mirror of
https://github.com/BerriAI/litellm.git
synced 2026-09-17 23:51:30 +00:00
Slice 1 of the Claude Code Compatibility Matrix: the thinnest end-to-end path through every layer for a single (feature, provider) cell, so a future docs page can render a real green cell sourced from a real test. What landed in this repo: - tests/claude_code/manifest.yaml — feature manifest with one entry (basic_messaging_non_streaming) plus the v0 provider column order. - tests/claude_code/cli_driver.py — Claude Code CLI Driver. One entry point (run_claude); handles subprocess assembly, env overlay, stream-JSON parsing, and structured failure modes. `runner=` is a unit-test seam. - tests/claude_code/conftest.py — `compat_result` fixture (tagged-union recorder) + pytest_runtest_makereport hook that infers (feature, provider) from the file path and writes a structured compat-results.json artifact. - tests/claude_code/basic_messaging_non_streaming/test_anthropic.py — the one cell, parametrized over Haiku/Sonnet/Opus per the PRD's per-cell model rule. - tests/claude_code/matrix_builder.py — pure-function builder from (manifest, results, run-metadata) to the v1 JSON schema. Aggregates per- model results into one cell (pass iff all pass). build_from_paths is the thin I/O wrapper for the publisher. - tests/claude_code/sample_compatibility-matrix.json — hand-authored sample of the v1 JSON; copied to the docs repo by hand as part of this slice. - Unit tests: 10 driver tests (mocked subprocess), 9 compat_result tests, 10 matrix-builder golden-file tests. 29/29 pass. Key decisions: - (feature, provider) is inferred from file path, not declared in metadata — mirrors the PRD's "no drift" goal. - Driver injects subprocess via a `runner` kwarg so unit tests don't need the real `claude` CLI; production callers leave it default. - Builder is a pure function on Mappings/Sequences; load/write live in a thin `build_from_paths` wrapper. Golden-file tests pin the schema. - `_driver_unit_tests/` and `_builder_unit_tests/` are prefixed with `_` so the conftest's path-inference hook skips them and they don't pollute the matrix artifact. - `compat-results.json` added to .gitignore (CI-only output). Out of scope per CLAUDE.md (docs live in BerriAI/litellm-docs): - The MDX page `docs/tutorials/claude-code-compatibility` and the `<CompatibilityMatrix />` React component. The hand-authored compatibility-matrix.json (`sample_compatibility-matrix.json` in this repo) is the artifact those docs files will consume; opening that doc PR is the next step in this slice. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
194 lines
6.4 KiB
Python
194 lines
6.4 KiB
Python
"""Claude Code CLI Driver.
|
|
|
|
A thin wrapper around the `claude` CLI in headless mode. Every compatibility
|
|
test consumes only this module — tests must never shell out directly. This
|
|
keeps the subprocess assembly, stream-JSON parsing, and result shape in a
|
|
single place that can be unit-tested with a mocked subprocess.
|
|
|
|
The driver is deliberately small: it knows how to invoke the CLI, drain its
|
|
stream-JSON output, and return a structured `DriverResult`. Higher-level
|
|
matrix concerns (status aggregation, manifest lookup, JSON serialization)
|
|
live in `matrix_builder.py`.
|
|
"""
|
|
|
|
from __future__ import annotations
|
|
|
|
import json
|
|
import os
|
|
import subprocess
|
|
from dataclasses import dataclass, field
|
|
from typing import Any, Dict, List, Mapping, Optional, Sequence
|
|
|
|
CLAUDE_CLI_DEFAULT = "claude"
|
|
DEFAULT_TIMEOUT_SECONDS = 120
|
|
|
|
|
|
class ClaudeCLIError(RuntimeError):
|
|
"""Raised when the `claude` CLI cannot be invoked or returns a fatal error."""
|
|
|
|
|
|
@dataclass
|
|
class DriverResult:
|
|
"""Structured outcome of a single `claude` CLI invocation.
|
|
|
|
`text` is the assistant's final user-visible reply (joined across any
|
|
intermediate `assistant` events for non-streaming runs). `events` is the
|
|
raw list of stream-JSON objects emitted by the CLI, preserved so test
|
|
authors can write feature-specific assertions (tool calls, cache hits,
|
|
usage) without re-parsing stdout.
|
|
"""
|
|
|
|
text: str
|
|
events: List[Dict[str, Any]] = field(default_factory=list)
|
|
exit_code: int = 0
|
|
stderr: str = ""
|
|
usage: Optional[Dict[str, Any]] = None
|
|
duration_ms: Optional[int] = None
|
|
|
|
|
|
def run_claude(
|
|
*,
|
|
prompt: str,
|
|
model: str,
|
|
base_url: str,
|
|
api_key: str,
|
|
extra_env: Optional[Mapping[str, str]] = None,
|
|
extra_args: Optional[Sequence[str]] = None,
|
|
cli_path: str = CLAUDE_CLI_DEFAULT,
|
|
timeout: float = DEFAULT_TIMEOUT_SECONDS,
|
|
runner: Optional[Any] = None,
|
|
) -> DriverResult:
|
|
"""Invoke `claude` once in headless stream-JSON mode and return the result.
