Re: Solaris boostrap breakage

[Mark Mitchell <mark at codesourcery dot com>]

>>>>> "Richard" == Richard Henderson <rth@redhat.com> writes:

    Richard> On Fri, Mar 09, 2001 at 06:36:59PM -0800, Mark Mitchell
    Richard> wrote:
    >> I have completed the binary search, and verified that it is
    >> this patch:
    >> 
    >> * config/sparc/sparc.c (function_arg_record_value_3): Fix
    >> calculation of the number of integer registers required.
    >> (function_arg_record_value): Likewise.

    Richard> Err, I'm having trouble believing that.

    >> I know that Kenner says this is Sparc64 only -- but this is
    >> definitely the patch that triggers the problem.

    Richard> You've set a breakpoint on this function and it hits _at
    Richard> all_?  Cause 32-bit early exits from all this stuff e.g.

I didn't try that.

But, I did exhaustively do the CVS binary search, and I've verified
using `cvs diff' that the only change is your change to sparc.c.  (I
didn't just rely on the ChangeLog; I actually looked at the complete
set of diffs.)

The problem involved here is very, very strange.  Kaveh and I have
both had a very hard time reproducing the problem outside of the
actual `make bootstrap' phase.  It's entirely likely that your change
is only the trigger for some much more subtle memory error; the
changing layout of sparc.o could, for example, in conjunction with the
environment set up by `make', could be together touching some piece of
memory somehow.

However, there's no question that it is with this change that the
problem shows up.  It would be good if you could track it down.

Otherwise, we'll just have to pull out the patch and hope nothing else
tickles the same bug -- I don't have the time to try to figure out
what's being tickled by this.

Sorry,

--
Mark Mitchell                   mark@codesourcery.com
CodeSourcery, LLC               http://www.codesourcery.com