🚀 go-pugleaf

I submitted a report for this bug to bug-gnu-utils yesterday, but am
not really satisfied that this is entirely a gas bug.

Software: gcc-2.2.2 (configure i386-sequent-bsd --gas --with-gnu-ld)
          [ld is from g++-1.39.  the fact that binutils-2.0 won't build
           for this platform "out of the box" is another matter.]
Hardware: Sequent Symmetry running DYNIX 3.0.12 in the ucb universe.

Bug:      While building the stage2 compiler, genattr cores in
          init_rtl().  The problem appears to be that the stage1
          compiler is (for a reason unknown to me) putting ".align 2"s
          in the middle of the assembly generated for the function
          init_rtl() in rtl.c.

	.align 2
.globl _init_rtl
_init_rtl:
	pushl %ebp
	movl %esp,%ebp
	subl $32,%esp
	pushl %edi
	pushl %ebx
	xorl %edx,%edx
	.align 2           <---------------- this guy right here.
L201:
	movl _rtx_format(,%edx,4),%edi
	xorb %al,%al
	movl $-1,%ecx

	  The system assembler correctly fills the void with "nop"s,
          but gas is filling the void with zeros.  Obviously gas has a
          bug, but I am left scratching my head and wondering why gcc
          felt it had to align the label (since gcc-1.39 did not do
          this).  I have built the compiler without gas, but I need
          gas to handle the stabs generated by g++.  I am open to
          patches and/or suggestions on how to proceed...  :-)

--
Ronald Cole           +----------------------+  internet: rjc@unify.com
Software Engineer II  | This space for rent. |  uucp:     uunet!unify!rjc
Unify Corporation     +----------------------+  voice:    +1 916 928 6238
          "THE BILL OF RIGHTS --- Void where prohibited by law"