libfat : maybe you need to loose some weight :D

Written by cafealpha 3 comments
Classified in : Saturn, Programming, Mylife Tags : saturn, memcart, kai, fat32, library
So as you may know, I'm (still) developing the hardware/firmware/everything else for my SD card-based Saturn Cartridge.

One the main module for the firmware of this cartridge is File System access library.

As you may know, there are many File System existing in real worls : FAT, NTFS, exFat, ReiserFs, etc.
And in my project, I choose to use FAT32 in order to access files and directories.
Why FAT32 ? Just because it offers the best rate of advantages vs disadvantages in my project scope.
Advantages :
 - Available everywhere, from Windows 98 to every Linux distribution.
 - Detailed (?) specs available on MS website, lots of free source code available elsewhere.
 - Little resource requirement (at least 1~2KB of work RAM, and few tenths of program memory for code).
Disadvantages :
 - A little hard to understand why internal data structure is so messy ... more understandable after reading FAT16 specs.
 - Complex Long FileName support, and patented by MS too.
 - Write performances may be poor under certain conditions. Example : fragmented file system + large partition size.
 - Free source code available everywhere is sometimes buggy / bloated.

At the very beginning of the project (nearly 4 years ago !), I choose DharmaniTech FAT32 library , and it was easy to use, so that I could read files without many problems on software development side.
But it only supports <512 bytes file write, and a lot of advanced features weren't available, so I then switched to thinfat32 library .
Thinfat32 library was a pain to adapt to Saturn, and some advanced features weren't available, or didn't worked, so I rapidly switched to libfat .

Libfat is great, works without any major trouble, and even supports FAT16
So I kept libfat as FAT32 library for around 3 years.
But one problem of libfat is ... its weight (<- yeah, I wanted to put this pun somewhere ).
 - Sector caching is cool, but library doesn't work without it, so we need at least 8KB of work RAM for it.
 - Source code is very complete, but also very complex, so unless you aren't a talented developer, so can't customize it easily ... and unfortunately I'm not a talented developer
 - Library supports nearly everything for file/directory manipulation, so size after compilation is a little big (around 50KB IIRC).

You may ask : "why complaining about few KBs of RAM ? Saturn got plenty of RAM to use !"
Well, yes and no. My project (on game game backup from/to SD card) of resources :
 - Work RAM : 24KB ... not contiguous : 16KB+8KB
 - Program memory : work RAM, or cartridge ROM (1MB). I choose to run program directly from ROM.

And, in a (maybe) future hardware project, I may need to port FAT32 library to microcontroller, so few resources will be mandatory.

So last month, I dropped libfat, and started to look for a simpler & better FAT32 library.
When selecting a starting point for development, my choice pointed to thinfat32, not because it was perfect, but because I had some experience with it in the past, so that I know where I should start to fix things

First, I dropped some features I don't need :
 - Long FileName when writing files : LFN can be read, but only 8.3 files can be written ... it's enough for my needs
 - Long path : changed maximum size for file path from MAX_PATH (260 bytes) to 96 bytes ... it's enough for me, and saves some RAM.
 - Only one file open at once : it saves some RAM.
 - Other ? (don't remember)

Then started to fix/debug/etc the library :
 - Big/low endian support ... Saturn and PC endianness is different
 - Directory creation fix.
 - File creation in non-root directory fix.
 - Added test bench under windows ... there was one stub in original sources, but with very limited test range ...
 - Better log support.
 - FS info support.
 - FAT16 support ... not working yet (), but at least it works better than in original sources
 - File append fix ... original sources support file append ... only in the case source file is multiple of 512 bytes ... this bug took me many time to spot
 - Clock support when setting file timestamp ... original sources supported only fixed (dummy) timestamp.
 - File listing within a directory.
 - Source code cleanup.
 - MANY fixes.

Complaints about thinfat32 library may look severe, but I really thanks its author to publicly release it !
Thanks to him, I know have something that fits my needs, and have a deeper knowledge regarding FAT32 structure

Last but not least, here are some screenshots :
data/images/20151021_thinfat_testprog.tb.png
Test program contents :
In this screenshot, I first create one directory,
then write 3 files inside it.


data/images/20151021_thinfat_testbench.tb.png
Test program execution - just in order to show
a command line screenshot on my blog

data/images/20151021_thinfat_testresults.tb.png
Result in (virtual) test drive :
one directory and 3 files are created !

This is just the beginning of the development : I still need to test under Saturn emulator (yabause ), then on real hardware (Saturn + custom cartridge ).

Read more libfat : maybe you need to loose some weight :D

Rss feed of the tag