LFN support

[thejpster]

Was:

The LFN code assumes the code points are all valid UTF-16 and not surrogate pairs exist. This may not be a valid assumption. We should do a proper UTF-16 to UTF-8 conversion.

Now:

• Show LFNs when iterating directory Add LFN support when iterating directories #157
• Handle surrogate pairs in the UTF-16 input Add LFN support when iterating directories #157
• Delete LFNs when deleting a directory entry
• Open file using LFN (fn open_lfn_in_dir(&self, dir: RawDirectory, name: &str, mode: Mode)
• Open directory using LFN (fn open_dir(&self, dir: RawDirectory, name: &str)

Note that opening an existing file is easier, as we can match a &str against the LFN entries on disk. Creating a new file is more difficult because we need to:

a) work out how many 16-bit code units are required to store the filename
b) work out how many directory entries are required to store the long filename chunks and the real directory entry (13 code units per chunk)
c) find a gap in the directory with that many consecutive deleted directory entries, or at the end of the directory (using any deleted directory entries which are located at the end)
d) write out the LFN chunks as well as the actual directory entry

[jonathanpallant]

See #157

[thejpster]

I wanted to see how characters outside the BMP are encoded on Windows 11 on a FAT16 filesystem.

The file was called Smiley 😀.txt.

[2024-10-27T11:47:34Z DEBUG embedded_sdmmc::fat::volume] LFN Contents true 1 90 [0053, 006d, 0069, 006c, 0065, 0079, 0020, d83d, de00, 002e, 0074, 0078, 0074]
[2024-10-27T11:47:34Z TRACE embedded_sdmmc::filesystem::filename] LFN push 't'
[2024-10-27T11:47:34Z TRACE embedded_sdmmc::filesystem::filename] LFN push 'x'
[2024-10-27T11:47:34Z TRACE embedded_sdmmc::filesystem::filename] LFN push 't'
[2024-10-27T11:47:34Z TRACE embedded_sdmmc::filesystem::filename] LFN push '.'
[2024-10-27T11:47:34Z TRACE embedded_sdmmc::filesystem::filename] LFN push '?'
[2024-10-27T11:47:34Z TRACE embedded_sdmmc::filesystem::filename] LFN push '?'
[2024-10-27T11:47:34Z TRACE embedded_sdmmc::filesystem::filename] LFN push ' '
[2024-10-27T11:47:34Z TRACE embedded_sdmmc::filesystem::filename] LFN push 'y'
[2024-10-27T11:47:34Z TRACE embedded_sdmmc::filesystem::filename] LFN push 'e'
[2024-10-27T11:47:34Z TRACE embedded_sdmmc::filesystem::filename] LFN push 'l'
[2024-10-27T11:47:34Z TRACE embedded_sdmmc::filesystem::filename] LFN push 'i'
[2024-10-27T11:47:34Z TRACE embedded_sdmmc::filesystem::filename] LFN push 'm'
[2024-10-27T11:47:34Z TRACE embedded_sdmmc::filesystem::filename] LFN push 'S'

We can see that 😀 (\U{1F600}) was encoded as ['\U{D83D}', '\U{DE00}'], i.e. as a surrogate pair from the D800-DFFF range. Therefore to turn an LFN back into a String, we need to handle surrogate pairs.

[thejpster]

Note that worst case, a filename could consist of 255 codepoints, each between \U{0800} and \U{D7FF}. These would each be encoded as 3 bytes in UTF-8, meaning the worst-case size for an LfnBuffer is 765 bytes.