Re: update dwarf2 asm unwind info [hppa64-*-* failures]

[Steve Ellcey <sje at cup dot hp dot com>]

On Wed, 2008-08-13 at 22:08 -0400, John David Anglin wrote:
> > Since this didn't help with shared libraries and after re-reading Dave's
> > comments about the pc-relative issues I tried doing a build with
> > DIFF_EXPR_OK undefined.  My test case works (with static or shared
> > libraries) and I am running the g++ tests which look good so far.
> > My question is:  Is this a reasonable thing to do?  I don't pretend to
> > understand all the ramifications of not setting this but on the surface,
> > it seems to fix the problem.
> 
> Possibly.  However, I'm getting confused as to why you don't see
> the "invalid or unsupported encoding in .cfi_personality".  How
> are personality calls being made?  We used to use DW_EH_PE_aligned
> for this case as this results in a plabel being generated (P') in
> the assembler output for.eh_frame.  I presume you still have Richard's
> change installed with .eh_frame in the data quadrant.
> 
> I thought we needed DIFF_EXPR_OK for dwarf2 debugging info.
> 
> Dave

I don't know how personality calls are being made, I'll try to look into
that.  I did keep all of Richard's changes in place in addtion to
unsetting DIFF_EXPR_OK.  In the end it did cause about 50 regressions in
the C++ compiler and library testsuites, so it is not a clean fix for
this problem.  Is it time to lie and make GCC think that the .cfi stuff
isn't supported on HPPA 64?

Steve Ellcey
sje@cup.hp.com