macOS's unzip decodes names in a ZIP "made by" MS-DOS as code page 437, even with the UTF-8 flag set

Fact

The unzip that ships with macOS (Info-ZIP UnZip 6.00, Apple build) ignores general-purpose flag bit 11 ("names are UTF-8") when an entry's version made by host byte is 0, MS-DOS. It converts the name from code page 437 instead, so a UTF-8 name comes out as ??????, and on APFS, which rejects the resulting bytes, extraction fails with write error (disk full?). Python's zipfile, ditto (Archive Utility) and bsdtar honour the flag and extract the same archive correctly, and unzip -t reports it as fine.

Why it matters

A hand-written ZIP writer usually copies the minimal DOS header layout and sets bit 11, which passes every check except extraction in a terminal. The error then points at disk space rather than at file names.

How to apply

• Write version made by as Unix: high byte 3, e.g. 0x0314.
• Then set Unix permissions in the upper 16 bits of the external file attributes, e.g. (0o100644 << 16) >>> 0. With Unix origin and attributes 0, unzip extracts every file with mode 000.
• Test with a real extraction (unzip -d out archive.zip) on macOS, not only with unzip -t.