: cleanup/proxy-handlers-shared-helpers @e1c0bbe6health: 200 POST /v1/messages -> HTTP200 smoke-ok POST /v1/chat/completions -> HTTP200 smoke-ok stats: requests=2 failed=0 tokens_saved=13574 savings_percent=99.82 proxy.log WARNING/ERROR lines: (only the pre-existing offline-mode notice and Kompress offline-prefetch warnings) $ HOME=$T/h HEADROOM_WORKSPACE_DIR=$T/ws CODEX_HOME=$T/c python -W always -c "<drive StreamingMixin/OpenAI/Anthropic helpers>" anthropic SSE usage: {'input_tokens': 12, 'cache_read_input_tokens': 3, 'cache_creation_input_tokens': 9, 'cache_creation_ephemeral_5m_input_tokens': 4, 'cache_creation_ephemeral_1h_input_tokens': 5} openai SSE null usage: {'input_tokens': 0, 'output_tokens': 7, 'cache_read_input_tokens': 0} responses usage: (100, 0, 40, 60, 60) hook usage extra: 1 3 restore prefix: ([{'role': 'system', 'content': 'a'}], 1) retrieve tool: True True False ``` ## Runtime Rollout Safety - Rollout-managed feature(s): none - Minimum rollout channel: stable - Stable/default behavior changed: one edge case only: a boolean usage count (`True`) read as 1 at the six old coercion sites and now reads as 0, as `provider_usage` already did; no provider sends a boolean count - Kill switch / disable path: not applicable (refactor) - Unsafe override required: no - Qualification impact: none - Rollback path: revert this commit ## Review Readiness - [x] I have performed a self-review - [x] This PR is ready for human review ## Checklist - [x] My code follows the project's style guidelines - [x] I have performed a self-review of my code - [x] I have commented my code, particularly in hard-to-understand areas - [ ] I have made corresponding changes to the documentation - [x] My changes generate no new warnings - [x] I have added tests that prove my fix is effective or that my feature works - [x] New and existing unit tests pass locally with my changes - [x] I did **not** edit `CHANGELOG.md` — it is generated by release-please from my Conventional Commit PR title (a CI guard enforces this) ## Screenshots (if applicable) N/A ## Additional Notes Part of maintainer-requested cleanup #4136 (dead code, duplication, debuggability), split into small PRs for review. - All removed names are private (leading underscore). None is in an `__all__`, a lazy export or the consumer import index, and none appears in docs. - The four new shared functions are additive, and none is added to any `__all__`. - Exact behaviour difference: the six old copies computed `max(int(value), 0)`, so `True` gave 1. `usage_int` returns the default (0) for booleans, as `provider_usage._int` already did. Every other input gives the same result as before. - Documentation: N/A, because no user-visible behaviour changes. 🤖 Generated with [Claude Code](https://claude.com/claude-code)
228 lines
9.1 KiB
Python
228 lines
9.1 KiB
Python
"""Tests for the thinking/visible output-token split.
|
|
|
|
The property under test throughout is that **unknown is not zero**. Every
|
|
provider reports this differently and one reports it not at all, so the failure
|
|
mode to guard against is a fabricated zero quietly asserting "no thinking
|
|
happened" and corrupting every average computed over mixed traffic.
|
|
"""
|
|
|
|
from __future__ import annotations
|
|
|
|
from headroom.proxy.outcome import RequestOutcome
|
|
from headroom.proxy.thinking_tokens import (
|
|
ThinkingTokens,
|
|
anthropic_thinking_text,
|
|
extract_from_usage,
|
|
extract_thinking_tokens,
|
|
)
|
|
|
|
|
|
class TestOpenAI:
|
|
def test_responses_details(self):
|
|
payload = {
|
|
"usage": {"output_tokens": 900, "output_tokens_details": {"reasoning_tokens": 700}}
|
|
}
|
|
assert extract_thinking_tokens(payload) == ThinkingTokens(tokens=700, inferred=False)
|
|
|
|
def test_chat_details(self):
|
|
payload = {
|
|
"usage": {
|
|
"completion_tokens": 500,
|
|
"completion_tokens_details": {"reasoning_tokens": 320},
|
|
}
|
|
}
|
|
assert extract_thinking_tokens(payload).tokens == 320
|
|
|
|
def test_zero_reasoning_is_a_real_zero_not_unknown(self):
|
|
"""A reported 0 is a claim: the provider says no thinking happened.
|
|
It must not be collapsed into None."""
