Re: bss section not getting zero-initialised vars

[John Vickers <John dot Vickers at pace dot co dot uk>]

Mark Mitchell wrote:
> 
> >>>>> "Jamie" == Jamie Lokier <jamie.lokier@cern.ch> writes:
> 
>     Jamie> Richard Henderson wrote:
>     >> > Yes, this is true.  gcc doesn't have the optimization where
>     >> we move > zero inited things into bss.
>     >>
>     >> Yes, and there's a good reason historical for this.  Providing
>     >> an initialization is a (reasonably portable) way to force the
>     >> variable into the data section, where it can then be edited in
>     >> the binary.
> 
> I think this is one of those places where the embedded folks and the
> non-embedded folks don't see eye to eye.

I think a lot of embedded folks would like to have such an optimisation,
in order to shrink their ROM/Flash images.

I guess gcc would provide something like -fno-initialised-variables-in-bss
or what have you.

Some embedded folks even try to write portable, compliant code...

On the other hand, I once had to program a family of embedded 68k boxes
which had battery backed RAM.  Bss was *not* zeroed on leaving
reset:  variables which needed to be non-volatile were not explicitly
initialised, so they went into the battery-backed bss.  Names witheld
to protect the guilty.

I hope that any feelings of nausea or giddyness this message may have
caused will pass quickly.

Regards,

John.