pytc.df.thc.convert_store_to_rank_major¶

pytc.df.thc.convert_store_to_rank_major(src, dst, *, x_dataset='X', x_rm_dataset='X_rm', row_block=8)[source]¶

Whole-store conversion: every dataset copied; X kept AND a rank-major twin appended.

The ISDF store carries more than X (K1/K3/D kernels, metadata); a store the driver can actually load needs all of it. Top-level datasets and file attributes are copied verbatim – including x_dataset itself, since legacy consumers read it as (nmo, nmo, rank) – and a rank-major twin x_rm_dataset is added alongside (with the x_layout attribute and a provenance stamp of the source X).

Writes go to <dst>.tmp and are atomically renamed into place, so an interrupted conversion never leaves a partial file at the destination.