cli improvements (#12465)

Features
- add multiple repos by their branch/repo
- generate `pip install` commands and `add_route()` code
![Screenshot 2023-10-27 at 4 49 52
PM](https://github.com/langchain-ai/langchain/assets/9557659/3aec4cbb-3f67-4f04-8370-5b54ea983b2a)

Optimizations:
- group installs by repo/branch to avoid duplicate cloning

libs/cli/langchain_cli/namespaces/serve.py

@@ -5,17 +5,17 @@ Manage LangServe application projects.
 import shutil
 import subprocess
 from pathlib import Path
-from typing import List, Optional
+from typing import Dict, List, Optional, Tuple
 
-import tomli
 import typer
 from langserve.packages import get_langserve_export, list_packages
 from typing_extensions import Annotated
 
 from langchain_cli.utils.events import create_events
 from langchain_cli.utils.git import (
+    DependencySource,
     copy_repo,
-    parse_dependency_string,
+    parse_dependencies,
     update_repo,
 )
 from langchain_cli.utils.packages import get_package_root
@@ -79,7 +79,10 @@ def add(
         Optional[Path], typer.Option(help="The project directory")
     ] = None,
     repo: Annotated[
-        List[str], typer.Option(help="Shorthand for installing a GitHub Repo")
+        List[str], typer.Option(help="Install deps from a specific github repo instead")
+    ] = [],
+    branch: Annotated[
+        List[str], typer.Option(help="Install deps from a specific branch")
     ] = [],
     with_poetry: Annotated[
         bool,
@@ -94,80 +97,112 @@ def add(
     langchain serve add git+ssh://git@github.com/efriis/simple-pirate.git
     langchain serve add git+https://github.com/efriis/hub.git#devbranch#subdirectory=mypackage
     """
+
+    parsed_deps = parse_dependencies(dependencies, repo, branch, api_path)
+
     project_root = get_package_root(project_dir)
 
-    if dependencies is None:
-        dependencies = []
+    package_dir = project_root / "packages"
+
+    create_events(
+        [{"event": "serve add", "properties": dict(parsed_dep=d)} for d in parsed_deps]
+    )
+
+    # group by repo/ref
+    grouped: Dict[Tuple[str, Optional[str]], List[DependencySource]] = {}
+    for dep in parsed_deps:
+        key_tup = (dep["git"], dep["ref"])
+        lst = grouped.get(key_tup, [])
+        lst.append(dep)
+        grouped[key_tup] = lst
+
+    installed_destination_paths: List[Path] = []
+    installed_exports: List[Dict] = []
 
-    # cannot have both repo and dependencies
-    if len(repo) != 0:
-        if len(dependencies) != 0:
-            raise typer.BadParameter(
-                "Cannot specify both repo and dependencies. "
-                "Please specify one or the other."
+    for (git, ref), group_deps in grouped.items():
+        if len(group_deps) == 1:
+            typer.echo(f"Adding {git}@{ref}...")
+        else:
+            typer.echo(f"Adding {len(group_deps)} dependencies from {git}@{ref}")
+        source_repo_path = update_repo(git, ref, REPO_DIR)
+
+        for dep in group_deps:
+            source_path = (
+                source_repo_path / dep["subdirectory"]
+                if dep["subdirectory"]
+                else source_repo_path
             )
-        dependencies = [f"git+https://github.com/{r}" for r in repo]
+            pyproject_path = source_path / "pyproject.toml"
+            if not pyproject_path.exists():
+                typer.echo(f"Could not find {pyproject_path}")
+                continue
+            langserve_export = get_langserve_export(pyproject_path)
 
-    if len(api_path) != 0 and len(api_path) != len(dependencies):
-        raise typer.BadParameter(
-            "The number of API paths must match the number of dependencies."
-        )
+            # default path to package_name
+            inner_api_path = dep["api_path"] or langserve_export["package_name"]
 
-    # get installed packages from pyproject.toml
-    root_pyproject_path = project_root / "pyproject.toml"
-    with open(root_pyproject_path, "rb") as pyproject_file:
-        pyproject = tomli.load(pyproject_file)
-    installed_packages = (
-        pyproject.get("tool", {}).get("poetry", {}).get("dependencies", {})
-    )
-    installed_names = set(installed_packages.keys())
+            destination_path = package_dir / inner_api_path
+            if destination_path.exists():
+                typer.echo(
+                    f"Folder {str(inner_api_path)} already exists. " "Skipping...",
+                )
+                continue
+            copy_repo(source_path, destination_path)
+            typer.echo(f" - Downloaded {dep['subdirectory']} to {inner_api_path}")
+            installed_destination_paths.append(destination_path)
+            installed_exports.append(langserve_export)
 