|
|
payload = {
|
|
"usage": {"completion_tokens": 40, "completion_tokens_details": {"reasoning_tokens": 0}}
|
|
}
|
|
result = extract_thinking_tokens(payload)
|
|
assert result.tokens == 0
|
|
assert result.known is True
|
|
|
|
def test_usage_without_details_is_unknown(self):
|
|
payload = {"usage": {"completion_tokens": 40}}
|
|
assert extract_thinking_tokens(payload).tokens is None
|
|
|
|
def test_bare_usage_dict_entry_point(self):
|
|
usage = {"output_tokens_details": {"reasoning_tokens": 128}}
|
|
assert extract_from_usage(usage).tokens == 128
|
|
assert extract_from_usage({}).tokens is None
|
|
assert extract_from_usage(None).tokens is None
|
|
|
|
def test_booleans_are_not_counts(self):
|
|
"""``isinstance(True, int)`` is True in Python; a bool here means a
|
|
malformed payload, not a count of one."""
|
|
usage = {"completion_tokens_details": {"reasoning_tokens": True}}
|
|
assert extract_from_usage(usage).tokens is None
|
|
|
|
|
|
class TestGemini:
|
|
def test_thoughts_token_count(self):
|
|
payload = {"usageMetadata": {"candidatesTokenCount": 200, "thoughtsTokenCount": 1500}}
|
|
assert extract_thinking_tokens(payload).tokens == 1500
|
|
|
|
def test_absent_thoughts_is_unknown(self):
|
|
payload = {"usageMetadata": {"candidatesTokenCount": 200}}
|
|
assert extract_thinking_tokens(payload).tokens is None
|
|
|
|
|
|
class TestAnthropic:
|
|
"""Anthropic reports no thinking count at all — the interesting case."""
|
|
|
|
def test_no_estimator_means_unknown_never_zero(self):
|
|
payload = {
|
|
"content": [
|
|
{"type": "thinking", "thinking": "a b c d e f"},
|
|
{"type": "text", "text": "done"},
|
|
]
|
|
}
|
|
result = extract_thinking_tokens(payload)
|
|
assert result.tokens is None, "without a tokenizer we do not know, and must say so"
|
|
assert result.inferred is False
|
|
|
|
def test_estimator_produces_an_inferred_count(self):
|
|
payload = {"content": [{"type": "thinking", "thinking": "one two three four"}]}
|
|
result = extract_thinking_tokens(payload, estimator=lambda t: len(t.split()))
|
|
assert result.tokens == 4
|
|
assert result.inferred is True, "a derived number must never look reported"
|
|
|
|
def test_content_present_but_no_thinking_is_a_real_zero(self):
|
|
"""Distinguishable from 'not an Anthropic response': a response that
|
|
carries content blocks genuinely did no thinking."""
|
|
payload = {"content": [{"type": "text", "text": "hello"}]}
|
|
assert extract_thinking_tokens(payload).tokens == 0
|
|
|
|
def test_unrecognised_payload_is_unknown(self):
|
|
assert extract_thinking_tokens({"foo": "bar"}).tokens is None
|
|
assert extract_thinking_tokens(None).tokens is None
|
|
assert extract_thinking_tokens("not a dict").tokens is None
|
|
|
|
def test_redacted_thinking_still_counts(self):
|
|
"""Redacted blocks carry no readable text but were still billed."""
|
|
payload = {"content": [{"type": "redacted_thinking", "data": "xxxx yyyy"}]}
|
|
result = extract_thinking_tokens(payload, estimator=lambda t: len(t.split()))
|
|
assert result.tokens == 2
|
|
|
|
def test_thinking_text_skips_malformed_blocks(self):
|
|
payload = {
|
|
"content": [
|
|
"not a dict",
|
|
{"type": "thinking"},
|
|
{"type": "thinking", "thinking": None},
|
|
{"type": "thinking", "thinking": "real"},
|
|
]
|
|
}
|
|
assert anthropic_thinking_text(payload) == "real"
|
|
|
|
def test_a_raising_estimator_degrades_to_unknown(self):
|
|
"""Accounting must never cost a caller their response."""
