Re: bgb, references for mailing lists?...

[cr88192 sydney]

err, bfd... I am not terribly knowlegable about it and I am not sure how 
well it would for for things other than object-files.

for the fs I was imagining using an object mesh like design. before I had 
planned on an "semantic" fs model (and that is about what I had coded). I 
had come to realize that that model would not work well for what I was 
imagining, so I came up with a new one. I had figured: why not rip off what 
I can from scheme? some details do not so cleanly map over but I can work 
them out.

misc: a gc was supprisingly easy to write, though mine was based on my mm 
which used the first fit algorithm. I have been planning on writing an 
alternate allocator for small stuff (a cell based or buddy allocator), as my 
allocator has an overhead of 32 bytes/object. I had imagined a cell or buddy 
allocator would be inefficient when dealing with larger objects. sorry I 
don't know how the guile gc/allocator works... I had imagined 2 allocators 
would be complicated though. I had written part of a cells allocator, it 
works by allocating objects and mapping cells and a bitmap into them, my 
cell allocator does not use best fit, as I imagine that most objects are 
only only a few cells. each cell objects has an overhead of 12 bytes (4 
bytes header, 4 bytes backlink, 4 bytes block-head reference). the cell 
index is computed from the address (I could do without the block-head 
reference, but more general code expects it).
I do crappy at this.

I doubt I can use guile in my kernel, as I imagine it might be a lot of work 
to rewrite to work in it. I might have to write my own... actually this 
almost take priority over new fs, as they would need to be effectivly 
interconnected.

my current fs stuff is a cheap vfs rip... with support for: fat (crappy, 
hackish, buggy); devfs (lists known devices); and ramfs (ram based fs, not 
persistent, used as root). it is buggy, and it is heirarchical (as when 
first writting it I didn't have many other ideas, except relational and I 
didn't want to try to make my old dbms code work as a fs). the "semantic" 
model in my oppinion is just a tweaked out heirarchical model.
I imagine an object mesh could emulate the other models... my early fs spec 
involved using vectors to emulate b-trees. I will not go into the details.

I hope I don't overpromote this...


_________________________________________________________________
Chat with friends online, try MSN Messenger: http://messenger.msn.com