apply-patch keeps refusing a stanza that deletes a file whose bytes here are not text, and the asymmetry with the crlf deletion rule is now a recorded decision rather than a comment: format-patch never writes a deletion of a file that was binary on the authors side, since the seam classifies that row as binary and the exporter omits it, so such a stanza is either one loot wrote for other bytes or a hunkless one from elsewhere, and git apply refuses the git spelling of the latter without full index hashes, which apply-patch does not read. ADR 0082 and the comments at the deletion arm and in the module header say so, and a new test pins both halves in one fixture: the export of a change deleting a binary and a text file carries a stanza for the text file alone, the git binary deletion stanza spliced into that patch is refused and nothing is written, and the exported patch then applies over the same bytes. moving the binary refusal below the deletion arm turned it red (11 passed, 1 failed), and letting the seam treat a binary deletion as text turned it red at the exporter half (0 passed, 1 failed). no behaviour changed, and the workspace suite is green (#1809) 1bb3f26f · dbf3dbe6…
Renames are not followed. loot's tree maps a path to an address, so a rename is a delete and an add. This list is the history of the name, not of the bytes.