fix: harden io path validation and error handling

[lipikaramaswamy]

Summary

• Add input path validation in read_input flow: fail fast when path does not exist or is not a file.
• Normalize read/write filesystem and pandas failures into clear InvalidInputError messages with path context.
• Ensure output parent directories are created before writing, and add regression tests for these edge cases.

Type of Change

• Bug fix
• New feature
• Breaking change
• Documentation update
• Refactoring

Testing

• Tests pass locally
• Added/updated tests for changes
  • Verified new failure modes return InvalidInputError instead of raw OS/pandas errors

Related Issues

Closes #12

[andreatgretel]

src/anonymizer/engine/io/reader.py:49

InvalidInputError is used for both validation problems and IO failures here. If a caller catches it to handle bad config, it'll also catch unrelated IO problems like permission errors or corrupt files.

Could be better to add an AnonymizerIOError(AnonymizerError) for the try/except blocks and keep InvalidInputError for the validation cases. Not blocking, just a thought for a follow-up.

---

src/anonymizer/engine/io/reader.py:38

nit: these exists() + is_file() checks could live as a Pydantic @field_validator on AnonymizerInput.source, so bad paths get caught at config time instead of read time. Keeps _load_dataframe focused on just parsing.

---

src/anonymizer/engine/io/reader.py:44 / writer.py:21

suggestion: adding logger.debug before reads/writes and logger.exception in the except blocks would help with debugging without adding noise.

---

src/anonymizer/engine/io/writer.py:17

nit: {".csv", ".parquet"} shows up in both reader and writer. Could pull it into a shared constant like SUPPORTED_FORMATS so adding a format later is a one-line change.

[lipikaramaswamy]

src/anonymizer/engine/io/reader.py:49

InvalidInputError is used for both validation problems and IO failures here. If a caller catches it to handle bad config, it'll also catch unrelated IO problems like permission errors or corrupt files.

Could be better to add an AnonymizerIOError(AnonymizerError) for the try/except blocks and keep InvalidInputError for the validation cases. Not blocking, just a thought for a follow-up.

src/anonymizer/engine/io/reader.py:38

nit: these exists() + is_file() checks could live as a Pydantic @field_validator on AnonymizerInput.source, so bad paths get caught at config time instead of read time. Keeps _load_dataframe focused on just parsing.

src/anonymizer/engine/io/reader.py:44 / writer.py:21

suggestion: adding logger.debug before reads/writes and logger.exception in the except blocks would help with debugging without adding noise.

src/anonymizer/engine/io/writer.py:17

nit: {".csv", ".parquet"} shows up in both reader and writer. Could pull it into a shared constant like SUPPORTED_FORMATS so adding a format later is a one-line change.

Thanks @andreatgretel for the thoughtful review. I've addressed 3/4 items -

• Added shared IO format constants so reader/writer don’t duplicate suffix definitions.
• Moved source path checks (exists / is_file) into AnonymizerInput via Pydantic @field_validator, so invalid paths fail at config/input creation time.
• Introduced AnonymizerIOError and now use it for read/write runtime failures, while keeping InvalidInputError for validation-style issues.

Let's defer logging changes to a dedicated follow-up PR so we can apply a consistent logging policy across the codebase in one diff. Can capture in #14 or a follow up to it.