Re: Size optimization for objects

[Eryndlia Mavourneen <eryndlia@gmail.com>]

On Friday, July 19, 2013 12:02:38 AM UTC-5, Randy Brukardt wrote:
> "Rego, P." <pvrego@gmail.com> wrote in message 
> 
> news:ca34e347-b53e-4119-b6d1-0cba6f40be79@googlegroups.com...
> ...
> > However when I add all the column, the result also seems incoherent, 
> > around 8GB.
> > The real bla.out generated is ~478kB.
> 
> I wouldn't expect the *uninitialized* data to have any effect on the size of 
> the .out file, either. The contents of the .o files are combined to create 
> the .out, and I don't think the uninitialized data appears in it, either, 
> just being a length given to the system loader. (That's definitely how it 
> works on Windows, and Windows NT copied [not very well] a lot of Unix stuff, 
> mainly because they started with an existing Unix targeted C compiler.)
> 
>                              Randy.

That's also the way it works on VMS and RSX.  Depending on the loader, uninitialized data is typically left uninitialized (you get whatever happens to be in that memory location upon loading) or zero-filled.

The development of the Win32 kernel, and I suppose a number of other parts of Windows NT, were led by Dave Cutler, who did the design of RSX (on the PDP-11) and then VMS on the VAX (CISC) and Alpha (RISC) machines.  This is a big reason you have device driver forking in Windows, which was introduced into RSX-11M first and then later used in VMS, instead of some other scheme (such as privileged, high-priority tasks for processing data emitted by device drivers as used in RSX-11D, also designed by Dave).

-- Eryndlia (a.k.a. KK1T)