Re: cross compiler questions

[Jeffrey A Law <law at cygnus dot com>]

  In message <199711282126.QAA19627@altera.gaithersburg.md.us>you write:
  > - Before building the cross-compiler, I put the native C library
  > includes and libs in /usr/local/rs6000-ibm-aix3.2.5/include and
  > /usr/local/rs6000-ibm-aix3.2.5/lib, via symlinks to NFS mounts.  Can I
  > remove all those symlinks once the build is done, or does gcc still
  > need them?
You're still going to need them since they'll be referenced when you
actually cross compile something for the rs6000-ibm-aix3.2.5 target.

  > I build the cross-compiler on several different NFS servers, one for
  > each architecture that I am using.  The machines that need to run them
  > then mount from the appropriate server.  Some of these machines are
  > actually rs6000-ibm-aix3.2.5 machines that have no native compiler.
  > So I build the native compiler and NFS export it from a server, and
  > mount it where it is needed.  Except that this does not work, because
  > the header files are not on the machines that don't have a compiler.
So, is there no /usr/include on the client machines?

Header files for native builds first come from gcc's private directory,
which is set up and installed automatically.

Then gcc will search /usr/include


You might be able to define SYSTEM_INCLUDE_DIR when building cccp.c
to point to a directory that is an NFS mounted copy of /usr/include
from your server.

  > and some .a files.  It seems as though the build process could be
  > streamlined quite a bit by getting the .a and include files from
  > another machine that has already done a cross-build, and just making
  > the necessary executables.
Yes, but this adds quite a bit of complexity to the build/install
procedure for little benefit -- what you're trying to do is pretty
unusual.

  >  The rs6000-ibm-aix3.2.5 cross-build takes
  > about 4 hours on a couple of the machines I am using, and a lot of it
  > seems to result from building varieties of a couple of the .a files.
Yes.  They're multilibs.  For example it might be building both hardware
and software floating point libraries, libraries for specific variants
of the rs6000/ppc, etc etc.

If you don't use -m options, then you can probably avoid building
multilibs with the "--disable-multilib" configure option.


  > - Are there any instructions on how to do builds, and the various
  > options that are available?  I have been assuming that the
  > documentation is not up to date, hopefully I am wrong.
The documentation is out of date, but I doubt any docs will really
describe how to set up what you're trying to do :-)

If you want to see what the egcs docs are going to look like, see
http://www.cygnus.com/egcs/install

jeff