v0.4.32

Added

• (fs) one no-alloc interface, a generic mount, formatters and partition writers

Added

• (fat) Volume::format / format_at: the allocator-free driver formats.
  FAT12, FAT16 or FAT32 and the cluster size are chosen from the volume's
  size the way mkfs.fat and Windows choose them (FAT32 from 512 MiB, with
  Microsoft's cluster table and the data region on a cluster boundary), or
  set through FormatOpts. Only metadata is written, a sector at a time
  through one sector of stack, boot sector last. fsck.vfat passes every
  flavour at 512- and 4096-byte sectors, fresh and after use, and inside an
  MBR partition; the hosted Fat32 reads them.
• (exfat) Volume::format / format_at, likewise: Microsoft's default
  cluster sizes (or VolumeFormatOpts::cluster_size), both boot regions with
  their checksum, one FAT, the allocation bitmap, a label, and the
  specification's recommended up-case table — the same 5836 bytes
  mkfs.exfat writes, so non-ASCII names fold as they do on any other card.
  The boot regions are written last. fsck.exfat, which checks the boot and
  up-case checksums, passes volumes from 8 MiB to 4 GiB at both sector sizes
  and inside a GPT partition; the hosted Exfat reads them. (The hosted
  Exfat::format still writes an ASCII-only table.)
• (device) device::mbr::write / set_entry: write a partition table, or
  change one slot of the one there, through a SectorDriver and a sector of
  scratch — boot code kept, CHS fields filled, the layout validated (inside
  the medium, no overlaps) before anything is written. Type-byte constants
  (FAT32_LBA, EXFAT, LINUX, …) and FIRST_LBA come with it. sfdisk
  reads back what they write, including changes to tables it wrote.
• (device) device::gpt::write / set_entry / erase: lay a GPT down
  (protective MBR, both headers and arrays, CRCs computed a sector at a time),
  change one entry of an existing table — rewriting both copies, which
  repairs a damaged one — or remove a table. Backup copy first, primary
  last, so a torn write leaves one whole table. Layout says where usable
  sectors start and end and where the 1 MiB-aligned first partition goes;
  Guid::random_v4 makes a well-formed GUID from the caller's random bytes.
  sgdisk -v passes the tables and sgdisk -i reads back every field; the
  device_tables fuzz target now runs set_entry over arbitrary media (it
  found an overflow on a header pointing its array off the medium, fixed).
• (fs) fs::volume::format and FormatAs: format any compiled-in
  filesystem through the generic layer — littlefs on a card through
  SectorFlash — and get an AnyVolume back. FormatAs::sd_card picks what
  the SD specification requires for a card's size: FAT up to 32 GiB, exFAT
  above.
• (fs) fs::volume: one interface over every allocator-free driver.
  fs::fat::Volume, fs::exfat::Volume and fs::littlefs::Volume all
  implement its Volume, VolumeFile and VolumeDirIter traits, so code
  written once — open, read, write, seek, truncate, list, create and remove
  — runs on any of them. Sizes are u64 and names bytes (with name_str)
  whatever the driver stores; a listing never reports FAT's . and ..;
  every driver error answers kind() with a shared ErrorKind, so generic
  code can tell NotFound from Io without knowing whose error it holds.
  The traits are generic rather than dyn, need no allocator, and cost
  nothing over calling the driver directly. The drivers' own APIs are
  unchanged.
• (fs) fs::mount / fs::volume::probe: find out what a card holds and
  mount it. The whole medium is tried, then each GPT or MBR partition in
  order, and at each start the compiled-in filesystems are recognised by
  their own signatures — exFAT boot sector, a FAT BPB that validates, a
  littlefs superblock log — never by a partition type. The result is an
  AnyVolume, which implements the same traits, and whose AnyError keeps
  the driver's error whole. probe answers without taking the card. Checked
  against volumes mkfs.fat (inside an sgdisk GPT), mkfs.exfat (whole
  card and MBR slot) and littlefs's C implementation (512-, 1024- and
  4096-byte blocks, after enough churn to compact the superblock) wrote: each
  is found, edited through the traits, and passed by its own tool afterwards.
  A new volume_mount fuzz target feeds probe and mount_found arbitrary
  media.
• (device) device::SectorFlash: a SectorDriver presented as a
  FlashDriver, one erase block per run of sectors — littlefs on an SD card
  or eMMC part, where never overwriting live metadata is exactly what a card
  pulled mid-write needs. fs::mount uses it to mount littlefs found on a
  card. Erase really writes 0xff, as the flash contract promises the
  driver.

Changed

• (fat) DirEntry::name returns &'a str, borrowed from the iterator
  rather than from the entry, like exFAT's and littlefs's already did. Code
  that compiled before still does.