[patch] minor toplevel configure.in simplification

J. Johnston jjohnstn@redhat.com
Thu Jul 11 14:32:00 GMT 2002

Doug Evans wrote:
> Nathanael Nerode writes:
>  > I want to do more cleanup on the stuff related to deciding what to do with
>  > newlib, but I have to figure out what the correct behavior of '--without-newlib'
>  > on 'always use newlib' targets is.  I'm thinking it should mean 'use installed
>  > newlib, not the one in the tree', but I'd love it if someone else could
>  > confirm that that's reasonable behavior.
> I wonder if it depends on one's meaning of "always".
> Is it a misnomer?
> There should be no reason a user of any port can't supply their
> own libc/libm/blah.
> Thus methinks there's no such thing as "always use newlib"
> [instead it should be "default to newlib"]
> and --without-newlib means what it says it means: don't use newlib,
> not even an installed one.

You should probably be including the Cygwin group in this discussion.  Cygwin requires
newlib because some of the library is used from newlib and some of the library is
added or replaced in winsup.  Cygwin takes the newlib libc and libm apart and reassembles
it with its own pieces added.  It isn't a well-documented split, especially regarding which
header files come from newlib and what comes from winsup.  Currently, winsup has a
hard dependency on newlib being in the build and source trees (see winsup/Makefile.common) so without changes to winsup, configuring --without-newlib
must be prohibited when building target-winsup.

I can see merit in allowing specification of a pre-built or installed newlib to winsup.  I have
serious doubts that a generic C library could be simply substituted in newlib's place but
I will leave the evaluation of that to the Cygwin folks.  Perhaps we could extend --with-newlib 
to allow --with-newlib=xxxx to allow specification of where to find an installed newlib.

-- Jeff J.

More information about the Newlib mailing list