-    package_dir = project_root / "packages"
+    if len(installed_destination_paths) == 0:
+        typer.echo("No packages installed. Exiting.")
+        return
 
-    create_events(
-        [{"event": "serve add", "properties": {"package": d}} for d in dependencies]
-    )
+    cwd = Path.cwd()
+    installed_desination_strs = [
+        str(p.relative_to(cwd)) for p in installed_destination_paths
+    ]
 
-    for i, dependency in enumerate(dependencies):
-        # update repo
-        typer.echo(f"Adding {dependency}...")
-        dep = parse_dependency_string(dependency)
-        source_repo_path = update_repo(dep["git"], dep["ref"], REPO_DIR)
-        source_path = (
-            source_repo_path / dep["subdirectory"]
-            if dep["subdirectory"]
-            else source_repo_path
+    if with_poetry:
+        subprocess.run(
+            ["poetry", "add", "--editable"] + installed_desination_strs,
+            cwd=cwd,
         )
-        pyproject_path = source_path / "pyproject.toml"
-        if not pyproject_path.exists():
-            typer.echo(f"Could not find {pyproject_path}")
-            continue
-        langserve_export = get_langserve_export(pyproject_path)
+    else:
+        cmd = ["pip", "install", "-e"] + installed_desination_strs
+        cmd_str = " \\\n  ".join(installed_desination_strs)
+        install_str = f"To install:\n\npip install -e \\\n  {cmd_str}"
+        typer.echo(install_str)
 
-        # detect name conflict
-        if langserve_export["package_name"] in installed_names:
-            typer.echo(
-                f"Package with name {langserve_export['package_name']} already "
-                "installed. Skipping...",
-            )
-            continue
+        if typer.confirm("Run it?"):
+            subprocess.run(cmd, cwd=cwd)
+    if typer.confirm("\nGenerate route code for these packages?", default=True):
+        chain_names = []
+        for e in installed_exports:
+            original_candidate = f'{e["package_name"].replace("-", "_")}_chain'
+            candidate = original_candidate
+            i = 2
+            while candidate in chain_names:
+                candidate = original_candidate + "_" + str(i)
+                i += 1
+            chain_names.append(candidate)
 
-        inner_api_path = (
-            api_path[i] if len(api_path) != 0 else langserve_export["package_name"]
+        api_paths = [
+            str(Path("/") / path.relative_to(package_dir))
+            for path in installed_destination_paths
+        ]
+
+        imports = [
+            f"from {e['module']} import {e['attr']} as {name}"
+            for e, name in zip(installed_exports, chain_names)
+        ]
+        routes = [
+            f'add_routes(app, {name}, path="{path}")'
+            for name, path in zip(chain_names, api_paths)
+        ]
+
+        lines = (
+            ["", "Great! Add the following to your app:", ""] + imports + [""] + routes
         )
-        destination_path = package_dir / inner_api_path
-        if destination_path.exists():
-            typer.echo(
-                f"Endpoint {langserve_export['package_name']} already exists. "
-                "Skipping...",
-            )
-            continue
-        copy_repo(source_path, destination_path)
-        # poetry install
-        if with_poetry:
-            subprocess.run(
-                ["poetry", "add", "--editable", destination_path], cwd=project_root
-            )
+        typer.echo("\n".join(lines))
 
 
 @serve.command()

libs/cli/langchain_cli/utils/git.py

@@ -2,7 +2,7 @@ import hashlib
 import re
 import shutil
 from pathlib import Path
-from typing import Optional, TypedDict
+from typing import Dict, List, Optional, Sequence, TypedDict
 
 from git import Repo
 
@@ -17,13 +17,25 @@ class DependencySource(TypedDict):
     git: str
     ref: Optional[str]
     subdirectory: Optional[str]
+    api_path: Optional[str]
+    event_metadata: Dict
 
 
 # use poetry dependency string format
-def parse_dependency_string(package_string: str) -> DependencySource:
-    if package_string.startswith("git+"):
+def parse_dependency_string(
+    dep: Optional[str],
+    repo: Optional[str],
+    branch: Optional[str],
+    api_path: Optional[str],
+) -> DependencySource:
+    if dep is not None and dep.startswith("git+"):
+        if repo is not None or branch is not None:
+            raise ValueError(
+                "If a dependency starts with git+, you cannot manually specify "
+                "a repo or branch."
+            )
         # remove git+
-        gitstring = package_string[4:]
+        gitstring = dep[4:]
         subdirectory = None
         ref = None
         # first check for #subdirectory= on the end
@@ -62,16 +74,77 @@ def parse_dependency_string(package_string: str) -> DependencySource:
             git=gitstring,
             ref=ref,
             subdirectory=subdirectory,
+            api_path=api_path,
+            event_metadata={"dependency_string": dep},
         )
 
