Stream image data in load() instead of reading into memory

[Honny1]

No description provided.

[Honny1]

/packit retest-failed

[inknos]

Edit. I was playing with cursor locally and didn't realize it uploaded the review, didn't want to do that but here we are.

My notes:

I was going to simply approve the pr since I think it's good. didn't think about the cursor nits to be honest and I prefer concise code as you did much more

---

Cursor:

Review: Approve with minor suggestions

The change is correct, well-scoped, and CI is fully green. The core benefit — avoiding loading multi-GB tarballs entirely into memory when file_path is used — is real and valuable. The APIClient._Data type alias already includes IO, confirming file-like objects are a supported data type.

Correct and positive

• Memory efficiency: Path.open("rb") streams data to requests, which reads it incrementally during the HTTP POST. Large tarballs no longer occupy full memory.
• Context manager usage: Both Path.open() and io.BytesIO support the with protocol. The file handle is properly closed when the with block exits.
• Generator safety: The returned generator does not depend on the stream. By the time return _generator(response.json()) executes, the HTTP request is complete and the response body is parsed. The generator only calls self.get(item) on image names — no risk of accessing a closed stream.
• Test update is correct: Mocking pathlib.Path.open to return io.BytesIO(b"mock tarball data") is a proper mock for the new streaming behavior. The import io addition is needed.
• Type safety: The _Data union type in podman/api/client.py explicitly includes IO, so passing a file object is type-safe.

Minor suggestions (non-blocking)

1. Unnecessary wrapping for the data path: When data (bytes) is provided, wrapping it in io.BytesIO(data) adds a copy and context manager overhead for no streaming benefit — the bytes are already in memory. The old code passed data directly to post(). Consider keeping a fast path that avoids the wrapper when only data is provided. This is a style/performance nit — the current unified with is more concise and still correct.

2. PR description is empty: Adding a one-liner like "Avoid loading entire tarball into memory when file_path is used; stream the file data directly to the HTTP request instead" would help reviewers and future archaeology.

3. Edge case data=b"": Verified that the validation if not data and not file_path correctly catches empty bytes (not b"" is True), raising PodmanError. No issue.

LGTM — nice focused improvement.