Re: Solving the config.h nightmare ?

[Daniel Reed]

On Sun, 23 Apr 2006, Bob Friesenhahn wrote:
> On Sun, 23 Apr 2006, Daniel Reed wrote:
> > When is the format for the buffer set, at GraphicsMagick ship time, 
> > GraphicsMagick build time, dependent program ship time, dependent program 
> > build time, or dependent program run time?
> In this case, at GraphicsMagick build time.  Certainly the design can be

But then we're back to ABI breakage. I build GraphicsMagick 
--with-buffertype=uint16_t, build dependentprog against your installed 
headers, then later rebuild GraphicsMagick --with-buffertype=uint32_t. Or 
I install a copy of dependentprog built by someone who had used a uint32_t 
while my GraphicsMagick package (possibly from another package repo) was 
built using a uint8_t.

Variable API is possibly convenient, but not all that robust. OpenGL, for 
example, has you pass pixel buffers around as void pointers with 
accompanying type, size, stride, etc. information, so it can choose an 
internal code path that handles your data the best, no matter what format 
you provide it in. And, I mean, that's what my favorite real time games 
are written using :)


> Some configuration options do need to be 
> remembered by the installed header files.

I am not sure that is true for any reason other than short-term 
convenience, which I do not believe is worth the long-term problems 
associated with basing an API (and hence the ABI) on any sort of 
variability.

--
Daniel Reed <address@hidden>      http://shell.n.ml.org/n/        
http://naim.n.ml.org/
... There is a lot of food in a supermarket, too, but a supermarket
isn't the best place to hold a dinner party. -- Christopher Faylor