-    elif package_string.startswith("https://"):
-        raise NotImplementedError("url dependencies are not supported yet")
+    elif dep is not None and dep.startswith("https://"):
+        raise ValueError("Only git dependencies are supported")
     else:
+        # if repo is none, use default, including subdirectory
+        base_subdir = Path(DEFAULT_GIT_SUBDIRECTORY) if repo is None else Path()
+        subdir = str(base_subdir / dep) if dep is not None else None
+        gitstring = (
+            DEFAULT_GIT_REPO
+            if repo is None
+            else f"https://github.com/{repo.strip('/')}.git"
+        )
+        ref = DEFAULT_GIT_REF if branch is None else branch
         # it's a default git repo dependency
-        subdirectory = str(Path(DEFAULT_GIT_SUBDIRECTORY) / package_string)
         return DependencySource(
-            git=DEFAULT_GIT_REPO, ref=DEFAULT_GIT_REF, subdirectory=subdirectory
+            git=gitstring,
+            ref=ref,
+            subdirectory=subdir,
+            api_path=api_path,
+            event_metadata={
+                "dependency_string": dep,
+                "used_repo_flag": repo is not None,
+                "used_branch_flag": branch is not None,
+            },
+        )
+
+
+def _list_arg_to_length(arg: Optional[List[str]], num: int) -> Sequence[Optional[str]]:
+    if not arg:
+        return [None] * num
+    elif len(arg) == 1:
+        return arg * num
+    elif len(arg) == num:
+        return arg
+    else:
+        raise ValueError(f"Argument must be of length 1 or {num}")
+
+
+def parse_dependencies(
+    dependencies: Optional[List[str]],
+    repo: List[str],
+    branch: List[str],
+    api_path: List[str],
+) -> List[DependencySource]:
+    num_deps = max(
+        len(dependencies) if dependencies is not None else 0, len(repo), len(branch)
+    )
+    if (
+        (dependencies and len(dependencies) != num_deps)
+        or (api_path and len(api_path) != num_deps)
+        or (repo and len(repo) not in [1, num_deps])
+        or (branch and len(branch) not in [1, num_deps])
+    ):
+        raise ValueError(
+            "Number of defined repos/branches/api_paths did not match the "
+            "number of dependencies."
+        )
+    inner_deps = _list_arg_to_length(dependencies, num_deps)
+    inner_api_paths = _list_arg_to_length(api_path, num_deps)
+    inner_repos = _list_arg_to_length(repo, num_deps)
+    inner_branches = _list_arg_to_length(branch, num_deps)
+
+    return [
+        parse_dependency_string(iter_dep, iter_repo, iter_branch, iter_api_path)
+        for iter_dep, iter_repo, iter_branch, iter_api_path in zip(
+            inner_deps, inner_repos, inner_branches, inner_api_paths
         )
+    ]
 
 
 def _get_repo_path(gitstring: str, repo_dir: Path) -> Path:

libs/cli/pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "langchain-cli"
-version = "0.0.7"
+version = "0.0.8"
 description = "CLI for interacting with LangChain"
 authors = ["Erick Friis <erick@langchain.dev>"]
 readme = "README.md"

libs/cli/tests/test_utils.py

@@ -1,3 +1,5 @@
+from typing import Dict, Optional
+
 import pytest
 
 from langchain_cli.constants import (
@@ -8,44 +10,74 @@ from langchain_cli.constants import (
 from langchain_cli.utils.git import DependencySource, parse_dependency_string
 
 
+def _assert_dependency_equals(
+    dep: DependencySource,
+    *,
+    git: Optional[str] = None,
+    ref: Optional[str] = None,
+    subdirectory: Optional[str] = None,
+    event_metadata: Optional[Dict] = None,
+) -> None:
+    assert dep["git"] == git
+    assert dep["ref"] == ref
+    assert dep["subdirectory"] == subdirectory
+    if event_metadata is not None:
+        assert dep["event_metadata"] == event_metadata
+
+
 def test_dependency_string() -> None:
-    assert parse_dependency_string(
-        "git+ssh://git@github.com/efriis/myrepo.git"
-    ) == DependencySource(
+    _assert_dependency_equals(
+        parse_dependency_string(
+            "git+ssh://git@github.com/efriis/myrepo.git", None, None, None
+        ),
         git="ssh://git@github.com/efriis/myrepo.git",
         ref=None,
         subdirectory=None,
     )
 
