[openbeosnetteam] Re: One small step for net team, but... but one small step forward?

> Well, in these cases it doesn't matter much, but I think _howmany() 
> would be better.

Will fix that soon, then.
 
> If I implemented it correctly, it doesn't depend on FD_SIZE at all, 
it 
> just takes the number of descriptors and use this to create the 
> internal array.
> But I don't think this is important to fix for the R5 implementation 
- 
> I guess this is a *very* rarely used feature, and documenting the bug 
> for R5 should be enough.

It's NOT implemented correctly (fd_set fields in a struct, no dynamic 
sizing of the array).
But, yeah, you right, it doesn't worth it to fix it.
Just need to remember this issue.

Let's add a docs/develop/net/release_notes.html soon ;-)
 
> Ah, okay, I just misread you then; they are using get_module() to 
load 
> modules, they don't use load_addon() or anything like this anymore.

Yeap, exactly.

> Although I think it would be possible to build them once as kernel 
> modules, having to care for set_sem_owner(sem, B_SYSTEM_TEAM) would 
be 
> a hack, I think :-)
> There are probably other issues as well.

That's pretty much all, in fact. And already taken care in old code.
#ifdef _KERNEL_MODE set_sem_owner() #endif is not that bad.
spawn_kernel_thread() is mapped to spawn_kernel() by net_server.
dprintf() to printf(), etc.

The final objective is to have a address space agnostic module code as 
much as possible, 
so that the stack debug platform could be use to coding new module, 
test them, fix them 
before having to do it in KDL for the more serious bugs...
 
-Philippe

--
Fortune Cookie Says:

There are people so addicted to exaggeration that they can't tell the
truth without lying.

Other related posts: