RE: stdio&unistd for cross compilation

[John Vickers <jvickers at acorn dot com>]

On Tue 21 Apr, Don Bowman wrote:
> > I can see where this would have been possible back when gcc 
> > had a much simpler
> > libgcc library.  However, I doubt that the usefulness of this is great
> > enough that we should change what we are doing.  As we add 
> > more features to
> > gcc, libgcc.a is unavoidably becoming more system specific.
> > 
> > The runtime libraries, libf2c, libstdc++, libobjc, have 
> > always been system
> > specific, so it makes no sense for EGCS as a whole to be 
> > system independent.
> 
> Its still possible.
> The runtime libraries are not used on (all) embedded systems.
> It was in fact possible until Friday to build egcs (and all previous gcc)
> without using target headers.
> With this innocuous change, it become significantly more difficult
> to configure and build gcc/egcs.

Indeed.

How is egcs supposed to be configured if it is to be used to write a new
operating system kernel, or to target a bare machine ?

The doc says you must get suitable header files.  This
documented bejaviour has always been at least arguably broken, for 
a cross-compiler, but it hasn't been an issue in the past,
because the headers haven't really been necessary (-Dinhibit_libc).

Perhaps for bare targets, it would be possible to assume
ANSI prototypes for the (few) needed functions, and to provide
a list of required functions in the doc.

John.