Re: MachTen Egcs Patches

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

  In message <v04011700b1d6b8ac93ee@[205.180.86.20]>you write:
  > 	1. Addition of CC1_LDFLAGS to provide for inclusion of
  > -Xlstack=1048576 during the construction of CC1.  This is essential to
  > create a large enough stack so that CC1 will operate at all.
Is this needed during all the stages, or just when building the stage1
compiler?

If the former, then you're better off defining BOOT_LDFLAGS in your x-machten
file.  If it's the latter, then I'm sure we've got a hook in one of the
xm files to do this.


  > 	2. -fno-builtin-alloca and -fbuiltin-alloca.  We include the
  > no-builtin-alloca to use our own real memory stack management
  > __builtin_alloca which comes from libc.  After base sizing we attempt to
  > grow the stack on a block-by-block basis (allocated via malloc).  The
  > blocks are managed as a separate list which at appropriate points in the
  > processing or when requested, the list is pruned according to the current
  > stack pointer.  This is essential to support non-virtual memory based stack
  > emulation.
I still not not see why this is essential to support non-vm based systems.

Or is the problem that your stacks are too tiny?  If that is the case, then
we'll need to look at other solutions, particularly since they effect both
host and target issues.

  > 	3.  We added the -Xl command line option to support linker options.
  > Specifically -Xlstack.
I assume you're trying to make it easy for folks to pass options to the linker?
-Wl,option is the way to do this.

  > 	4. In tree.c we provide for the optional override of the alloc and
  > free with the OBSTACK_CHUNK_ASSIGNMENT.
No.  This will not be accepted.  Overriding the obstack stuff is verboten.


In general, if you find yourself making a change to a machine independent file
you're going to need to explain them in a rather detailed manner.

I still don't see a copyright on file with the FSF.

jeff