committed: merge with libada-branch

[Joseph S. Myers]

On Tue, 10 Feb 2004, Zack Weinberg wrote:

> Well, /a priori/ it seems to me that moving the sources will allow the
> build system to be made simpler and easier to understand.  It's also a
> golden opportunity - since you are renaming them all anyway - to
> switch to a more comprehensible naming scheme (you can't build GCC on
> a filesystem limited to 8.3 names anyway - so using krunched names
> doesn't gain anything anymore - and how about config/OSNAME/foobar.adb
> instead of XYfoobar.adb?)

And how about automatically generating some of the configuration files -
for example, those just containing definitions of target-specific
constants - from the target's C headers?  Although autoconfiscating for
some aspects of target behaviour is difficult, some such as constant
values can be determined at compile time.  In general the amount needing 
to be done to port to a new target should be minimised - especially if 
both the target CPU and the target OS are already supported, little should 
need to be done to support the combination of that CPU and OS.

> There is brief documentation of how to specify a set of multilibs in
> the GCC internals manual, but I don't know of any documentation
> anywhere of how to write a multilib-enabled language runtime Makefile.

The src tree has etc/configure.texi but I think that is now thoroughly out
of date.

-- 
Joseph S. Myers
jsm@polyomino.org.uk