|
|
|
|
The CLI is pointed at a LiteLLM proxy via `ANTHROPIC_BASE_URL` /
|
|
`ANTHROPIC_AUTH_TOKEN`, so the same code path exercises every provider
|
|
column — only the model id and the proxy's routing differ between
|
|
invocations.
|
|
|
|
`runner` is an injection seam used by the unit tests: by default we call
|
|
`subprocess.run`, but the test suite swaps in a fake that yields canned
|
|
stream-JSON. Production callers should never set it.
|
|
"""
|
|
if not prompt:
|
|
raise ValueError("prompt must be a non-empty string")
|
|
if not model:
|
|
raise ValueError("model must be a non-empty string")
|
|
if not base_url:
|
|
raise ValueError("base_url must be a non-empty string")
|
|
if not api_key:
|
|
raise ValueError("api_key must be a non-empty string")
|
|
|
|
cmd: List[str] = [
|
|
cli_path,
|
|
"--print",
|
|
"--output-format",
|
|
"stream-json",
|
|
"--verbose",
|
|
"--model",
|
|
model,
|
|
prompt,
|
|
]
|
|
if extra_args:
|
|
cmd.extend(extra_args)
|
|
|
|
env = {**os.environ, **(extra_env or {})}
|
|
env["ANTHROPIC_BASE_URL"] = base_url
|
|
env["ANTHROPIC_AUTH_TOKEN"] = api_key
|
|
|
|
run_fn = runner or subprocess.run
|
|
try:
|
|
completed = run_fn(
|
|
cmd,
|
|
env=env,
|
|
capture_output=True,
|
|
text=True,
|
|
timeout=timeout,
|
|
check=False,
|
|
)
|
|
except FileNotFoundError as exc:
|
|
raise ClaudeCLIError(
|
|
f"claude CLI not found at {cli_path!r}; install with `npm i -g @anthropic-ai/claude-code`"
|
|
) from exc
|
|
except subprocess.TimeoutExpired as exc:
|
|
raise ClaudeCLIError(f"claude CLI timed out after {timeout}s") from exc
|
|
|
|
events = _parse_stream_json(completed.stdout or "")
|
|
text = _extract_assistant_text(events)
|
|
usage = _extract_usage(events)
|
|
|
|
return DriverResult(
|
|
text=text,
|
|
events=events,
|
|
exit_code=completed.returncode,
|
|
stderr=completed.stderr or "",
|
|
usage=usage,
|
|
)
|
|
|
|
|
|
def _parse_stream_json(stdout: str) -> List[Dict[str, Any]]:
|
|
"""Parse newline-delimited JSON emitted by `claude --output-format stream-json`.
|
|
|
|
Lines that don't parse as JSON are silently skipped — the CLI occasionally
|
|
emits debug output we don't care about, and a single malformed line should
|
|
not abort the whole run. Real failure modes surface via exit code.
|
|
"""
|
|
events: List[Dict[str, Any]] = []
|
|
for line in stdout.splitlines():
|
|
line = line.strip()
|
|
if not line:
|
|
continue
|
|
try:
|
|
obj = json.loads(line)
|
|
except json.JSONDecodeError:
|
|
continue
|
|
if isinstance(obj, dict):
|
|
events.append(obj)
|
|
return events
|
|
|
|
|
|
def _extract_assistant_text(events: Sequence[Mapping[str, Any]]) -> str:
|
|
"""Concatenate the text content of every `assistant` event in order.
|
|
|
|
The non-streaming `--print` path emits a single `assistant` event whose
|
|
`message.content` is a list of content blocks. We walk the blocks and
|
|
join every `text` block — the CLI prints other block types (e.g.
|
|
`tool_use`) which we ignore for the basic-messaging case.
|
|
"""
|
|
chunks: List[str] = []
|
|
for event in events:
|
|
if event.get("type") != "assistant":
|
|
continue
|
|
message = event.get("message") or {}
|
|
content = message.get("content")
|
|
if isinstance(content, str):
|
|
chunks.append(content)
|
|
continue
|
|
if not isinstance(content, list):
|
|
continue
|
|
for block in content:
|
|
if not isinstance(block, dict):
|
|
continue
|
|
if block.get("type") == "text" and isinstance(block.get("text"), str):
|
|
chunks.append(block["text"])
|
|
return "".join(chunks)
|
|
|
|
|
|
def _extract_usage(events: Sequence[Mapping[str, Any]]) -> Optional[Dict[str, Any]]:
|
|
"""Return the most recent `usage` block seen on any event, if any.
|
|
|
|
The CLI surfaces token + cache usage on the final `result` event for
|
|
non-streaming runs, but earlier events also carry partial usage in some
|
|
versions; taking the last non-empty one is the safe default.
|
|
"""
|
|
last: Optional[Dict[str, Any]] = None
|
|
for event in events:
|
|
usage = event.get("usage")
|
|
if isinstance(usage, dict) and usage:
|
|
last = usage
|
|
continue
|
|
message = event.get("message")
|
|
if isinstance(message, dict):
|
|
inner = message.get("usage")
|
|
if isinstance(inner, dict) and inner:
|
|
last = inner
|
|
return last
|