Compress tool outputs, logs, files, and RAG chunks before they reach the LLM. 20% fewer tokens for coding agents, 60-95% fewer tokens for JSON, same answers. Library, proxy, MCP server.
Python agentaianthropicclaude-codecompressioncontext-engineeringcontext-windowcursorfastapilangchainllmmcpopenaiprompt-engineeringproxypythonragtoken-optimizationtokenstypescript
from __future__ import annotations
import re
import pytest
from headroom.transforms.log_compressor import (
LogCompressor,
LogCompressorConfig,
LogFormat,
)
def _malformed_short_summary_diagnostic_log() -> list[str]:
lines = [f"INFO setup output {i}" for i in range(50)]
lines.append("=== diagnostic: short test summary info unavailable")
lines.extend(f"FAILED outside.py::test_{i}" for i in range(10))
lines.extend(f"ERROR outside.py::test_{i}" for i in range(10, 20))
return lines
def test_log_compressor_compress_and_ccr_paths() -> None:
"""Phase 3e.5: `compress()` is now a single Rust call, so this test
exercises end-to-end behavior instead of monkeypatching internal
helpers (which the old orchestration relied on)."""
compressor = LogCompressor(LogCompressorConfig(enable_ccr=True, min_lines_for_ccr=3))
short = compressor.compress("a\nb")
# Below min_lines_for_ccr (3 lines from "a\nb" = 2 lines) → verbatim
assert short.format_detected is LogFormat.GENERIC
assert short.compression_ratio == 1.0
# Real npm log to exercise format detection + CCR end-to-end. Build
# a long enough corpus so compute_optimal_k drops below the
# min_compression_ratio_for_ccr=0.5 threshold.
npm_lines = ["npm WARN deprecated x"] * 30 + ["npm ERR! something broke"] * 5
npm_content = "\n".join(npm_lines)
result = compressor.compress(npm_content)
assert result.format_detected is LogFormat.NPM
assert result.original_line_count == 35
assert result.compressed_line_count < result.original_line_count
# Short input below min_lines_for_ccr returns verbatim with ratio 1.0
# (no compression attempted).
too_short = compressor.compress("x\ny")
assert too_short.compression_ratio == 1.0
assert too_short.cache_key is None
def _pytest_reproduction(failure_count: int, line_ending: str = "\n") -> str:
lines = [f"pytest setup output {i}" for i in range(20)]
for i in range(20):
lines.extend(
[
f"FAILED traceback detail {i}",
f"E AssertionError: failure detail {i}",
]
)
lines.append("=========================== short test summary info ===========================")
lines.extend(
f"FAILED tests/test_generated.py::test_case{i} - AssertionError: failure {i}"
for i in range(failure_count)
)
lines.append(
f"========================= {failure_count} failed in 1.00s ========================="
)
return line_ending.join(lines)
def _retained_or_named(output: str, node_id: str) -> bool:
for line in output.splitlines():
if line.startswith(("FAILED ", "ERROR ")):
entry_id = line.split(" ", 1)[1].split(" - ", 1)[0]
if entry_id == node_id:
return True
if line.startswith("[") and "; omitted: " in line:
named = line.split("; omitted: ", 1)[1].removesuffix("]").split(", ")
if node_id in named:
return True
return False
@pytest.mark.parametrize(("failure_count", "first_regression"), [(16, 13), (20, 13)])
def test_real_compress_preserves_or_names_issue_3814_middle_failures(
failure_count: int, first_regression: int
) -> None:
content = _pytest_reproduction(failure_count)
assert len(content.splitlines()) >= 50
compressor = LogCompressor(LogCompressorConfig(enable_ccr=False))
result = compressor.compress(content)
assert result.compressed_line_count < result.original_line_count
assert result.compression_ratio < 1.0
last_regression = 13 if failure_count == 16 else 17
for i in range(first_regression, last_regression + 1):
node_id = f"tests/test_generated.py::test_case{i}"
assert _retained_or_named(result.compressed, node_id), node_id
def _pytest_issue_log(failure_count: int) -> str:
"""Build the issue #3814 reproduction with a distinct E message per failure."""