-    assert parse_dependency_string(
-        "git+https://github.com/efriis/myrepo.git#subdirectory=src"
-    ) == DependencySource(
+    _assert_dependency_equals(
+        parse_dependency_string(
+            "git+https://github.com/efriis/myrepo.git#subdirectory=src",
+            None,
+            None,
+            None,
+        ),
         git="https://github.com/efriis/myrepo.git",
         subdirectory="src",
         ref=None,
     )
 
-    assert parse_dependency_string(
-        "git+ssh://git@github.com:efriis/myrepo.git#develop"
-    ) == DependencySource(
-        git="ssh://git@github.com:efriis/myrepo.git", ref="develop", subdirectory=None
+    _assert_dependency_equals(
+        parse_dependency_string(
+            "git+ssh://git@github.com:efriis/myrepo.git#develop", None, None, None
+        ),
+        git="ssh://git@github.com:efriis/myrepo.git",
+        ref="develop",
+        subdirectory=None,
     )
 
     # also support a slash in ssh
-    assert parse_dependency_string(
-        "git+ssh://git@github.com/efriis/myrepo.git#develop"
-    ) == DependencySource(
-        git="ssh://git@github.com/efriis/myrepo.git", ref="develop", subdirectory=None
+    _assert_dependency_equals(
+        parse_dependency_string(
+            "git+ssh://git@github.com/efriis/myrepo.git#develop", None, None, None
+        ),
+        git="ssh://git@github.com/efriis/myrepo.git",
+        ref="develop",
+        subdirectory=None,
     )
 
     # looks like poetry supports both an @ and a #
-    assert parse_dependency_string(
-        "git+ssh://git@github.com:efriis/myrepo.git@develop"
-    ) == DependencySource(
-        git="ssh://git@github.com:efriis/myrepo.git", ref="develop", subdirectory=None
+    _assert_dependency_equals(
+        parse_dependency_string(
+            "git+ssh://git@github.com:efriis/myrepo.git@develop", None, None, None
+        ),
+        git="ssh://git@github.com:efriis/myrepo.git",
+        ref="develop",
+        subdirectory=None,
     )
 
-    assert parse_dependency_string("simple-pirate") == DependencySource(
+    _assert_dependency_equals(
+        parse_dependency_string("simple-pirate", None, None, None),
         git=DEFAULT_GIT_REPO,
         subdirectory=f"{DEFAULT_GIT_SUBDIRECTORY}/simple-pirate",
         ref=DEFAULT_GIT_REF,
@@ -53,9 +85,13 @@ def test_dependency_string() -> None:
 
 
 def test_dependency_string_both() -> None:
-    assert parse_dependency_string(
-        "git+https://github.com/efriis/myrepo.git@branch#subdirectory=src"
-    ) == DependencySource(
+    _assert_dependency_equals(
+        parse_dependency_string(
+            "git+https://github.com/efriis/myrepo.git@branch#subdirectory=src",
+            None,
+            None,
+            None,
+        ),
         git="https://github.com/efriis/myrepo.git",
         subdirectory="src",
         ref="branch",
@@ -66,7 +102,10 @@ def test_dependency_string_invalids() -> None:
     # expect error for wrong order
     with pytest.raises(ValueError):
         parse_dependency_string(
-            "git+https://github.com/efriis/myrepo.git#subdirectory=src@branch"
+            "git+https://github.com/efriis/myrepo.git#subdirectory=src@branch",
+            None,
+            None,
+            None,
         )
     # expect error for @subdirectory
 
@@ -77,16 +116,21 @@ def test_dependency_string_edge_case() -> None:
     # this could be a ssh dep with user=a, and default ref
     # or a ssh dep at a with ref=b.
     # in this case, assume the first case (be greedy with the '@')
-    assert parse_dependency_string("git+ssh://a@b") == DependencySource(
+    _assert_dependency_equals(
+        parse_dependency_string("git+ssh://a@b", None, None, None),
         git="ssh://a@b",
         subdirectory=None,
         ref=None,
     )
 
     # weird one that is actually valid
-    assert parse_dependency_string(
-        "git+https://github.com/efriis/myrepo.git@subdirectory=src"
-    ) == DependencySource(
+    _assert_dependency_equals(
+        parse_dependency_string(
+            "git+https://github.com/efriis/myrepo.git@subdirectory=src",
+            None,
+            None,
+            None,
+        ),
         git="https://github.com/efriis/myrepo.git",
         subdirectory=None,
         ref="subdirectory=src",