Summary

Five bugs found while reading the backend services that cause unhandled crashes or silent data corruption.

Bug 1 — services/code_generation.py: AttributeError when model not found

File: tensormap-backend/app/services/code_generation.py

generate_code() accesses model_configs.file_id without checking if model_configs is None first. If the model name doesn't exist in the DB, this raises AttributeError and returns 500 instead of a clean 404.

# Current — crashes if model not found
model_configs = db.exec(...).first()
file = db.exec(select(DataFile).where(DataFile.id == model_configs.file_id)).first()  # AttributeError

# Fix
if model_configs is None:
    raise ValueError(f"Model '{model_name}' not found in database")

Bug 2 — services/code_generation.py: KeyError for unknown problem type

File: tensormap-backend/app/services/code_generation.py

_map_template() uses options[problem_type_id] (dict key access) instead of options.get(problem_type_id). Any unrecognized model_type raises KeyError → 500 with no useful message.

# Current — KeyError on unknown type
return options[problem_type_id]

# Fix
return options.get(problem_type_id)  # returns None, caller raises ValueError with clear message

Bug 3 — services/model_run.py: AttributeError when model not found

File: tensormap-backend/app/services/model_run.py

Same pattern as Bug 1. _run() does not check if model_configs is None before accessing .model_type, crashing with AttributeError.

# Current — crashes if model not found
model_configs = db.exec(...).first()
if model_configs.model_type == ProblemType.IMAGE_CLASSIFICATION:  # AttributeError

Bug 4 — services/model_run.py: KeyError / crash when target_field is None or missing from CSV

File: tensormap-backend/app/services/model_run.py

features.drop(model_configs.target_field, axis=1) crashes with KeyError if:

• target_field is None (user hasn't configured training yet)
• target_field doesn't exist as a column in the CSV

The error message is unhelpful and the root cause is unclear to the user.

Bug 5 — services/data_upload.py: Uploading a duplicate filename silently overwrites the file on disk

File: tensormap-backend/app/services/data_upload.py

If a user uploads a file with the same name as an existing file, the new file overwrites the old one on disk, but a new DB record is created. The old DB record now points to a file with different content — causing silent data corruption.

Fix: Check for existing (file_name, file_type) in DB before saving. Return 409 Conflict if duplicate found.

Steps to Reproduce

• Bug 1 & 2: Call POST /model/code with a model_name that doesn't exist in the DB
• Bug 3: Call POST /model/run with a model_name that doesn't exist in the DB
• Bug 4: Call POST /model/run on a model where target_field is None or was deleted from the CSV
• Bug 5: Upload the same CSV file twice

Expected Behavior

Clean error responses (400/404/409) with descriptive messages instead of unhandled 500 crashes or silent data corruption.