Re: [Pnet-developers] pnetC: mman.h & co.

[Rhys Weatherley]

On Wednesday 13 August 2003 07:09 am, Mario D.Santana wrote:

> I tried tackling expat first and hit some errors.  Looks like it needs
> stuff from sys/mman.h so I grabbed glibc and started figuring that all
> out.  But the glibc stuff for mman.h (grep PROT_READ) seems to be
> platform-dependent and, having never implemented a C library before, I
> wanted to get some direction before I borked up the platform-dependent
> bits of pnetClib.  I also hit some syntax errors I'll submit as bugs to
> savannah.

The mman.h stuff (mmap, munmap, etc), doesn't make much sense in the pnet 
environment.  There's no way to memory-map a file in the C# library, for 
example.  And the /dev/zero memory allocation thing is just plain evil.

I suggest creating a set of dummy system calls that always return an error 
indication.  Hopefully expat is smart enough to back off to a non-mmap 
implementation if the call fails.  If not, you'll need to fix expat - an mmap 
using application that cannot back off is broken, by definition, since mmap 
won't work over pipes and sockets.

> Meanwhile, I'll go try and tackle the socket stuff.  (So you may hear
> from me again real soon.)

That will certainly be easier.  For ideas as to how it needs to be done, look 
at "pnetC/libc/unistd/unistd_glue.cs".  Basically, all socket-related 
functions need to be redirected through to "System.Net.Sockets.Socket" in 
"System.dll".

Cheers,

Rhys.