[YPARK]

The remaining commands get first-class views, under the one view rule from #42: on a terminal they open, and --batch-process or DATA_BEANS_BATCH=1 runs them straight through. MCP and qc's internal pooling always run in batch.

Browse

• histogram opens the stat explorer on the --stat and --dim asked (StatExplorer::on_stat); -o values are still written.
• row-names / column-names open a names browser (new interactive::list_tui): / filters (a regular expression when the text is one), Space marks, a marks the shown entries, w writes the marked (or shown) names, y copies them. A file -o is still written.
• list-h5 / list-zarr show the file as a tree in the same browser; Enter or the arrows open and close groups, and w writes paths. Batch output is byte-for-byte what it was.

Plan before writing

New interactive::plan_tui: a table of what will be written, notes, and settings on keys. Each change redraws the plan, Enter writes, and q writes nothing. A plan that can't be written says why, and Enter waits.

• merge-backend: each input's rows and columns, the rows every input has and those only some have, and the output size. c sets common rows and s squeeze. Files whose rows differ block until c.
• merge-mtx: the MTX sets found and their common rows; s squeeze.
• align-data: each file's data type and column group, and the rows and columns each keeps; r sets the data types.
• subsample: input against output size, and expected entries; n cells, g genes, s seed.
• split: train and test cells per fold; t test fraction, k folds, s seed, and g grid under --coord.

The merge, align, subsample and split bodies now take what the plan decided as arguments. Their batch behaviour is unchanged.

Import preview

Every from-* build ends with the rows' and columns' non-zeros on log bins, with the squeeze cutoffs marked. s squeeze and r/c cutoffs are settings. Enter keeps the build, squeezed if on; q deletes it.

Note

On a terminal, merge-*, align-data, subsample, split and from-* now wait for Enter. A script started from a terminal should export DATA_BEANS_BATCH=1 or pass --batch-process. The README says so.

Version

0.10.10. A patch release: the new library API (list_tui, plan_tui, StatExplorer::on_stat, zarr_io::output_path) is additive.

Testing

• cargo fmt --check, cargo clippy --all-targets --features sim -- -D warnings, and cargo test (219 lib, 97 bin, and the integration suites) pass.
• cargo check --all-targets passes with --features hdf5, --features alg, and --no-default-features --features aux,tui.
• New tests: list view (paths, filter with ancestors, folding, write marked/shown, drawing), plan view (blocked Enter, typed numbers, cancel, a setting's key never taken for moving), on_stat, the zarr and zip tree parsers, merge and align plan counts, split and subsample settings, and import-preview counts and histogram lines.
• Driven by hand in tmux on a small matrix: list-zarr tree, column-names write, histogram on mean with Tab to rows, merge-backend blocked then merged with common rows, split to 3 folds, subsample, from-mtx keep with squeeze and discard.