lines = [
"============================= test session starts =============================",
"collected 400 items",
"",
]
lines += [
f"tests/test_module_{i:02d}.py .......................... [ {i:2d}%]" for i in range(40)
]
lines.append("=================================== FAILURES ==================================")
for i in range(1, failure_count + 1):
lines += [
f"____________________ test_case{i:03d} ____________________",
"",
"> assert result == expected",
f"E AssertionError: mismatch in test_case{i:03d}",
"",
"tests/t.py:42: AssertionError",
]
lines.append("=========================== short test summary info ===========================")
lines += [
f"FAILED tests/t.py::test_case{i:03d} - AssertionError: mismatch"
for i in range(1, failure_count + 1)
]
lines.append(f"=============== {failure_count} failed, 380 passed in 41.02s ==============")
return "\n".join(lines)
@pytest.mark.parametrize("failure_count", [60, 200])
def test_real_compress_binding_cap_keeps_totals_and_first_error_line(failure_count: int) -> None:
out = LogCompressor(LogCompressorConfig(enable_ccr=False)).compress(
_pytest_issue_log(failure_count)
)
compressed = out.compressed
assert f"{failure_count} failed, 380 passed" in compressed
assert "E AssertionError: mismatch in test_case001" in compressed
# Every entry is still kept or named: kept + listed + K == total.
kept = sum(1 for line in compressed.splitlines() if line.startswith("FAILED tests/t.py::"))
named = compressed.splitlines()[-1].split("; omitted: ", 1)[1].removesuffix("]").split(", ")
overflow = int(named[-1][1:].removesuffix(" more")) if named[-1].startswith("+") else 0
listed = len(named) - (1 if overflow else 0)
assert kept + listed + overflow == failure_count
def test_real_compress_keeps_all_crlf_short_summary_entries_with_non_binding_cap() -> None:
content = _pytest_reproduction(20, "\r\n")
assert len(content.splitlines()) >= 50
compressor = LogCompressor(
LogCompressorConfig(
max_errors=2,
error_context_lines=0,
max_total_lines=1_000,
enable_ccr=False,
)
)
result = compressor.compress(content, bias=1_000)
assert result.compressed_line_count < result.original_line_count
retained_ids = {
line.split(" ", 1)[1].split(" - ", 1)[0]
for line in result.compressed.splitlines()
if line.startswith(("FAILED ", "ERROR "))
}
assert {f"tests/test_generated.py::test_case{i}" for i in range(20)}.issubset(retained_ids)
def test_real_compress_ignores_malformed_short_summary_diagnostic() -> None:
content = "\n".join(_malformed_short_summary_diagnostic_log())
compressor = LogCompressor(
LogCompressorConfig(
max_errors=2,
error_context_lines=0,
keep_summary_lines=False,
max_total_lines=1_000,
enable_ccr=False,
)
)
result = compressor.compress(content, bias=1_000)
assert result.compressed_line_count < result.original_line_count
assert result.compressed == (
"FAILED outside.py::test_0\n"
"FAILED outside.py::test_9\n"
"ERROR outside.py::test_10\n"
"ERROR outside.py::test_19\n"
"[67 lines omitted: 8 ERROR, 8 FAIL, 51 INFO]"
)
def test_omitted_marker_counts_only_dropped_lines_on_native_path() -> None:
lines = [f"INFO worker processed item {i}" for i in range(400)]
error = "ERROR worker failed: connection reset"
lines.insert(200, error)
result = LogCompressor(LogCompressorConfig(enable_ccr=False)).compress("\n".join(lines))
marker = next(line for line in result.compressed.splitlines() if "lines omitted:" in line)
counts = {
level: int(count) for count, level in re.findall(r"(\d+) (ERROR|FAIL|WARN|INFO)", marker)
}
assert error in result.compressed.splitlines()
assert counts == {"INFO": result.lines_omitted}
assert marker == f"[{result.lines_omitted} lines omitted: {result.lines_omitted} INFO]"
assert result.stats["errors"] == 1
assert result.stats["info"] == 400
@pytest.mark.parametrize("prefix", ["DEBUG ", "TRACE ", ""])
def test_non_severity_omissions_keep_count_notice_on_native_path(prefix: str) -> None:
error = "ERROR retained failure"
original = "\n".join([f"{prefix}progress step {i}" for i in range(100)] + [error])
result = LogCompressor(LogCompressorConfig(enable_ccr=False)).compress(original)
marker = re.search(r"(?m)^\[(\d+) lines omitted[^\]]*\]$", result.compressed)
assert marker is not None
assert int(marker[1]) == result.lines_omitted
assert not re.search(r"\d+ (ERROR|FAIL|WARN|INFO)\b", marker[0])
assert marker[0] == f"[{result.lines_omitted} lines omitted]"
assert error in result.compressed.splitlines()
assert result.stats["errors"] == 1
assert result.stats["total"] == 101