Re: [lwip-devel] [bug #20254] Initializing static/global variables

[Jonathan Larmour]

Goldschmidt Simon wrote:
> > So what about removing all the extra 
> >
> > xxxx = 0;
> > xxxx = NULL;
> >
> > for all globals and static variables, if there are any?
> >
> > These should be taken care of the the compiler clearing the 
> > BSS on startup, well all the compilers I use work this way.
> >
> > Should not make the code any larger
> >
> > Also change static tables to const tables (sockets.c) moves 
> > data that is only read into the rom data section, no more wasted ram
>
> I think that's a good idea. And that is why I'd like to leave the bug
> open, not because of the buggy compiler.

This bug was then closed, and I don't see a replacement about this. Did I 
miss it? Shall I open a new bug for this different issue - removing 
unnecessary zero initialisation of statics/globals, and making sure 
constant data is marked const?

Although I don't see what there is in sockets.c or anywhere else that can 
be const.

> Note that many pieces of embedded software that are supposed to be
> ultra-portable provide compile switches whether to initialize the data
> or not. I'll try to search for some C definition if the initialization
> to zero is mandatory or not.

It is mandatory.

It is also faster for the loader to initialise a large memory region (the 
bss section) to 0 all at once - it can write to memory at full speed, and 
it usually happens in assembler. And of course it definitely takes less 
code than initialising bits and pieces around the place. I can't envisage 
any circumstance where debugging bits individually would reduce memory use, 
and of course for many systems it would definitely be superfluous.

Jifl
--
eCosCentric Limited      http://www.eCosCentric.com/     The eCos experts
Barnwell House, Barnwell Drive, Cambridge, UK.       Tel: +44 1223 245571
Registered in England and Wales: Reg No 4422071.
------["The best things in life aren't things."]------      Opinions==mine