|
|
|
|
def boom(_: str) -> int:
|
|
raise RuntimeError("tokenizer exploded")
|
|
|
|
payload = {"content": [{"type": "thinking", "thinking": "x"}]}
|
|
assert extract_thinking_tokens(payload, estimator=boom).tokens is None
|
|
|
|
|
|
class TestVisibleSplit:
|
|
def test_visible_from(self):
|
|
assert ThinkingTokens(tokens=700).visible_from(900) == 200
|
|
|
|
def test_visible_from_is_none_when_unknown(self):
|
|
assert ThinkingTokens().visible_from(900) is None
|
|
|
|
def test_inferred_overshoot_clamps_at_zero(self):
|
|
"""An inferred count is on Headroom's tokenizer scale while
|
|
output_tokens is on the provider's; they can disagree on a short
|
|
response and the difference must not go negative."""
|
|
assert ThinkingTokens(tokens=120, inferred=True).visible_from(100) == 0
|
|
|
|
|
|
class TestRequestOutcome:
|
|
def _outcome(self, **kw) -> RequestOutcome:
|
|
base = {
|
|
"request_id": "r1",
|
|
"provider": "anthropic",
|
|
"model": "claude-sonnet-4",
|
|
"original_tokens": 1000,
|
|
"optimized_tokens": 800,
|
|
"output_tokens": 900,
|
|
"tokens_saved": 200,
|
|
"attempted_input_tokens": 1000,
|
|
}
|
|
base.update(kw)
|
|
return RequestOutcome(**base)
|
|
|
|
def test_defaults_preserve_every_existing_emit_site(self):
|
|
"""The fields are optional so none of the existing construction sites
|
|
need to change; an outcome that says nothing must report unknown."""
|
|
outcome = self._outcome()
|
|
assert outcome.thinking_tokens is None
|
|
assert outcome.thinking_inferred is False
|
|
assert outcome.turn_index == 0
|
|
assert outcome.visible_output_tokens is None
|
|
|
|
def test_visible_output_tokens_splits_the_total(self):
|
|
outcome = self._outcome(output_tokens=900, thinking_tokens=700)
|
|
assert outcome.visible_output_tokens == 200
|
|
|
|
def test_reported_zero_yields_full_visible(self):
|
|
outcome = self._outcome(output_tokens=900, thinking_tokens=0)
|
|
assert outcome.visible_output_tokens == 900
|
|
|
|
def test_turn_index_round_trips(self):
|
|
assert self._outcome(turn_index=7).turn_index == 7
|
|
|
|
|
|
class TestHandlerWiring:
|
|
"""The handler helper must actually produce a count.
|
|
|
|
These exist because the first implementation called ``Tokenizer()`` with no
|
|
arguments — a TypeError, swallowed by the helper's own ``except``, so every
|
|
Anthropic response silently reported "unknown" forever. Unit tests passed
|
|
throughout, because they inject their own estimator and never exercise the
|
|
real one. Only an end-to-end assertion on the helper catches that.
|
|
"""
|
|
|
|
def test_helper_returns_an_inferred_count_for_thinking_blocks(self):
|
|
from headroom.proxy.handlers.anthropic import _thinking_tokens_for
|
|
|
|
payload = {
|
|
"content": [
|
|
{"type": "thinking", "thinking": "Let me read the parser before editing it."},
|
|
{"type": "text", "text": "done"},
|
|
]
|
|
}
|
|
result = _thinking_tokens_for(payload)
|
|
assert result.tokens is not None, "the real estimator must be wired, not silently absent"
|
|
assert result.tokens > 0
|
|
assert result.inferred is True
|
|
|
|
def test_helper_reports_a_real_zero_when_nothing_was_thought(self):
|
|
from headroom.proxy.handlers.anthropic import _thinking_tokens_for
|
|
|
|
result = _thinking_tokens_for({"content": [{"type": "text", "text": "hi"}]})
|
|
assert result.tokens == 0
|
|
assert result.inferred is False
|
|
|
|
def test_estimator_is_built_once(self):
|
|
"""AnthropicTokenCounter loads a tiktoken encoding in __init__, so
|
|
constructing one per request would put a vocab load in the response
|
|
path."""
|
|
from headroom.proxy.handlers.anthropic import _thinking_estimator
|
|
|
|
assert _thinking_estimator() is _thinking_estimator()
|
|
|
|
def test_helper_never_raises_on_junk(self):
|
|
from headroom.proxy.handlers.anthropic import _thinking_tokens_for
|
|
from headroom.proxy.thinking_tokens import ThinkingTokens
|
|
|
|
for junk in (None, "string", 42, [], {"content": "not a list"}):
|
|
result = _thinking_tokens_for(junk)
|
|
assert isinstance(result, ThinkingTokens)
|
|
assert not result.tokens # nothing to count: unknown (None) or zero
|