Building EGCS from scratch

Building EGCS from scratch

[Steven J. Madsen]

I thought I knew what I was doing with all of this, but it seems that maybe
I don't after all.  The EGCS distribution has gotten so big at this point
that I'm at a loss where to start.  I can build GCC/G++ just fine, run the
tests and even install the compiler.  My problem is with everything else in
the top-level directory.

For example, it looks like there is a version of the standard C++ library
that is built.  However, when I'm doing this (on Linux, libc5, latest
everything), I get the following error while building libiostream.a:

make[1]: *** No rule to make target `stdio.list', needed by `libiostream.a'.

It seems that EGCS is sorely missing documentation on what needs to be built
for <foo> OS, and in what order.  I've been on the mailing list for a few
weeks now, and nothing of the sort has showed up in my mailbox.

Does anyone have any pointers?  (Sorry if this sounds frazzled, but I've
been spending the better part of my spare time the last two days trying to
grok this, and I'm just not getting it...)

-- 
Steve Madsen  <steve@moonglade.com>              PGP key available

Re: Building EGCS from scratch

[Jeffrey A Law]

  In message < 19970908234701.21678@calvin.moonglade.com >you write:
  > For example, it looks like there is a version of the standard C++ library
  > that is built.  However, when I'm doing this (on Linux, libc5, latest
  > everything), I get the following error while building libiostream.a:
  > 
  > make[1]: *** No rule to make target `stdio.list', needed by `libiostream.a'.
This is a bug in the linux config files in libio.  Jim posted a patch for
this problem a day or two ago.

  > It seems that EGCS is sorely missing documentation on what needs to be built
  > for <foo> OS, and in what order.  I've been on the mailing list for a few
  > weeks now, and nothing of the sort has showed up in my mailbox.
You can try the "bootstrap" target if you pick up the makefile fragment I
posted to the list a few days ago.

Or


configure  --srcdir=whereever
cd gcc
make bootstrap
make compare
cd ..
make

Jeff

Re: Building EGCS from scratch

[Joe Buck]

> > It seems that EGCS is sorely missing documentation on what needs to be built
> > for <foo> OS, and in what order.

Mike Stump writes:
> It isn't missing, though some of the simplicity seems to have been
> obscured over the years.  It is configure; make; make install, that's
> about it, and that is the order.

That procedure doesn't provide a correctly bootstrapped gcc at present for
egcs.  If you don't bootstrap, then the gcc binary depends on too many
variables, which will translate into hard-to-reproduce bugs.

Now, if a top-level make did a three-stage build and then built all the
other tools with the newly bootstrapped gcc, then we'd have the simplicity
you speak of.

Re: Building EGCS from scratch

[Mike Stump]

> Date: Mon, 8 Sep 1997 23:47:01 -0700
> From: "Steven J. Madsen" <steve@moonglade.com>
> To: egcs@cygnus.com

> It seems that EGCS is sorely missing documentation on what needs to be built
> for <foo> OS, and in what order.

It isn't missing, though some of the simplicity seems to have been
obscured over the years.  It is configure; make; make install, that's
about it, and that is the order.  Yes, it should really be this
simple, any deviation from this simplicity, is a bug, and will
probably be addressed in time.  There are many other options on this
basic scheme and other variations are possible.  I for one would like
to see the first configure become optional, so that the default make
target of the shipped Makefile does a configure, then a normal make.
That would mean we are then down to `make; make install' for the
normal case, and we'd be back to the original unix philosophy.  Cross
cases and special order things are harder, but that is ok.

Building this way is the most basic and simplest, and works on most
GNU software.  Hopefully it is easy enough to memorize.