ci.deps - Cirron Documentation

Report which optional extras (torch, tensorflow, pandas, datasets, SQL drivers, …) are installed, and, when called with required names, raise CirronDependencyError listing every missing dep at once with a combined pip install command. Use it at the top of a long training script so a missing extra raises immediately instead of 40 minutes into training with a cryptic ImportError, or inside library code that wraps the SDK and wants to gate on what’s present. ci.deps() is the in-process equivalent of the cirron doctor CLI, which inspects an installed cirron-sdk from outside the Python process. Both read from the same pyproject.toml extras list.

Signature

def deps(*required: str) -> dict[str, str | None]

Parameters

Name	Type	Purpose
`*required`	`str`	Optional. Names of extras that must be present. Raises if any are missing.

Each required name can be either the import name ("torch", "datasets", "sklearn") or the pyproject extras name ("hf" for datasets). Unknown names raise ValueError: that’s a caller bug, not a missing dep.

Returns

A dict keyed by import name. With no arguments, one entry for every known extra, mapping to the installed version string or None. With required names, only the requested entries (all present; missing ones would have raised).

Behavior

Zero import cost: uses importlib.util.find_spec + importlib.metadata.version, not __import__. Heavy frameworks (torch, tensorflow, transformers) are never actually loaded by the check itself. Cheap to call at module import time.
Fail-fast form lists every miss at once: one CirronDependencyError with every missing dep in the body and a combined pip install 'cirron-sdk[a,b,c]' command, rather than one error per missing dep.
Caller bug vs missing dep: unknown extras names raise ValueError, not CirronDependencyError. Catching CirronDependencyError never hides a typo in your code.

Examples

Report every extra

import cirron as ci

deps = ci.deps()
# {
#   'pandas': '2.3.3', 'polars': None, 'torch': '2.6.0',
#   'tensorflow': None, 'transformers': '4.41.0', 'datasets': None,
#   'sklearn': '1.5.0', 'safetensors': '0.4.3',
#   'boto3': None, 'psycopg': None, ...
# }

Keyed by import name so deps["polars"] reads naturally. Extras whose pyproject name differs from the import name (hf → datasets, sklearn → scikit-learn) show up under the import name.

Fail fast at script startup

import cirron as ci

ci.deps("torch", "pandas", "transformers")

# If transformers (say) isn't installed:
#
# CirronDependencyError: Missing required dependencies:
#   - transformers: pip install 'cirron-sdk[transformers]'
#
# If multiple are missing, the combined install command is also shown:
#
# CirronDependencyError: Missing required dependencies:
#   - torch: pip install 'cirron-sdk[torch]'
#   - pandas: pip install 'cirron-sdk[pandas]'
#   - transformers: pip install 'cirron-sdk[transformers]'
# Or install all together: pip install 'cirron-sdk[pandas,torch,transformers]'

Guard optional code paths

deps = ci.deps()

if deps["polars"]:
    df = ci.load("./data.parquet", as_="polars")
else:
    df = ci.load("./data.parquet")   # pandas fallback

Accept both import and extras names

# Equivalent: "hf" is the pyproject extras name, "datasets" is the import name.
ci.deps("hf")
ci.deps("datasets")

From a `Cirron` instance

Module-level ci.deps is sugar over the default instance; the method exists on Cirron for API symmetry with the other surface.

from cirron import Cirron

c = Cirron()
report = c.deps()                    # same output as ci.deps()
c.deps("torch", "pandas")            # same fail-fast semantics

Extras presence is process-global, so which instance you call on doesn’t change what the probe sees.

When a check returns `"unknown"`

If a module is importable but not pip-tracked (e.g. an editable install with no metadata, or a vendored copy), ci.deps() reports "unknown" instead of a version string. The extra is considered present, and ci.deps(name) does not raise.

What it does not do

Does not install anything. The error message names the pip command; the user runs it.
Does not probe the Cirron platform. Fully local.
Does not validate config. For that, see ci.profile().health().

cirron doctor

External CLI equivalent. Inspects an installed cirron-sdk from outside the Python process.

CirronDependencyError

Exception raised by the fail-fast form.

Installation

The full list of extras and what they add.

Profile lifecycle

profiler.health() for runtime state.

​Signature

​Parameters

​Returns

​Behavior

​Examples

​Report every extra

​Fail fast at script startup

​Guard optional code paths

​Accept both import and extras names

​From a Cirron instance

​When a check returns "unknown"

​What it does not do

​Related