Skip to content

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.

fix(code): preserve syntax in CCR retrieval comments (#4583)

Endidac8423965d97
diff --git a/headroom/transforms/code_compressor.py b/headroom/transforms/code_compressor.py
index 734c2ccc6..71bdb0b05 100644
--- a/headroom/transforms/code_compressor.py
+++ b/headroom/transforms/code_compressor.py
@@ -1510,8 +1510,9 @@ class CodeAwareCompressor(Transform):
# Use the actual config attribute (not the wrong name)
ttl_min = max(1, self.config.ccr_ttl // 60)
+ comment_prefix = _LANG_CONFIGS[detected_lang].comment_prefix
compressed += (
- f"\n# [{original_tokens - compressed_tokens} tokens compressed."
+ f"\n{comment_prefix} [{original_tokens - compressed_tokens} tokens compressed."
f"{summary_str}"
f" Retrieve more: hash={cache_key}."
f" Expires in {ttl_min}m.]"
diff --git a/tests/test_transforms/test_code_compressor.py b/tests/test_transforms/test_code_compressor.py
index c689f4976..4ac106c9c 100644
--- a/tests/test_transforms/test_code_compressor.py
+++ b/tests/test_transforms/test_code_compressor.py
@@ -15,6 +15,11 @@ from unittest.mock import patch
import pytest
import headroom.transforms.code_compressor as cc
+from headroom.cache.compression_store import (
+ CompressionStore,
+ clear_request_compression_store,
+ set_request_compression_store,
+)
from headroom.transforms.code_compressor import (
CodeAwareCompressor,
CodeCompressionResult,
@@ -210,6 +215,46 @@ def generate_go_code(n_functions: int = 3) -> str:
return "\n".join(lines)
[email protected](not TREE_SITTER_INSTALLED, reason="tree-sitter not installed")
+ ("language", "code", "comment_prefix"),
+ [
+ pytest.param("python", generate_python_code(), "#", id="python"),
+ pytest.param("javascript", generate_javascript_code(), "//", id="javascript"),
+ pytest.param("typescript", generate_javascript_code(), "//", id="typescript"),
+ pytest.param("go", generate_go_code(), "//", id="go"),
+ ],
+)
+def test_ccr_marker_preserves_code_syntax(language, code, comment_prefix):
+ """CCR metadata must be a comment in the final output, not just before storage."""
+ store = CompressionStore()
+ set_request_compression_store(store)
+ try:
+ compressor = CodeAwareCompressor(
+ CodeCompressorConfig(
+ min_tokens_for_compression=10,
+ enable_ccr=True,
+ fallback_to_kompress=False,
+ )
+ )
+ result = compressor.compress(code, language=language)
+
+ assert result.cache_key is not None
+ assert result.syntax_valid
+ # Parse the returned source, including the marker appended after the
+ # compressor's internal syntax check (#4478, item 4).
+ assert compressor._verify_syntax(result.compressed, CodeLanguage(language))
+ marker = result.compressed.splitlines()[-1]
+ assert marker.startswith(f"{comment_prefix} [")
+ assert f"Retrieve more: hash={result.cache_key}." in marker
+ assert "Expires in 5m.]" in marker
+ entry = store.retrieve(result.cache_key)
+ assert entry is not None
+ assert entry.original_content == code
+ finally:
+ clear_request_compression_store()
+
+
# =============================================================================
# TestCodeCompressorConfig
# =============================================================================