[YuangGao]

Closes #4269

Rationale for this change

• Appending an empty table creates a snapshot with no manifests, and _files / _all_files then call pa.concat_tables([]), which raises ArrowInvalid.
• Both now return an empty table with the files schema when there are no manifests to read.

Are these changes tested?

• test_inspect.py — files, data_files, delete_files, all_files, all_data_files and all_delete_files on a snapshot without manifests.

Are there any user-facing changes?

• These metadata tables return an empty table instead of raising.

[oresttokovenko]

@YuangGao What if a table already has data and someone overwrites it with an empty table (using the overwrite() method in pyiceberg/table/__init__.py): files() comes back empty, but all_files() still returns the old data file, is that what you'd expect? I added a diagram to help visualize this:

the situation you're merging a fix for

   flowchart TD
       A["no snapshot"] --> B["append(0 rows)"]
       B --> C["snapshot with 0 manifests"]
       C --> D["inspect.files() / all_files()"]
       D --> E["results = []<br/>unique_manifests = {}"]
       E --> F{"guard?"}
       F -- "before: none" --> G["ArrowInvalid"]
       F -- "after: not empty" --> H["empty_table(), 0 rows"]

Loading

what the new logic can introduce

flowchart TD
    A["append 3 rows"] --> B["snapshot 1: APPEND<br/>1 manifest, 1 data file"]
    B --> C["overwrite with an empty table"]
    C --> D["snapshot 2: DELETE<br/>1 manifest, 0 entries"]
    C --> E["snapshot 3: APPEND<br/>0 manifests"]
    D --> F["current snapshot = 3"]
    E --> F
    F --> G["files(): 0 rows"]
    F --> H["all_files(): manifests from snapshots 1 and 2<br/>unique_manifests = 2"]
    H -->|"guard on all snapshots"| I["1 row, history kept"]
    H -->|"guard on current snapshot only"| J["0 rows, history lost<br/>13 tests pass"]

Loading