[PATCH 2/2] first stab at NOFORK applet support

[Rob Landley]

On Tuesday 09 May 2006 4:27 pm, Bernhard Fischer wrote:
> Just a quick sidenote..
> Perhaps it'd be better to use bitfield for the pure true/false values like
> be_quiet, in_place:
> struct sed_globals
> 	/* options */
> 	__extension__ unsigned be_quiet:1;
> 	__extension__ unsigned in_place:3;/*or how many you need*/
> ...
> }
> Even with the implied padding you'd call for by only using 1+3, you'd
> save a little bit there, eventually.
>
> cheers,

Bitfields tend to save data at the expense of code size (the extra shift and 
mask), but now that it's in a structure that's definitely one of the options 
we've got.  (P.S.  I love bloat-o-meter.  It's really really really really 
cool.  If you haven't tried "make bloatcheck" yet, you should. :)

I just converted mdev.c in my local tree to use a similar struct for its 
measly two static variables, and now I'm working out the shared 
infrastructure to put them into a union.  Once that's in, we can convert 
other stuff over one at a time.

One thing I'm worried about is that the union might screw up the compiler's 
optimizer.  (The "strict aliasing" stuff, and such.)  I could always make 
union vs separate structs a config option, although I'd really rather not...

Rob
-- 
Never bet against the cheap plastic solution.