---

name: windows-memory-vsi description: "Routes Rasterio windowed I/O, block processing, in-memory files, archive/VSI paths, and custom opener workflows." disable-model-invocation: true metadata: disco-role: operating license: BSD 3-Clause

Windows, Memory Files, and VSI Paths

Use this sub-skill when a user wants to read or write rasters in windows, process blocks concurrently, keep rasters in memory, or open zip:// / file:// / GDAL VSI paths.

Typical requests

• "How do I read only a window of a raster?"
• "How do I use MemoryFile instead of a temporary file?"
• "How do I open a raster inside a zip file?"
• "How do I process blocks concurrently?"

What this sub-skill owns

• Window, Window.from_slices, from_bounds, window_index, get_data_window, and block-aligned window helpers.
• MemoryFile, ZipMemoryFile, and safe local byte-stream workflows.
• zip://, zip+file://, file://, and the rasterio._path/GDAL VSI path conventions.
• Custom opener use cases and the practical limits around sidecar files.
• Safe windowed concurrency patterns.

What it excludes

• Dataset profile and basic write-time metadata choices, which live in dataset-io-profiles.
• Geometry masks/rasterization, which live in features-masks.
• Reprojection, WarpedVRT, and merge/stack workflows, which live in reprojection-merge-vrt.
• rio CLI flag selection, which lives in rio-cli.

Read first

• references/api-reference.md for the verified Window and MemoryFile signatures.
• references/workflows.md for the common windowed read/write, MemoryFile, and VSI recipes.
• references/troubleshooting.md for invalid windows, closed MemoryFiles, and opener/VSI limitations.
• scripts/check_install.py for a repo-independent smoke check.

Helper scripts

• scripts/windowed_copy.py — safe block-processing helper adapted from the thread-pool example.
• scripts/memoryfile_smoke.py — open a local file in memory and print a compact summary.
• scripts/vsi_smoke.py — open a zip:// / file:// URI and print the dataset summary.

Workflow shape

1. Decide whether the task is pixel-based or world-coordinate-based.
2. Use Window and block_windows when the task is about chunks of pixels.
3. Use get_data_window when you need the tight valid-data extent of a masked raster.
4. Use MemoryFile when the input is bytes or when a temporary file would be awkward.
5. Use zip:// or file:// URIs when the dataset is already packaged locally.
6. Fall back to a real filesystem path when a custom opener would hide required sidecar files.

Decision points

• Window.from_slices is the right tool when the user already has Python-style row/column slices.
• from_bounds is the right tool when the user already has spatial bounds and the transform is known.
• MemoryFile is ideal for bytes in RAM, but a closed MemoryFile cannot be reused.
• ZipMemoryFile is useful for archive members, but not every sidecar-dependent workflow fits through it.
• The concurrent helper should start conservative and lock reads/writes separately.

Common mistakes

• Confusing row/column order with x/y order.
• Treating a closed MemoryFile like an open file object.
• Assuming a custom opener will expose sidecar mask or metadata files.
• Using a window that extends beyond the dataset bounds without checking the expected behavior.

Good validation path

• tests/test_windows.py::test_read_with_window_class
• tests/test_windows.py::test_window_from_bounds
• tests/test_windows.py::test_round_window_to_full_blocks
• tests/test_windows.py::test_data_window_full_2d
• tests/test_memoryfile.py::test_initial_bytes
• tests/test_memoryfile.py::test_zip_file_object_read
• tests/test_path.py::test_read_vfs_zip

What a future agent should be able to do here

A future agent should be able to answer common windowed-IO or in-memory-file tasks using this sub-skill plus the bundled files, without reopening the original repo.