Code Organization

Encapsulation

Use wrapper methods instead of exposing internal dependencies. This provides better encapsulation and makes refactoring easier.

Bad example:

python

class DownloadService:
    def __init__(self):
        self.repository = HuggingFaceRepository()

    def download_model(self, id: str):
        # Directly accessing internal API of repository
        repo_info = self.repository.api.repo_info(id)
        ...

Good example:

python

class HuggingFaceRepository:
    def get_repo_info(self, id: str) -> RepoInfo:
        """Get repository information from HuggingFace Hub."""
        return self.api.repo_info(id)

class DownloadService:
    def __init__(self):
        self.repository = HuggingFaceRepository()

    def download_model(self, id: str):
        # Using wrapper method - better encapsulation
        repo_info = self.repository.get_repo_info(id)
        ...

Benefits:

•Hides implementation details
•Makes it easier to add logging, caching, or error handling
•Simplifies testing (mock the wrapper instead of internal dependencies)
•Allows changing the underlying implementation without affecting callers

File Modularity

Never put everything in one file. Split large files into focused modules with single responsibilities.

When to split:

•File exceeds 150 lines
•Class has more than 5 distinct responsibilities
•Logic can be grouped into clear, reusable modules

How to split:

•Group related functions into separate files
•Create modules by responsibility (e.g., repository.py, filters.py, file_downloader.py)
•Keep main service file as a thin orchestration layer
•Use clear, descriptive filenames that indicate purpose

Example structure:

code

features/downloads/
  ├── services.py          # Main orchestration only
  ├── repository.py        # Repository operations
  ├── file_downloader.py   # Low-level file operations
  └── filters.py           # Filtering logic

Function Design

One function, one responsibility. Each function should do exactly one thing and do it well.

Bad example:

python

def process_user(user_data):
    # Validates, saves, and sends email - too many responsibilities
    if not user_data.get('email'):
        raise ValueError('Email required')
    db.save(user_data)
    send_welcome_email(user_data['email'])
    return user_data

Good example:

python

def validate_user_data(user_data):
    if not user_data.get('email'):
        raise ValueError('Email required')

def save_user(user_data):
    return db.save(user_data)

def send_welcome_email(email):
    # Only handles email sending
    ...

Code Clarity

Use descriptive variable names in loops. Never use single letters that don't convey meaning.

Bad examples:

python

for p in components_scopes:  # What is 'p'?
for i in users:              # 'i' suggests index, but it's a user
for x in files:              # Meaningless

Good examples:

python

for component_scope in components_scopes:
for scope in components_scopes:  # If 'scope' is clear in context
for user in users:
for file_path in files:

Minimize comments—write self-documenting code instead.

•Only add comments for non-obvious business logic or workarounds
•Never comment on what the code does (code should be clear enough)
•Only comment on why it does it (when it's not obvious)
•Add comments for hacky solutions that can't be avoided

Bad examples:

python

# Increment counter
counter += 1

# Loop through users
for user in users:
    # Process user
    process_user(user)

Good examples:

python

# Workaround for HuggingFace API returning inconsistent revision formats
# See: https://github.com/huggingface/huggingface_hub/issues/1234
revision = getattr(repo_info, 'sha', None) or 'main'

# Skip lock acquisition here because this method is always called
# within a context that already holds the lock
self._update_state_unsafe(new_state)

Using Unique Identifiers

Use database IDs or unique identifiers for adapter/instance names. When working with multiple instances of objects (adapters, plugins, etc.), use database IDs or UUIDs for guaranteed uniqueness.

Bad example:

python

for idx, config in enumerate(lora_configs):
    adapter_name = f"lora_{idx}"  # Index can change if list order changes
    self.pipe.load_lora_weights(config.file_path, adapter_name=adapter_name)

Good example:

python

for config in lora_configs:
    adapter_name = f"lora_{config.id}"  # Database ID is stable and unique
    self.pipe.load_lora_weights(config.file_path, adapter_name=adapter_name)

Benefits:

•Guaranteed uniqueness
•Stable across reorderings
•Easier to debug (can trace back to database)
•More predictable behavior