Implementation:Langchain ai Langchain Check Diff CI Detection

Overview

Concrete CI diff-analysis tool that maps changed files to affected packages and generates test matrices, provided by the check_diff.py script in the LangChain monorepo.

Description

The check_diff.py script is the core engine of LangChain's incremental CI system. It receives a list of changed files as command-line arguments, analyzes which packages are affected, builds a dependency graph, and outputs JSON test matrix configurations for GitHub Actions.

Key functions:

• all_package_dirs(): Discovers all package directories by globbing for pyproject.toml files under libs/.
• dependents_graph(): Builds a mapping of package names to their dependent packages by parsing pyproject.toml dependencies and extended_testing_deps.txt files.
• add_dependents(): Given a set of changed directories, adds all packages that depend on them (except core dependents, which are handled separately).
• _get_configs_for_single_dir(): Generates test configurations for a single directory with appropriate Python versions.
• _get_pydantic_test_configs(): Generates Pydantic version compatibility test configurations.
• _get_configs_for_multi_dirs(): Orchestrates test matrix generation across all affected directories for a given job type.

File-to-package mapping rules:

• Files under libs/partners/<name>/ map to libs/partners/<name>.
• Files under libs/core, libs/langchain, etc. map to the corresponding LANGCHAIN_DIRS entry and trigger extended testing on all downstream packages.
• Files under .github/workflows, .github/scripts trigger extended tests on all core packages.
• Files under libs/standard-tests trigger tests on standard-tests itself plus select partner packages (openai, anthropic, mistralai, fireworks, groq).

Ignored partners: The IGNORED_PARTNERS list (currently ["huggingface"]) excludes unstable packages from dependent testing while still testing them when directly modified.

Output format: Each job type produces a JSON array of configuration objects:

[
  {"working-directory": "libs/partners/deepseek", "python-version": "3.10"},
  {"working-directory": "libs/partners/deepseek", "python-version": "3.14"}
]

Usage

This script is invoked by the check_diffs GitHub Actions workflow. It is not typically run manually, but can be used for debugging CI behavior.

Code Reference

Source Location: .github/scripts/check_diff.py, Lines 1-333

Entry point signature:

if __name__ == "__main__":
    files = sys.argv[1:]
    # ... maps files to dirs_to_run categories
    # ... builds dependents graph
    # ... generates per-job configs
    for key, value in map_job_to_configs.items():
        json_output = json.dumps(value)
        print(f"{key}={json_output}")

Key constants:

LANGCHAIN_DIRS = [
    "libs/core",
    "libs/text-splitters",
    "libs/langchain",
    "libs/langchain_v1",
    "libs/model-profiles",
]

IGNORE_CORE_DEPENDENTS = False

IGNORED_PARTNERS = [
    "huggingface",
]

Import:

# Not imported as a library; invoked as a script
python .github/scripts/check_diff.py <file1> <file2> ...

I/O Contract

Input	Type	Description
sys.argv[1:]	list[str]	Changed file paths relative to repo root
libs/**/pyproject.toml	TOML files	Package dependency declarations (read at runtime)
libs/core/uv.lock	TOML file	Core lock file for Pydantic version detection

Output	Type	Description
lint=<json>	stdout line	JSON array of lint job configurations
test=<json>	stdout line	JSON array of unit test job configurations
extended-tests=<json>	stdout line	JSON array of extended test configurations
compile-integration-tests=<json>	stdout line	JSON array of integration test compile configurations
dependencies=<json>	stdout line	JSON array of dependency check configurations
test-pydantic=<json>	stdout line	JSON array of Pydantic compatibility test configurations
codspeed=<json>	stdout line	JSON array of performance benchmark configurations

Usage Examples

Manual invocation for debugging:

# Simulate a change to the deepseek partner package
cd /path/to/langchain
python .github/scripts/check_diff.py libs/partners/deepseek/langchain_deepseek/chat_models.py

Expected output:

lint=[{"working-directory": "libs/partners/deepseek", "python-version": "3.10"}, {"working-directory": "libs/partners/deepseek", "python-version": "3.14"}]
test=[{"working-directory": "libs/partners/deepseek", "python-version": "3.10"}, {"working-directory": "libs/partners/deepseek", "python-version": "3.14"}]
extended-tests=[]
compile-integration-tests=[{"working-directory": "libs/partners/deepseek", "python-version": "3.10"}, {"working-directory": "libs/partners/deepseek", "python-version": "3.14"}]
dependencies=[{"working-directory": "libs/partners/deepseek", "python-version": "3.10"}, {"working-directory": "libs/partners/deepseek", "python-version": "3.14"}]
test-pydantic=[...]
codspeed=[{"working-directory": "libs/partners/deepseek", "python-version": "3.13"}]

Simulating a core change (triggers dependents):

python .github/scripts/check_diff.py libs/core/langchain_core/language_models/chat_models.py
# Output includes libs/core plus dependent packages in test/lint matrices

Related Pages

• Principle:Langchain_ai_Langchain_CI_Configuration