Invalid URL schemes are not properly handled and trigger fallback file resolution crashes

[yasqdb]

Bug

When providing an unsupported URL scheme (ex: ftp://), the CLI validates the input correctly but does not stop execution afterward.
Instead of exiting cleanly, the process continues and attempts to treat the invalid URL as a local file path, which leads to a filesystem error.

Steps to reproduce

Run any unsupported URL scheme (all URL's other than 'http' or 'https') :
ex : "docling ftp://some-server/file.pdf"

Docling version

Docling version: 2.90.0
Docling Core version: 2.74.0
Docling IBM Models version: 3.13.0
Docling Parse version: 5.9.0
Python: cpython-313 (3.13.13)
Platform: Windows-11-10.0.26200-SP0

Python version

Python 3.13.13

[dosubot]

Thanks for the detailed report! I can confirm the root cause.

The issue is in resolve_source_to_stream() in docling_core/utils/file.py [1]. When a URL like ftp://some-server/file.pdf is passed, Pydantic's AnyHttpUrl validator rejects it (since it only accepts http:// and https://), but the resulting ValidationError is caught by an overly broad except ValidationError block that assumes "not a valid URL → must be a file path." It then tries to read ftp://some-server/file.pdf as a local filesystem path, which naturally fails.

except ValidationError:
    try:
        local_path = TypeAdapter(Path).validate_python(source)
        stream = BytesIO(local_path.read_bytes())
        ...

There's no check to distinguish between an unsupported URL scheme (has ://) and a legitimate local file path. A fix would be to detect URL-like inputs (containing ://) and reject unsupported schemes explicitly before falling back to file path resolution — something like:

if "://" in str(source):
    scheme = str(source).split("://")[0].lower()
    if scheme not in ("http", "https"):
        raise ValueError(
            f"Unsupported URL scheme: '{scheme}'. Only http:// and https:// are supported."
        )

Note that this lives in docling-core, not in the main docling repo, so a fix would need to be submitted there [2].

To reply, just mention @dosu.

---

Share context across your team and agents. Try Dosu.