PFS3 status update

[Piru]

I've been working on PFS3 by using a cross compiler on AMD64 linux box. This has prevented me from testing much of the changes so far, but I will do that shortly when I'll have access to my MorphOS box again.

So far I've:

• replaced all assembler parts with C code
• fixed several small bugs in the original code
• ported the code to compile with GCC, while making sure the original SAS/C build works ok as well
• ported the filesystem and the tools to MorphOS
• removed all static data from the filesystem making it possible to include it in the MorphOS boot.img

To do:

• add support for new MorphOS dospackets
• add a resethandler to commit any pending changes to disk before reboot
• testing all the new code and port, fixing any issues
• fixing the remaining serious bugs (MakeAnodeBitmap superindex array out of bounds access, beta warning #1, softlink support, PFSDoctor zeropage hits)
• possibly create a MUI GUI for PFSDoctor

Some links that might be of interest:

• SVN tree
  
• Bug tracker
  
• Feature requests
  

[Piru]

you've got to be kidding? the Perfect File System 3 code? you ported it?
i thought it was closed source, i.e. proprietary, paid for software.

It became open source recently: See http://www.amiga.org/forums/showthread.php?t=52358

i always used this file system, was so much faster and more reliable than ffs, tests i have done show it to be faster than sfs too.

Indeed. It's far the best filesystem on amigoid systems. It has some small tiny issues I intend to fix though.

Err, i'd love to see this file system back in development if you did manage to contact the author and get the code!
All i can say is wow. well done!

Getting the code wasn't my doing (see the other thread for details) but now that the source code is available I intend to help as much as I can.

[Piru]

Are you gonna include binaries for both 3.x and MorphOS platforms in the future releases? (I hope so).

This hasn't been decided yet and needs some careful planning. A filesystem is such a critical component that some serious testing should be done before a new version is released. No such testing/release planning has been done yet.

My plan is to include the PFS3 is some future MorphOS release at least. The filesystem will be inside the boot.img, making it possible to use PFS3 easily with the supported Mac systems, too. Without being inside the boot.img you would be limited to systems with RDB support or to mounting via mountlists.

Also the current 68k version functions just fine with MorphOS already, and since the filesystem is translated by the Trance JIT is likely to be exactly as fast as the native version will be.

For now I've omitted committing the MorphOS Makefiles in the SVN. This is because I have been unable to test much my new code. I wouldn't want anyone to hose their data with a buggy filesystem. Once I've tested and bugfixed the code I'll add the Makefiles. At that stage anyone with the SDK installed can build the filesystem and play with it. Obviously even then I'd seriosly recommend only doing that with some spare partitions and avoid using the SVN version in production environments.

[Piru]

What gcc arguments do you recommend to compile it with for fastest performance on a CS 060?

Just a simple -m68060 will suffice?

Last I played with 68k target gcc it was back in the 2.95 time. Back then gcc produced far worse code for 68k than SAS/C. I don't know if things has changed since, but in general -O2 with the flag specifying the CPU target (in this case -m68060) is usually the best choice. -O3 is worth trying as well, but usually it only leads to very minimal speed up with the expense of the binary bloating up. As always try couple of different optimization flag combinations and benchmark your code.

[Piru]

Anyone experience with vbcc? Got a mail from Stefan Haubenthal who wants to make PFS compatible with it.

I think my GCC fixes should go a long way fixing it for VBCC as well. I also made sure that the alignment patches (#pragma pack(2)) work with VBCC, too. No doubt there will be some things to fix with VBCC still.

I haven't used VBCC myself though.

[Piru]

@Michiel & @Piru

A friend of Michiel Pelt (PFS3 author) said to me few weeks ago that he found a bug :

'By the way... pfs3 progress is slow, we found a bug in it at the 4gb disksize threshold. We are working on it'


It's fixed ?

Well, depends on if he talks about these bugs:
https://sourceforge.net/tracker/?func=detail&aid=3290333&group_id=532591&atid=2163215
https://sourceforge.net/tracker/?func=detail&aid=3291407&group_id=532591&atid=2163215

I don't know of any "4gb disksize threshold" bug, but these are the closest. They're related to large filesystem (needing superindex blocks) and not 4GB disksize limit, though.